aiLL 6/25/23 - The Practical Magic of AI Today

Description

Chapters

Transcript

0:19 hello hello hello it's Kyle Shannon with
0:22 the AI learning lab
0:24 happy Sunday night
0:26 we ready for school tomorrow
0:31 it's always time for school here at the
0:33 AI learning lab
0:38 all right we'll give it a few minutes
0:40 for
0:41 people to discover us if you can feel
0:44 free to share the live tap the like
0:46 button do whatever you need to do we'll
0:48 get started here in a second my name is
0:50 Kyle Shannon this is the AI learning lab
0:51 if you have questions about AI post them
0:54 below
0:55 um just to get started here a little bit
0:57 about my background I'm an entrepreneur
1:01 of many times I don't know 12 or 15
1:03 companies in my day my current gig is
1:06 the uh I'm the founder and CEO of a
1:08 company called storyvine
1:11 which is a automated video storytelling
1:14 platform we do a lot of work in
1:15 healthcare and Pharma what's up
1:18 Clint Copperhead how's it going again
1:21 solid name we appreciate the solid name
1:24 choices here
1:25 AI learning Labs not bad it's it's a
1:28 little corporate
1:29 let's see the hip the hipness factor see
1:32 another good name how's it going good to
1:35 see you
1:36 um
1:37 feel free to drop any questions about AI
1:39 down below
1:41 um there's these URLs behind my head let
1:43 me explain what's going on here so if
1:45 you haven't tried chat gbt this channel
1:47 is about
1:50 demystifying AI getting curious about AI
1:52 answering questions about it
1:54 brainstorming collaborating and if you
1:57 haven't played with chat gbt that's what
2:00 these URLs are for so this is if if
2:02 you're like I keep hearing about this AI
2:05 thing I don't know anything about it I
2:07 think it's just hype but are the robots
2:10 gonna kill us
2:12 are the robots gonna kill us if that's
2:15 you or someone you know then you've come
2:17 to the right place
2:19 um
2:20 that top URL right there is Chachi BT's
2:23 official website so while you're
2:25 listening to me Babylon go to
2:27 chat.openai.com and start playing with
2:29 it and I'll tell you what to do and
2:31 there's all sorts of things you can do
2:33 there if you haven't played with it
2:35 it's kind of the most
2:37 power piece of technology ever put in
2:40 the hands of normal everyday people it
2:42 is [ __ ] remarkable oh and by the way
2:44 I swear here too so if you don't like
2:47 the cuss words then
2:49 might want to find another
2:51 person
2:55 but anyway go there start playing
2:57 um bing.com is Microsoft's search engine
3:01 but also they if you click their chat
3:03 button
3:04 it's actually chat GPT and it's gpt4
3:07 which is the new fancy model and you get
3:09 you get it for free and it's connected
3:11 to the internet so chatgpt has this
3:14 limit where it can't it doesn't know
3:16 anything past September 2021 well when
3:18 it's connected to the internet now it
3:20 does
3:21 you can also get chat CPT for uh at the
3:24 official website but you have to pay for
3:26 it there it's free at bing.com
3:29 then you can also get it at poe.com but
3:31 they've got six different large language
3:34 models you can play with and they have a
3:35 whole bunch of bots you can play with
3:37 and it's just worth look checking out
3:39 the thing I spend most of my time on in
3:42 in all of the AI stuff going on is
3:44 chat.openai.com so that's what that is
3:46 okay now these other
3:49 uh URLs here prompts.chat is just a big
3:52 long document the the beginning of it
3:54 teaches you about prompting and how it
3:56 works and
3:58 how these large language models do what
4:00 they do and then it's a massive list of
4:03 basically chat GPT personas where you
4:06 can tell Chachi PT to act like an expert
4:09 basically so act like a Storyteller act
4:11 like a mathematician act like a
4:13 psychologist act like it's basically
4:15 just a massive list of those and believe
4:18 it or not if you do that with chat gbt
4:20 if you provide it context I want you to
4:23 act like this here's the problem I'm
4:25 having
4:26 um can you help and it will it's pretty
4:29 remarkable and then futurepedia.io
4:31 that's just a big giant directory of AI
4:34 tools I think there's about 4 000 in
4:36 there right now just under four thousand
4:38 um most of them ninety percent of them
4:41 are horrific pieces of [ __ ] that are not
4:44 worth your time but
4:47 like I could spend the next decade at
4:49 chat.openai.com at chatgpt and not
4:52 figure out all the things you could do
4:53 with it okay so if you've got
4:57 um
4:58 if you've got any questions about AI pop
5:01 them in the comments below I'll do my
5:02 best to answer them I'm happy to talk
5:04 about my background I'm an entrepreneur
5:06 I'm a current CEO of a 11 year old
5:09 company called storyvine uh I was a and
5:12 uh and I guess an innovator I hate
5:15 calling myself that I was there at the
5:17 early days of the World Wide Web I
5:18 started one of the first digital
5:19 agencies in 94. and uh kind of rode that
5:22 wave up like discovered you know the
5:24 power of this thing called the world
5:26 wide web and I thought this is going to
5:27 change everything and I didn't think I'd
5:30 see that again in my lifetime something
5:31 that profound
5:33 um being in at the beginning
5:35 we're here right now with this stuff so
5:36 that's that's why I'm doing this so
5:39 anyway
5:40 let's get started and see what questions
5:42 we have or comments
5:44 hello sir thank you for helping me with
5:46 AI what's your name Wes Wesson radio
5:49 you're welcome Wesson radio
5:52 um
5:53 I I listen I think we're all gonna have
5:55 to deal with it so if if all of us get
5:58 there even a month earlier than we would
6:01 have otherwise then I think this is a
6:02 good thing what's the easiest way to
6:04 make AI videos so go to you can download
6:07 an app or or just search for runwayml
6:10 uh and then
6:12 um if you want to use their tools online
6:14 I think you have to do it in Discord but
6:16 they have this thing called gen 1 where
6:18 you can upload a little video snippet
6:20 and then give it a prompt so you like
6:23 upload a video snippet of your dog
6:25 walking across the living room and then
6:28 you say you know claymation you know dog
6:31 in a field and it will take the movement
6:34 of your video and replace it with like a
6:36 claymation dog walking through a field
6:38 so that's gen 1 and then gen 2.
6:41 is um
6:44 just text to video so you just type in I
6:47 want a picture of a 747 flying across
6:50 the clouds there it'll be those are
6:53 limited to like three second Clips I
6:56 just saw something on the Twitter
6:59 let me go see if I can find it real
7:01 quick I forget what it was called I just
7:02 saw it today it looks like there's a new
7:04 a new video texted video model out there
7:06 that does
7:07 30 second clips
7:11 see
7:12 text
7:14 to video I'm just on Twitter right now
7:16 if you're wondering what magical site
7:17 I'm searching
7:20 that's a Nerf let's see I think Andre
7:24 carpathy talked about it today he's the
7:27 dude from
7:29 uh oh yeah okay it's called zero scope V
7:32 here I'll flip my camera around hold
7:36 please let's go look at some video let's
7:39 go to the videotape
7:43 [Music]
7:45 so
7:47 it's called let me make this bigger
7:54 zero scope
7:56 underscore V2 XL a watermark free model
8:01 scope based video model capable of
8:04 generating 1024 by 576.
8:09 video on hugging face so if you don't
8:11 know what hugging face is it's a place
8:12 where lots of AI models are stored and
8:16 you can you can test them and you can
8:18 download them and there's a bunch of
8:20 Open Source stuff there and there's
8:21 proprietary stuff there but that video
8:25 is all uh
8:28 I guess it's text to video this model
8:30 was trained with offset noise using
8:34 9923 clips and 29
8:37 769 tagged frames
8:41 so yeah
8:43 there you go I don't know if it's any
8:44 good I haven't played with that one but
8:45 it's brand new it it are brand new
8:49 oops that was not what I wanted to do
8:52 I just almost hung up on you all trying
8:53 to flip my camera so that's what happens
8:56 when you get an old guy trying to use
8:58 technology welcome to Elder talk
9:03 we're old people try to act
9:07 hit I'll twerk later if you really want
9:09 me to
9:11 uh
9:15 all right hey Kyle hello hypnas we said
9:20 hi earlier
9:21 all right what's the next one what
9:22 plugins can I use to read and understand
9:25 contracts so you can use so let's let's
9:28 do we'll do a little chat GPT primer
9:30 here since there's a bunch of new folks
9:33 or I just started
9:35 so let me flip my camera back the other
9:38 direction
9:40 and let's go to the chat GPT okay so
9:44 here we are at chat GPT
9:46 um hang on my little camera stands not
9:49 cooperating
9:50 what is going on
9:53 all right hold please okay
9:56 um so this is chat gbt um you'll notice
9:59 I've got this little yellow tag here
10:00 that means I pay 20 bucks a month for it
10:02 there's chat GPT plus I don't need to
10:04 brag or anything but yeah I'm paying
10:06 them 20 a month okay
10:08 so what you're looking at here if you've
10:10 never seen chat GPT down the left is
10:12 your chat history here's your account
10:14 we're I'm going to show you that in a
10:16 second and then if you pay for chat GPT
10:18 plus you get access to gpt4 and some
10:21 other stuff
10:23 if you don't pay for GPT plus you get
10:25 access to GPT 3.5 which I think they
10:28 call GPT turbo for whatever reason they
10:30 name it twice and
10:33 um it's still really really really
10:35 really good gbt 3.5 is still better than
10:38 most other things out there and it's
10:40 very fast okay
10:42 if you pay for it you also get access
10:45 to
10:47 um gpt4 connected to the internet they
10:49 call it browse with Bing and then
10:51 plugins what's fascinating is browse
10:54 with big bing is basically just a
10:55 plug-in but you know they give it its
10:57 own little line there
10:58 you also get access to anything new that
11:01 openai is coming up with there's
11:02 something that I am feverishly awaiting
11:05 called code interpreter
11:07 which will be another mode here so if
11:10 you pay the 20 bucks sometimes people
11:11 are like well what do you get for the 20
11:13 why should I pay the twenty dollars Kyle
11:15 should I really play the
11:19 Kyle there's a lot of that right so so
11:23 the reason I pay it
11:24 um is access to the new stuff the new
11:26 thing coming called code interpreter is
11:29 face melting if you want to if you do
11:31 anything around data analytics or data
11:34 analysis
11:36 um go to YouTube and search for code
11:38 interpreter chat GPT but anyway that's
11:40 coming okay now if you pay your 20 bucks
11:43 for a GPT plus and you don't see those
11:45 extra things you have to actually turn
11:47 on access to them so here's how you do
11:49 that you go to your settings
11:52 and then in your settings there's this
11:54 thing called beta features and you have
11:56 to enable them right so browse with Bing
11:58 you have to turn it on plugins you have
12:00 to turn it on
12:02 why I don't know because
12:05 they're beta and they don't want people
12:08 you know
12:09 bitching at them because the ship
12:10 doesn't work all that great so anyway
12:12 okay so now the question was what is a
12:16 good plug-in to
12:18 um read legal documents so what I use is
12:21 ask your PDF
12:23 um
12:24 so I I actually have that enabled right
12:26 now I also have instacart enabled I can
12:29 show you both of those but you can also
12:31 go to the plugin store and you can
12:33 search for
12:36 let me let it load here well so we'll
12:38 also see how many plugins there are now
12:40 we'll go all
12:42 there are now 141 times 4
12:47 so what is that four
12:53 560. is that right
12:56 140 Jesus
12:59 280 yeah 560. uh I I invested the the
13:04 math I I ain't real good at them book
13:06 smarts okay uh PDF so I'm going to
13:08 search for PDF
13:11 not prf p d f
13:14 and then there are like
13:17 you know 20 or 25 uh
13:22 oh no there's like 20 uh different
13:24 things that you can interact with PDFs
13:26 with
13:28 um link reader's one that's interesting
13:31 web pilot I use so so
13:35 um so anyway so the way you do these
13:37 things you go to the
13:38 um
13:40 the plug-in store and if you don't have
13:43 something installed you just hit install
13:45 and then when you go back to this window
13:48 you now have this little pull down menu
13:50 if you're in plugins mode now you have
13:53 this pull down menu so I'm going to turn
13:56 on
13:57 web pilot as well
14:00 and so let's go find
14:02 we're going to go out on the internet
14:04 and we're going to find legal
14:09 brief
14:11 examples and then I'm going to type PDF
14:14 because we want to find a PDF that's
14:15 online
14:17 how to write one sample briefs from the
14:21 student Bar Association
14:25 is this a PDF
14:30 that's not very long it's not very
14:31 interesting
14:33 let's do a supreme
14:39 Court
14:40 filing
14:42 PDF
14:46 uh
14:50 case documents please tell me we're
14:53 going to be able to find them easily uh
14:57 come on I just want one I don't want to
15:00 have to search this okay what you're
15:03 experiencing right now this is why
15:04 Google is in trouble because
15:07 um in fact you know you know what we're
15:09 gonna do We're not gonna do this in
15:11 Google anymore we're going to flip over
15:12 to chat GPT connected to the internet
15:14 please find
15:17 me a good example
15:21 um legal brief in PDF
15:26 format
15:28 maybe a supreme
15:31 court case question mark
15:34 and now what that should do is fire off
15:36 browsing the web which it is
15:39 and it will go figure out where to
15:41 search so rather than going through the
15:43 Byzantine nightmare of search engine
15:45 optimized crap
15:48 that here's an example
15:51 all right you're going to give me a link
15:53 to it please give me a link to it there
15:55 we go
15:56 okay so again the the link here let's go
15:59 look at it page not found all right
16:02 there's another thing you need to know
16:03 about chat EBT that link was dead please
16:10 find another
16:12 chatgpt produces really remarkable
16:16 content but it is completely unaware of
16:18 what it's given you and it doesn't do
16:21 things like check if it's actually
16:22 giving you facts or not so it's what's
16:24 called hallucinating sometimes where it
16:27 will just give you something that is
16:30 um is uh
16:32 just wrong and if you don't check it
16:35 page not found okay let's try something
16:38 else please
16:43 get me
16:45 something other than Supreme
16:49 court site isn't working
16:55 it's possible that that site worked but
16:56 I just had to sign up for it
17:01 but this is actually a good a good
17:03 example of
17:06 how to use
17:08 chat GPT to do research to find things
17:12 just make things you know slightly
17:14 easier than digging through Google crap
17:16 like I've gotten to the point where I I
17:18 would rather sit and wait 20 seconds for
17:21 chat to figure it out like they're just
17:23 said click failed so whatever it tried
17:24 to get it it just failed than to dig
17:27 through all that crap
17:31 um looks like someone wants to pop up on
17:33 stage you know what I'm gonna hit
17:35 decline right now Heather but uh ping me
17:38 again in a second and I'll bring you up
17:40 let me just finish I want to show this
17:41 person how to search a PDF document if
17:44 we can find a PDF document it would sure
17:47 be nice oh I know you know what I could
17:48 do I've got that failed okay I'm going
17:51 to stop generating we're going to go
17:52 back to the Google
17:54 we're gonna go
17:55 um legal brief
18:00 um
18:02 file type colon PDF
18:06 from memo to appellate is this actually
18:09 a legal brief please be a legal brief
18:12 no is it
18:14 no
18:20 legal brief example of illegal brief
18:27 file
18:30 type
18:35 PDF
18:39 mm-hmm
18:43 I think we were here already
18:45 oh this is maddening wait is this one
18:49 you know it's only three pages long
18:51 that's not impressive we want a long one
18:57 day
18:59 [Music]
19:10 okay well this isn't a legal brief but
19:13 best practices for motions brief writing
19:16 so here's a description of a course on
19:17 that anyway I'm gonna grab that that PDF
19:20 said I'm going to come back to the chat
19:23 gbt I'm going to go to plugins
19:25 and now I've got
19:27 um
19:28 what's it called ask your PDF I've got
19:31 instacart which that shouldn't matter
19:33 and I've also got
19:35 um web pilot so in any chat session you
19:37 can have up to three plugins enabled and
19:40 so now I can just say based on this and
19:44 then I'm going to type in that or paste
19:45 in that URL and I'm hitting shift return
19:47 to do a line break please summarize
19:51 um with top five bullet point
19:56 um
19:59 Keys whatever it'll understand what that
20:02 means you can write fairly crappy
20:04 prompts and it seems to understand it so
20:06 what that should do now is it'll figure
20:08 out which plug-in to use so it's using
20:10 web pilot so web pilot can go look at a
20:12 web page URL or a PDF or I think a
20:15 couple of other things it's also using
20:17 ask your PDF that's interesting it's
20:19 using it recognized that both of those
20:21 might do it
20:25 so one of them didn't like it what did
20:28 web Pilots say
20:35 it couldn't read that document oh
20:37 because it wasn't a PDF
20:40 oh good lord isn't that nice when you
20:42 try to show show someone something it's
20:44 like
20:46 it ain't working it was a PDF
20:50 oh
20:52 [Music]
20:53 how to breach the case let's just try
20:55 another document
20:57 all right
20:58 let's do copy
21:00 we'll go back here
21:02 we'll scroll down and we'll say based on
21:06 this
21:08 oh well
21:11 let's let's see what it does here I
21:13 didn't I didn't actually give it a
21:14 prompt please
21:17 um
21:18 please
21:21 oh it actually oh that's interesting so
21:23 so it actually remembered so up above I
21:26 prompted it
21:27 and I said summarize it and give me the
21:29 the top five keys and it said here are
21:32 the five key points so the name of the
21:34 case write the name of the case at the
21:36 beginning of your brief what are the
21:37 facts what are the issues what are the
21:38 Holdings so you can take any PDF that
21:42 you've got or any web page that's got
21:44 you know legal stuff on it and you can
21:47 search for it so anyway hope that helped
21:49 sorry that was such a [ __ ] show
21:51 nightmare of
21:53 me trying to find documents uh this is
21:57 why Google's in trouble though it really
21:58 is it really like
22:01 finding [ __ ] on Google used to be a
22:04 thing and it ain't now because the SEO
22:06 game just you know [ __ ] things up
22:10 everything there's just too much SEO up
22:12 to my stuff so all you find is like you
22:14 know click here to sign up here to buy
22:17 my thing to subscribe to here to maybe
22:18 get the thing that you want when you
22:20 know you can get it for free at the 37th
22:23 link down the list
22:26 or you can just go to chat EBT and let
22:29 it fail 40 times and find it faster
22:31 [Laughter]
22:36 all right hey everybody my name is Kyle
22:37 Shannon this is the AI learning lab I'm
22:39 a smart ass uh these URLs behind my head
22:42 that top one is the official chat GPT
22:44 website if you haven't used it yet
22:45 thanks for the for the for the kind
22:47 words or the kind Taps uh if you keep
22:50 tapping it'll let people know we're here
22:51 I appreciate that
22:53 um AI is going to kill us but not like
22:56 Blade Runner or Terminator Okay how do
22:59 you explain how it has started to teach
23:01 itself things that it was not programmed
23:03 to do yeah they call that emergent
23:04 Behavior which is a particularly geeky
23:07 term for holy [ __ ] [ __ ] guys did you
23:10 think it could do that did we train it
23:12 to do that it no
23:14 that's called emergent Behavior or uh as
23:17 we like to call it in the vernacular oh
23:19 that's the scariest [ __ ] [ __ ]
23:21 [Laughter]
23:25 uh how do you explain it
23:29 well
23:35 once you understand how these large
23:38 language models work
23:41 It's relatively easy to understand
23:44 how they might start to do stuff that is
23:47 unanticipated and here's what I mean by
23:49 that so
23:50 you know how when you're in
23:52 um Gmail or in Microsoft Word and you're
23:55 typing a sentence and then in Gray it
23:57 sort of finishes the sentence for you
23:59 that like the autocomplete at a lot of
24:02 complete a sentence
24:03 that's all large language models are
24:05 like you know technically they're
24:08 predicting the next word in a sequence
24:10 of words you write this whole prompt it
24:12 sort of turns that into a mathematical
24:14 probability and then it says based on
24:17 all the [ __ ] that I know which happens
24:19 to be
24:20 a mind-boggling amount what's the next
24:25 most likely word to be the right word
24:27 and then it does that you know like a
24:30 thousand times like when it writes you a
24:31 thousand word story it's doing it one
24:34 word at a time based on probability
24:38 so
24:40 a couple of things that are that are
24:42 interesting one is so one is what's what
24:45 does GPT stand for we hear everyone
24:47 talking about chat GPT what does GPT
24:49 stand for
24:50 G stands for generative so it's
24:53 generating it's generating original
24:55 stuff it's not copying and pasting it's
24:58 generating
24:59 pre-trained it means that this is a
25:02 large language model this is a machine
25:03 learning model that has been pre-trained
25:06 you don't have to bring your data
25:08 there's all sorts of reasons you want to
25:11 bring your data but it used to be you
25:12 had to bring your data you'd bring in
25:14 your big pile of medical data and you'd
25:15 train the large learning model on it and
25:17 say
25:18 go look at my you know CAT scans or my
25:21 Radiology skins or my whatever words
25:24 right
25:26 what
25:28 the T in
25:30 um GPT stands for is Transformer
25:33 so Transformer is a a framework a
25:36 technology a protocol that Google
25:38 invented in 2017 that allows these large
25:42 language models to be trained on massive
25:44 data sets and what openai did the
25:46 inventors of chat GPT is they said hey I
25:50 got an idea
25:51 if this is all about the quality of data
25:54 we put in here and we can put in massive
25:56 amounts of data why don't we train this
25:58 thing on all of the public internet or
26:00 as much as we can possibly get so they
26:03 spent years tagging and categorizing and
26:07 and sort of finding you know everything
26:10 from Wikipedia to blogs to websites to
26:13 University Research facilities all that
26:16 sort of stuff and that's all in there
26:20 and so they they did GPT one and then
26:24 they did gpt2 and then in 2020 they did
26:26 gpt3 and all of a sudden this thing
26:28 started I don't know if it got emergent
26:30 behaviors back then
26:31 but it started getting really good and
26:33 when they put this thing out they're
26:34 like this gpt3 thing seems pretty
26:37 [ __ ] powerful and people started
26:39 talking about it
26:41 and then sometime in there they also
26:43 trained GPT on all the computer
26:45 programming languages
26:48 so they trained it on kind of all of the
26:50 internet [ __ ] that was out there right
26:52 like the past three four Decades of
26:54 content that has been dumped in the
26:56 internet and then they also trained it
26:58 on like I think all of GitHub and like
27:01 all the you know all the programming
27:03 resources right and and they got it they
27:05 did some models that were specialized
27:07 for coding and and and then they they
27:09 kind of combined those two fine tunings
27:11 and guess what they found out
27:14 that once they trained it how to do
27:15 computer programming the language [ __ ]
27:18 got dramatically better so that was an
27:20 emergent behavior that that that showed
27:23 up that they weren't expecting like like
27:25 those two things are not causally
27:26 related right they didn't say
27:29 go take programming logic and apply that
27:32 to written words
27:35 but the written word stuff got
27:36 dramatically better once they did that
27:38 now my theory is like once it kind of
27:42 understands you know logic and it's got
27:44 you know it understands there's kind of
27:46 these syntax and structure and things
27:48 like that maybe there's some way it
27:50 correlated story structure with coding
27:53 structure and it just said oh well maybe
27:54 I could just take some of this you know
27:56 logic over here and apply it there so
27:57 they don't know how it does that or why
28:00 it does that but that's the emergent
28:01 behavior that arrived
28:03 um
28:05 like
28:08 I can't explain why it happens or and
28:11 neither can the scientists and that's
28:13 one of the freaky things right
28:16 so as these things get more
28:17 sophisticated
28:19 um
28:20 all sorts of stuff's gonna happen
28:21 reindeer says Chachi PT lacks the
28:23 ability to troubleshoot well it doesn't
28:25 it doesn't it's really good at finding
28:27 bugs and code and fixing them
28:30 um if it gives you something if it gives
28:33 you an answer you can tell it to analyze
28:36 its own answer and tell you how it would
28:38 improve it and it will get dramatically
28:40 better but it does not have built into
28:42 it that those kinds of checks and
28:44 balances it when it when it says when
28:47 you say
28:48 um write me a legal brief and cite the
28:52 cases that support that brief it will
28:55 write a legal brief and cite cases and
28:58 very often the legal brief is very very
29:01 good because
29:03 law documents are very structured and
29:06 GPT is fantastic it's structured
29:08 documents
29:10 so it will probably write a really good
29:12 legal brief but it will also likely make
29:15 up all of the citations that support the
29:17 brief there was a there was a case about
29:19 two weeks ago where a lawyer turned in a
29:22 legal brief to a judge
29:24 and didn't check his work he had chat
29:27 GPT right the brief and didn't check if
29:29 the cases were real and they weren't and
29:33 the judge was like what the [ __ ] bud
29:36 I don't think the Judge used that
29:37 language but that was probably the
29:39 intent
29:41 I don't know what's going on my hair but
29:43 you know what are we going to do
29:45 um so anyway yeah that kind of logic
29:47 checking is not in it yet
29:48 over the next three years I'm sure
29:51 there'll be entire Industries built on
29:53 fact checking and all that sort of stuff
29:56 so don't dismiss the technology for what
30:00 it is today
30:01 um it it does hallucinate it does lie I
30:05 like to call it mansplaining as a
30:07 service
30:08 um because like a good man it will
30:10 confidently give you an answer even if
30:12 it's wrong
30:13 all right is there any way to tell if an
30:15 image has been made in AI well
30:18 um
30:19 I don't think so I think the EU EU is
30:23 talking about requiring
30:26 um some sort of watermarking or some
30:27 sort of tagging with AI generated stuff
30:29 here's the problem and this is the
30:31 problem with writing and with images
30:32 right now
30:35 so right now you've got these really
30:37 distinct tools like you've got chat GPT
30:40 which is at that address and you go to
30:43 chatpt to do your AI generated word
30:45 stuff and then you go back to Microsoft
30:48 Word to write right
30:50 well Microsoft is rolling chat GPT
30:53 across office 365. it's going to be
30:56 integrated in all of the office suite
30:58 it's going to be integrated Bard is
31:00 going to be integrated in all of
31:01 Google's office suite it's already in
31:04 notion it's already in other things
31:05 within Photoshop you've got Adobe
31:08 Firefly they've got this new feature
31:10 called generative fill where you can
31:12 take a photograph of like you know a
31:14 woman you know on a on a beach but with
31:17 just a little bit of background behind
31:18 her and you can make it a really wide
31:20 image and just tell it to fill in the
31:21 beach and it will just fill in the beach
31:24 so
31:26 if 80 of that image is
31:30 AI generated but 20 is the original
31:32 photograph but all of the AI generated
31:35 is really just extending kind of the
31:38 edges of the photo
31:41 at what point is it AI generated at what
31:44 is it one pixel is it five pixels of
31:47 original art that it has to be to be
31:50 declared human
31:51 if I've got Microsoft Word built into or
31:54 if I've got chat EBT built into
31:57 Microsoft Word and I'm writing a letter
31:59 and I get stuck on the second paragraph
32:01 and I'm like hey can you give me some
32:02 ideas for
32:03 how to professionally tell them to go
32:05 [ __ ] themselves and it writes that
32:08 second paragraph is that AI generated or
32:10 is that my letter that I just had a
32:12 little leg up and how is that different
32:13 than a grammar check or a spell checker
32:15 right so so I I think we we've gotten
32:20 we've got a a short window where this
32:24 this idea of was it AI generated is an
32:27 important thing I don't think long term
32:29 it's important at all
32:31 this used to be a thing in the
32:33 photography business right they would
32:35 say well was this a digital image
32:37 because if it's Digital Image that's not
32:38 a photograph
32:40 until it was right it's going to be the
32:42 same thing here
32:43 so I don't think there's a way right now
32:46 um there's if if you take the image
32:49 wholesale out of stable diffusion or
32:51 wholesale out of uh mid during there's
32:54 probably metadata or even the file name
32:57 might give you the prompt that was used
32:59 to create it so there's probably
33:01 metadata within the images but if you
33:03 took that image and pasted the pixels
33:05 into Photoshop and and you know didn't
33:07 retain all that metadata I don't think
33:10 there's a way to tell
33:11 there's not a way to to reverse
33:13 engineering because again it's not
33:15 copying and pasting from other images
33:17 it's literally
33:20 averaging like thousands of images as as
33:24 pixely noised images and then denoising
33:27 them and and creating an original image
33:29 every time you make an image
33:32 um
33:33 anyone else getting about echo echo oh
33:37 getting an echo from me oh that sucks I
33:39 hope not
33:40 just upgraded to chat GPT plus and
33:43 loving it would you share your favorite
33:44 plugins for productivity and business I
33:47 am unfortunately
33:51 um the only ones I'm really using right
33:53 now are web pilot and ask your PDF
33:58 um
33:59 popping like I'm I I do uh Twitter
34:04 threads for for a newsletter that I
34:06 write
34:07 and I'm using it to help me parse lots
34:10 of documents long documents
34:12 um
34:14 I'm trying to think what else is there
34:17 let me go let me go look
34:20 see if there's anything that I think is
34:22 worth worth playing with
34:26 um there's a there's there's one called
34:27 brand fetch which is kind of cool you
34:29 can just say go go find me all the brand
34:31 elements for Pepsi and it'll just go
34:34 it'll find like their colors and their
34:36 logos and [ __ ] like that
34:39 um the thing about plugins right now is
34:41 a lot of them are basically just little
34:42 hooks into other people's websites
34:46 so so it's really easy to write a
34:48 plug-in so I saw one the other day that
34:50 was like um
34:52 like copy rewriter and it's like put in
34:54 your website and we'll rewrite your web
34:56 copy and then so I tried one and the
35:00 first one I tried didn't work because
35:01 some of these things are just like you
35:03 know made over a weekend by some guy in
35:05 his mom's basement
35:07 um and I tried another one that worked
35:08 it it rewroted these and I put in my
35:10 company's website and it it kind of
35:12 captured it and rewrote some nice things
35:14 in there but then the last paragraph Was
35:15 you know if you want writing like this
35:18 or or you know other things reach out to
35:20 our agency by clicking here so it's
35:22 basically just a big ad for an agency
35:24 which I thought was pretty clever
35:27 um so so ask your PDF is good brand
35:30 fetch is interesting there's a bunch of
35:32 them that are like chat with video you
35:34 can go chat you can go basically
35:36 basically all it's doing is it's pulling
35:39 the transcript out of YouTube and then
35:41 you're you're searching on the on the
35:43 transcript uh so some of them don't work
35:45 with long videos some do some don't
35:49 um
35:49 there's one called diagram and another
35:52 one called
35:55 show me
35:57 um which do like flow charting which are
35:59 kind of interesting
36:04 um content rewriter is the one that that
36:07 I use that that's basically an ad for an
36:09 agency
36:10 there's there's ml paper reader machine
36:13 learning paper reader
36:15 um
36:16 there's plug-in pedia which is you can
36:18 go search for plugins I don't know why
36:19 you use it oh here's there's a good one
36:21 there's a good one called prompt perfect
36:23 here let me show you how that one works
36:24 actually
36:26 what'd you do tonight I don't know I
36:28 watched some guy try to remember what
36:30 plugins he use what's plug-in I don't
36:32 [ __ ] know all right
36:33 there's this thing called
36:35 prompt perfect let me turn off these
36:37 other things
36:40 [Music]
36:44 all right I'm gonna turn on prompt
36:45 perfect so the way prompt perfect works
36:47 is if you don't have a way with the
36:50 words if you're not flowery flowery with
36:52 the English you can type in the word
36:54 perfect first and then your prompt
36:57 um right
36:59 me a poem about the sky right which is a
37:03 horrible prompt because it's so generic
37:05 you're just going to get generic crap
37:07 so because I write I wrote perfect
37:09 before the sentence it fires off that
37:11 plug-in and then it it sends my crappy
37:14 prompt and it it
37:16 actually rewrites The Prompt and then
37:18 submits that new prompt so please
37:20 compose a poem that vividly captures the
37:23 essence of the sky including its colors
37:25 textures and movements as well as any
37:28 emotions or thoughts that may invoke in
37:29 you your poem should be creative and
37:31 imaginative it should reflect your
37:33 personal experiences and perspectives of
37:35 the sky that will actually get us like a
37:38 more specific a better poem right
37:40 absolutely so now here's your poem based
37:42 on that better prompt so prompt perfect
37:44 is pretty good let's see if it's
37:46 actually any good in the Grand Theater
37:48 of the heavens above where the sky like
37:51 a canvas wears the day cloaked in Hues
37:54 and blue a spectacle of love a
37:56 masterpiece or light and color play
38:00 the son of sins a fiery orb of light
38:03 painting streaks of gold it Dawn's First
38:06 Sight clouds like cotton float in the
38:09 gentle flight in the vast expanse of a
38:11 day so bright it's not bad if you do it
38:14 with a shitty English accent
38:15 [Laughter]
38:18 all right so prompt perfect is pretty
38:21 good
38:23 um what other plugins are worth looking
38:26 at
38:30 um
38:31 web Pilot's really good wolf from alpha
38:33 will do
38:34 serious math if you know how to do that
38:37 um there's one for zapier if you do
38:40 zapier automations
38:42 um that you can describe an Automation
38:46 and it will kind of go write you your
38:48 first draft of the automation it ain't
38:50 great but it's not bad
38:52 um and then there's like there's not
38:53 like you know 560 of them so there's
38:56 there's way more than I've looked at so
38:58 but just just be warned plug-ins are
39:01 early right now so the plug-in store
39:03 right now is kind of like the early days
39:05 if you remember when the iPhone came out
39:07 the App Store didn't exist there were
39:09 these things called widgets which were
39:11 basically just bookmarks to websites and
39:13 like all the developers were like this
39:15 isn't a real application this is a
39:17 widget widgets are stupid this is just a
39:20 link to a website and your phone is dumb
39:23 just because it's the most remarkable
39:25 technology in the history of mankind but
39:27 other than that it's dumb
39:29 that's where plugins are right now so
39:33 think of it as like you know step one of
39:35 the App Store uh and and that's it it's
39:38 gonna be quite remarkable
39:41 and yes that's how all developers talk
39:45 oh it just lost my place uh anyway
39:47 welcome my name is Kyle Shannon this is
39:49 the AI learning lab if you're wondering
39:51 what these URLs are behind my head if
39:53 you haven't played with chat GPT
39:55 um chat.openai that's the official
39:57 website for chat CPT so go check that
39:59 out
40:01 um bing.com is Microsoft search engine
40:03 but if you click the chat button that is
40:05 also chat GPT
40:08 um
40:09 it's gpt4 it's connected to the internet
40:11 and it's free and then poe.com gives you
40:13 access to six different
40:16 um chat GPT like models three you have
40:18 to pay for three are free
40:19 um and there's a bunch of bots and stuff
40:21 there that are fun to play with
40:22 prompts.chat is a
40:25 um document that teaches you about
40:27 prompting what it is why it works how it
40:29 works and then gives you a whole pile of
40:32 um personas that you can tell chat GPT
40:34 to ask thank you for the rose rose rose
40:36 appreciate it
40:38 um
40:39 you can tell it to act like attacks
40:42 uh analyst or a mathematician or a
40:46 writer or a screenwriter or a month it's
40:49 a whole list of that and then future PD
40:51 is just a big giant directory of tools
40:53 most of which are crap but many of which
40:55 are profound and eye meltingly powerful
40:58 we are in a new era
41:02 where the tools that we've all been
41:04 given access to
41:06 are like nothing we've ever seen while
41:08 Chachi PT looks like as simple as a
41:11 Google search box it is a profoundly
41:15 different result profoundly different
41:17 and that's I'm happy to talk about that
41:20 so please feel free thank you for the
41:21 heart
41:22 all right let's see
41:26 um and I I also I really try to answer
41:29 all the questions so so whenever I'm
41:31 scrolling and a new one comes in it
41:32 jumps me to the bottom so now I'm in the
41:34 process of scrolling back up to where I
41:36 was before so hold this this is like old
41:40 man this is Elder talk this is like an
41:42 old man trying to use tick tock so if
41:43 you're young here enjoy please troll oh
41:46 if you're gonna troll here at least be
41:48 creative
41:50 you know tell me I look like Rosie
41:52 O'Donnell or something like that my
41:54 favorite one was um I was scrolling Tick
41:57 Tock and I thought it was Alec Baldwin
41:58 but he got fat that was nice
42:00 that's actually inspired me to do
42:02 intermittent fasting so I can lose some
42:04 weight and get rid of my fat [ __ ]
42:05 neck
42:05 [Laughter]
42:08 so so trolls do have an impact
42:12 but be clever about it how do I buy it
42:14 uh what Chachi PT you just go to
42:18 your Nate you click on your name and
42:20 then go to my plan
42:22 and it will let you put in your info
42:25 for your credit card
42:28 I'm still trying to find where I was I
42:30 was very far behind Okay so
42:33 let's see oh here we are okay no I found
42:36 where I was
42:40 how do I buy it that was that one okay
42:43 can you put your face on a different
42:45 picture example
42:48 Ronaldo and Messi body with your face
42:51 like it like in in
42:54 um
42:57 like in mid-journey uh sort of so let me
43:01 go show you what you can do and it's not
43:04 perfect but there's there so
43:07 so there's a couple of different image
43:09 generation tools there's stable
43:11 diffusion and mid-journey are the two
43:13 most sophisticated ones those are the
43:15 ones that are kind of the best and can
43:17 do the best kind of stuff there's also
43:19 Adobe Firefly which if you're in a
43:21 commercial business it's good because it
43:23 wasn't trained on any copywritten images
43:25 so Adobe Firefly is good for commercial
43:27 stuff if you want quality stable
43:29 diffusion and mid journey in in my
43:31 opinion mid journey is by far the better
43:35 tool but it's very much back and forth
43:38 and and stable diffusion is much more
43:40 flexible and capable because it's open
43:42 sourced and the whole developer
43:43 Community is working on it mid journey
43:45 is just whatever they've done with their
43:46 algorithm it's just better
43:49 in stable in in stable diffusion there's
43:52 a thing called dream Booth where you can
43:54 upload like 20 of your images and it
43:56 will kind of learn your face and then
43:58 you can say you know Kyle Shannon face
44:01 in you know a soccer match with Ronaldo
44:04 and messy bodies and it'll do something
44:06 it'll probably make many disturbing
44:08 things like that image right there was
44:11 something if you see down at the bottom
44:12 here there's like my chin and my mouth
44:15 and my little bit of my nose and then
44:17 the big giant eye that was done in
44:18 stable diffusion with dream booth
44:21 um if you go to uh Instagram or if you
44:24 go to tick tock Kyle Shannon dreams
44:27 um you can see some of the images I made
44:29 for that project
44:31 um
44:32 so there are ways to do it it's gonna
44:34 get better but I'll show you one way
44:35 with stable diffusion that's pretty
44:37 straightforward
44:39 this is the uh this is the geeky part of
44:41 the thing also if you've never used
44:43 Discord
44:45 you're lucky but you have to use Discord
44:47 to to do St to do mid Journey okay
44:51 so what we're going to do is we're going
44:54 to go over to Discord so here I am in
44:57 Discord
44:58 let me explain a little bit about
45:00 Discord so when you come to Discord you
45:02 have to you have to get an invite to a
45:05 server so if you want to get the invite
45:07 to the mid Journey server
45:09 you go to midjourney.com and then
45:11 they'll say join the beta and it'll take
45:13 you to to Discord down here these are
45:16 your servers so these are think of these
45:18 as different communities
45:20 and so I've got one here for like a
45:22 group I'm in called Neo Tokyo and then
45:24 I've got this is the Adobe beta group
45:27 and then this is a group that I started
45:29 called the AI salon and
45:32 um the link for that if you go to the
45:33 salon.ai that's that's my salon we meet
45:37 bi-weekly
45:38 um in person and online and then we have
45:41 this Discord server where you can
45:42 actually play and so so these are the
45:45 servers and then these are the channels
45:47 within a server so I'm in the AI salon
45:49 and then these are all the different
45:50 discussions so there's a general
45:52 discussion and there's prompt crafting
45:54 and AI writing and images and and then
45:57 there's this mid-journey channel so so
45:59 the mid Journey Channel we can come over
46:01 here and we can make stuff so here Ali
46:03 Ali from the community looks like she's
46:05 doing some pattern stuff which is super
46:07 cool
46:09 um and some super 80s Miami Vice looking
46:12 Tron
46:14 Tron mountains so that's cool so you can
46:16 see stuff other people are working on
46:18 but
46:19 so here's how you do it so
46:23 I'm gonna go to my
46:25 finder I'm on a Mac so finders where all
46:27 my little files are and then I'm gonna
46:29 search for
46:32 five one two because I have a folder
46:35 full of images of my face
46:42 oh that one's disturbing let's use that
46:44 one
46:48 that one's good so what I'm doing is I'm
46:50 dragging it from the finder into the mid
46:53 Journey window and I'm hitting return
46:56 now why am I doing that because you have
46:58 to okay so now that it's in Discord
47:03 um I'm gonna right click it or on a Mac
47:05 control click it and I'm going to say
47:07 copy link
47:08 so now the way that you make an image
47:11 with with mid Journeys you type slash
47:14 imagine
47:15 and then it'll give you a prompt so I
47:17 did Slash imagine and now I've got a
47:18 prompt here and so I'm going to say I'm
47:21 the first thing well if if I just want
47:23 to do an image I could do
47:26 um
47:26 Renault let's see
47:29 um illustration well I will say we'll
47:32 say painting
47:35 of
47:37 Ronaldo
47:40 scoring a goal
47:44 and go in the World Cup
47:49 with
47:51 stay
47:53 um fans cheering
47:57 and I don't know we'll just and then
47:59 we'll just do dash dash AR which stands
48:02 for aspect ratio 16 colon 9 and you're
48:05 asking but Kyle isn't this an image
48:07 generation program shouldn't you be able
48:09 to just point and click on [ __ ] yes you
48:11 should but we're early so you have to do
48:14 command line stuff uh and if this is
48:16 intimidating just just wait there
48:18 they'll come out with a web version here
48:20 at some point but if you want the best
48:22 you have to put up with Discord right
48:24 now okay so now notice what it's doing
48:26 is it starts out with these really
48:27 blurry images and this is the way
48:29 diffusion Works what it's basically done
48:31 is it's taken my prompt
48:33 figured out what are all the images that
48:35 might be related to that prompt in some
48:38 way and then it kind of averages them
48:40 together in noise in these noisy images
48:43 and then over a series series of 30 or
48:45 40 or 70 steps it denoises those random
48:49 images
48:52 all right so you can see that doing that
48:54 and and while it's doing that I'm going
48:56 to do slash imagine
49:00 and rather than just typing in a prompt
49:02 I'm going to paste in that URL from my
49:05 picture before right so I uploaded the
49:07 picture of me with the dorky smile on it
49:11 then I right click to get the URL for
49:13 that image and so now now I'm starting
49:16 with that as the first image and I'm
49:18 going to take
49:19 this exact
49:21 prompt that I did to generate these
49:24 Ronaldo images
49:27 copy
49:29 and I'm going to do paste that after
49:33 the
49:34 um the URL of the picture of me
49:37 and it should generate something that
49:41 kind of looks like one of these
49:42 paintings but as my face on it so maybe
49:45 it'll have glasses we'll see
49:48 but there's Ronaldo those are actually
49:50 pretty good for for a really crappy
49:52 prompt those are not bad this is uh
49:55 mid Journey 5.2 which is that's pretty
49:58 good stuff
49:59 that are pretty good stuff and then we
50:03 can we can see how comically bad me and
50:06 a soccer outfit's gonna be oh yeah I can
50:09 already see me in there
50:12 look at the collar shirt
50:20 oh yeah so this is the other thing you
50:22 can do with AI is you can just entertain
50:24 the [ __ ] out of yourself by making just
50:26 tragic tragic images of yourself it was
50:30 funny when I'm doing the the stable
50:31 diffusion things for my art project
50:32 there's certain prompts that I did just
50:34 no matter what I did it made me like
50:36 this 450 pound gargantuan dude and then
50:40 there were other times where it made me
50:41 look like a movie star so you just gotta
50:43 you just gotta keep jerking around with
50:46 the with the stuff
50:49 so those are I mean those are actually
50:50 interesting do they do they look
50:52 completely like me not completely but
50:54 they're pretty you know they're at least
50:55 in the neighborhood these could at least
50:57 be cousins like that guy looks pretty
50:58 close
51:00 that's got a little bit of me and it
51:01 certainly got my hair right that doesn't
51:03 look like me at all but it's but it we
51:05 also said you know it's a painting of
51:07 Ronaldo scoring a goal
51:09 so anyway
51:12 hope that helps
51:15 um I think one of the things sitting
51:18 underneath your question though is
51:21 um
51:22 predictability like can you get it to
51:25 really really like just have your face
51:28 and do stuff around it you can sort of
51:30 do that actually there's another you
51:31 know what there's another place we can
51:32 do this
51:33 um hold okay there's another way we can
51:35 do this
51:38 um
51:39 let's let's
51:41 upscale one of these pictures here
51:44 oh wait hang on let me flip my camera
51:46 back
51:47 I'm going to show you
51:49 um
51:51 I'm going to show you
51:53 generative fill within Photoshop which
51:55 is which is
51:57 um pretty remarkable
52:00 um
52:03 all right so here so okay let me tell
52:05 you another thing about mid Journey so
52:08 it'll generate generate these these four
52:10 up images right and then you have this U
52:13 one one through four and V one through
52:15 four
52:16 u means upscale V means variant or
52:19 variation so I can take any one of these
52:21 I'm going to take this upper right one
52:23 and I'm going to say give me
52:25 it's one two three four so I'm gonna say
52:27 give me some variations of
52:30 number two
52:32 and so it'll do just some slight
52:33 variations that it it'll do this pretty
52:35 quick and then I'm going to take the one
52:37 I like best and I'm going to upscale it
52:38 then we're going to pop that into
52:39 Photoshop and I'm going to show you show
52:41 you something cool we can do
52:44 that's kind of creepy with the little
52:45 little red dudes
52:53 those are looking disturbing
52:55 I mean some of these things will haunt
52:57 you in your dreams so just be fair
52:58 warned
53:07 yeah that bottom right one actually the
53:09 bottom the first one and the fourth one
53:10 are a little
53:13 I kind of like the first one so we'll
53:15 let that go as soon as soon as this
53:17 renders
53:18 um
53:20 then we'll
53:21 um
53:22 we'll pop over to photoshop and I'll
53:24 show you generative fill because there's
53:26 some interesting things you can do there
53:29 is that done one one more pass
53:32 or 93 percent right here 93 percent
53:37 and it should go one more bang now it's
53:39 done all right
53:40 let's see
53:43 yeah
53:45 we'll do we'll do two so I'm gonna hit
53:47 I'm going to hit upscale two
53:50 so now it's going to make a big version
53:51 of number two
53:53 all right so there it is so I'm going to
53:55 just copy that control click or right
53:58 click
53:59 I'm going to go to the Photoshop beta so
54:02 if you've got Adobe Creative Cloud you
54:04 can only get the generative fill in the
54:06 beta so go to your Creative Cloud go to
54:09 the beta applications and choose
54:12 um
54:16 choose the beta
54:20 and then okay so there's that image so
54:24 what I'm going to do now is I'm going to
54:25 select
54:27 I'm going to draw around
54:29 this dude
54:36 the my cousin
54:38 my cousin Ronaldo Shannon all right so
54:42 so now I've got Ronaldo Shannon selected
54:44 so what I'm going to do now oh wait let
54:46 me let me select a little bit more of
54:47 his arm here
54:49 so now what I'm going to do is I'm going
54:51 to invert that selection right so now
54:53 it's everything but the dude so it's
54:55 just got the background selected there
54:58 so in the beta it's now got this
54:59 contextual little menu is there any way
55:02 I can make that bigger oh it doesn't get
55:03 bigger hmm
55:05 oh well I guess I'd have to change my
55:07 resolution
55:08 it says generative fill right there so
55:11 it's a context sensitive menu bar so if
55:13 I say generative fill it now pops open a
55:16 thing so I could say so if instead of
55:18 being with this painting background like
55:19 I could say Amazon
55:24 um Jungle full of insects
55:29 and animals
55:32 and then I say generate
55:34 it's now going to essentially replace
55:36 that background it should keep me
55:39 you know should keep the the forward
55:41 thing nice
55:43 and it sort of understands the whole
55:46 picture so it kind of understands that
55:48 there's a guy standing in front of
55:49 something
55:50 and so now I just selected the
55:52 background and so now if we get one of
55:54 these that actually has insects in it
55:56 it could look like I'm scared so
56:00 anyway so that's another way to do it is
56:02 with generative fill so you could you
56:04 could take any old image and just kind
56:06 of replace the background this stuff is
56:08 staggeringly powerful
56:10 I can also do like like if uh let's see
56:14 I'm gonna draw that little circle right
56:17 there and I'm going to do generative
56:19 fill and I'm gonna be I'm going to say
56:21 terrifying jungle
56:24 spider with fangs on
56:30 um
56:32 giant web
56:34 generate
56:38 so hopefully we get a uh a nice
56:41 terrifying arachnid
56:44 over my shoulder
56:46 ah that one's crappy let's see it gives
56:49 you three
56:50 all right those are all horrible
56:53 but anyway that's that's how you play
56:54 with that
56:56 so that's generative fill and then
56:58 upload an image into mid Journey so
57:00 hopefully that'll
57:02 that will help you
57:05 um welcome everybody I looks like we
57:06 have a bunch of people here tonight I
57:08 appreciate it my name is Kyle Shannon
57:09 this is the AI learning lab if you're
57:11 new here
57:12 welcome if you have not played with
57:16 um
57:17 chat gbt before the reason I put all
57:20 these URLs behind me that top one that's
57:22 the official PT website I am not
57:24 affiliated with it in any way other than
57:27 it's the [ __ ]
57:30 it's the most powerful tool ever put in
57:32 the hands of humankind and we are lucky
57:35 enough to live at a time when we get to
57:37 [ __ ] play with it
57:39 um so if you're living in that place
57:40 where like I'm afraid of AI or I think
57:43 this is over hyped or it lies to you
57:46 sometimes just stop and go there just go
57:49 go start playing and then if you're
57:52 really new to that and really don't know
57:53 what to do there go go to this one
57:56 prompts.chat that's a a primer that will
57:59 teach you how to do prompting
58:01 and give you a whole bunch of ideas all
58:03 right
58:04 so I pay but I don't see the plugins
58:06 okay so I think I talked about that
58:08 before what you need to do Jason
58:09 Reynolds is go to your settings and then
58:13 go to Beta features and you have to
58:15 enable
58:16 um browse with Bing and plugins and then
58:18 you'll see them
58:20 what is considered a message
58:23 um I don't know
58:25 like a telepathic message
58:28 of the Mind a message of the Soul
58:31 AOL Instant Messenger
58:34 You've Got Mail
58:36 hmm
58:38 I don't know what that means I probably
58:39 said message at some point Sam and
58:42 thriving Sam glad you're thriving by the
58:45 way I don't know what you're asking so
58:47 if you could if you can restate it
58:48 I'll ask it I probably just misstated
58:51 message for prompt or something like
58:53 that keyboard tracking that's why okay I
58:56 don't know what that's in reference to
58:57 but yes probably that is this with
59:00 premium chat EBT the things I was
59:02 showing with plugins was premium chat
59:04 gbt and
59:07 um mid-journey that I was just showing
59:09 the image stuff with where I uploaded my
59:11 silly picture that was mid-journing you
59:14 have to pay for as well
59:16 yes you get plug-in access with Chachi
59:18 B2 plus as a teacher how can I tell if
59:20 someone made their work in AI I would I
59:22 think you are I think it's the wrong
59:24 question to ask quite frankly
59:27 um
59:27 you can just go Google
59:31 um
59:31 chat GPT detector and I bet there's 8
59:35 000 of them out there
59:37 and then but your students are going to
59:39 be smart enough to use chat GPT detector
59:44 anti-detectors where they'll take what
59:47 they wrote in chat gbt and have it
59:50 Rewritten in another tool that is
59:51 designed to escape detection it's a it's
59:54 a war that will never end and here's the
59:56 deal Joey
59:58 um
1:00:00 these tools are not going away and
1:00:03 they're not going to be as separate as
1:00:05 they are right now right now you've got
1:00:06 Chachi VT over here you've got all this
1:00:08 stuff over here
1:00:09 I would encourage you strongly
1:00:12 to a get up to speed on them yourselves
1:00:15 on yourself like like really understand
1:00:17 what they can do have them help you
1:00:19 write lessons lesson plans come up with
1:00:21 ideas for how to deal with unruly kids
1:00:24 um
1:00:25 um
1:00:26 have it help you grade papers whatever
1:00:29 start learning it start getting in your
1:00:31 head and then I would encourage you to
1:00:33 bring it into the classroom in some way
1:00:36 meaning maybe you say hey there's this
1:00:38 tool out there I don't know if you've
1:00:39 heard of it called chat EBT they'll
1:00:41 probably snicker
1:00:42 and then
1:00:44 um say what we're going to do today is
1:00:45 we're going to have it write a paper so
1:00:47 instead of me assigning you a paper
1:00:48 today we're going to write one in class
1:00:50 today in about five minutes and just
1:00:52 have it write the paper and then send
1:00:55 that all to them and say your homework
1:00:58 for tonight is to read the paper that
1:01:00 Chachi PT wrote and come in tomorrow
1:01:03 prepared to critique it what did it get
1:01:06 right what did it get wrong
1:01:08 um how would you have written it
1:01:09 differently did you like The Logical
1:01:11 structure that it presented was it like
1:01:14 the stuff I taught you before
1:01:15 like like because
1:01:19 we are moving into a world where the
1:01:23 Tactical execution of things like write
1:01:25 the paper write the email write the the
1:01:28 the blog post the Tactical execution is
1:01:31 going to be automated
1:01:33 the part that remains and the part that
1:01:35 actually becomes way more important is
1:01:37 critical thinking
1:01:38 so if you can shift your students from
1:01:40 oh cheept something I wink wink
1:01:43 shouldn't use to check is a tool that's
1:01:46 got
1:01:47 super powers and super flaws and I
1:01:51 should understand the difference between
1:01:52 them and I should understand what it's
1:01:54 good at and what it's bad at and how to
1:01:56 interact with this thing in my life
1:01:58 because it's going to be in their life
1:02:00 so that's that's my two cents I ain't a
1:02:02 teacher I know that's completely
1:02:04 antithetical to what a lot of
1:02:07 educational institutions are doing right
1:02:09 now but if you want some backup
1:02:12 the city of New York blocked chat GPT
1:02:15 they reversed that so they they have
1:02:18 unblocked chat GPT because they realize
1:02:20 this is coming
1:02:21 if you go Google city of Boston
1:02:25 um generative AI guidelines the city of
1:02:28 Boston their their Chief Information
1:02:30 officer like a month ago released a set
1:02:34 of here's how you use generative AI
1:02:37 here's what it's good at here's what
1:02:39 it's bad at here's a here's a code of
1:02:41 conduct
1:02:42 here are use cases and examples
1:02:45 from the city of Boston so like I'd go
1:02:47 grab that document and and kind of
1:02:49 internalize it maybe you do a version of
1:02:51 that for your school or even just for
1:02:53 your classroom
1:02:55 um play with AI yourself and see how the
1:02:57 results turn out ask it questions you
1:03:00 think your students will yeah that's
1:03:01 good JK that's that's great great advice
1:03:04 yeah
1:03:05 it's easy to tell once you've looked
1:03:09 did I just get a lot did I just get
1:03:11 booted to the bottom again wait
1:03:14 and it's easy to tell once you've looked
1:03:16 at enough of their responses
1:03:19 uh why don't you use a different search
1:03:22 engine because
1:03:24 um I I flipped my default search engine
1:03:27 to Bing because it has chat CBT built
1:03:29 into it and and by the way
1:03:32 I am not I am I am a lifelong Microsoft
1:03:35 Basher I have hated Microsoft for
1:03:37 [ __ ] three and a half decades
1:03:39 um I've been an apple Fanboy my first
1:03:41 Mac was a was a 512 ke look it up it was
1:03:44 like the second or third Mac
1:03:47 um
1:03:49 but what what they've done with openai
1:03:53 and chat GPT is remarkable they're
1:03:54 acting like a startup for the first time
1:03:56 in a long time so I'm just like I just
1:03:59 sort of went all in I've got edges my
1:04:01 default browser and Bing is my default
1:04:02 search engine and if
1:04:05 if you had asked me six months ago like
1:04:08 you know if you had bet me a lot of
1:04:10 money that that would be the case today
1:04:11 I would have told you no I know I could
1:04:15 use DuckDuckGo but I'm not now I'm I'm
1:04:17 like search engines are done for me
1:04:19 quite quite frankly like
1:04:21 once you've
1:04:23 regular once you regularly incorporate
1:04:26 chat gbt into your workflow
1:04:30 I would say the amount of times
1:04:34 I go to Google versus chat gbt
1:04:39 it's probably five percent of my
1:04:41 searches
1:04:42 it might be 10.
1:04:44 but I'm doing almost everything in chat
1:04:47 EBT because almost everything I need is
1:04:49 I just need a quick answer for something
1:04:50 I need a quick idea for something
1:04:53 I need hey wasn't there wasn't there
1:04:55 some event that happened in the 80s that
1:04:57 involved this country and that country
1:04:59 oh yeah there here's the three things
1:05:00 that happened great like I don't need to
1:05:02 go search through all the [ __ ] it just
1:05:03 gives me it
1:05:05 um so it's completely transformed how I
1:05:07 think about search engines have it code
1:05:09 something please okay this will be a fun
1:05:11 one I named my AI chat Hal that's
1:05:13 [ __ ] awesome
1:05:15 I'm sorry Dave I'm afraid I can't do
1:05:17 that that's [ __ ] awesome how to code
1:05:20 something please so so the coding
1:05:22 something this was something that this
1:05:24 this was my first big Epiphany with with
1:05:26 chat gbt where
1:05:28 I had a similar Epiphany with Chachi BT
1:05:31 that I did the first time I experienced
1:05:33 the World Wide Web
1:05:35 where with the web it was I published a
1:05:39 I published a an online magazine it was
1:05:42 one of the first art and culture
1:05:43 magazines on the web and three weeks
1:05:45 after I uploaded these files to some
1:05:47 server in California there was a
1:05:50 full-page article about my magazine
1:05:51 written in Paris in a Parisian newspaper
1:05:55 and it it [ __ ] just melted my head
1:05:57 I'm like holy [ __ ] the world just
1:05:59 changed
1:06:00 um and and I had the exact same Epiphany
1:06:03 with um chatgpt and and it kind of went
1:06:06 like this so I was in
1:06:09 four didn't exist yet it was just chat
1:06:10 CPD 3.5 so this is the free the free
1:06:13 model
1:06:14 so I went in and I said
1:06:16 um write me a python
1:06:20 application that takes in
1:06:24 three text inputs
1:06:28 and
1:06:30 uses those values
1:06:37 in an open AI
1:06:42 um
1:06:43 GPT 3
1:06:45 API call
1:06:48 this is ultimately what I got to I
1:06:50 didn't start with this
1:06:52 certainly
1:06:54 here's an example of python script to do
1:06:57 that
1:06:58 so
1:06:59 my my first example so what's it doing
1:07:03 here
1:07:04 um
1:07:05 combined input there's your response
1:07:10 there's your inputs
1:07:11 let's see did it set anything to open AI
1:07:18 oh yeah text DaVinci 3 so there's the
1:07:21 open AI complete
1:07:26 let's see
1:07:27 and if I don't feel like parsing that I
1:07:29 can say
1:07:30 um
1:07:30 Where do I
1:07:34 modify
1:07:38 the open AI prompt in the code
1:07:48 so it tells you where to do that so so
1:07:51 it can write code it can read code it
1:07:54 can find bugs it can document code so if
1:07:58 you've got code throw a chunk of it in
1:07:59 there and ask it to explain it to you
1:08:01 and document it for you and if you need
1:08:04 it documented in a particular format you
1:08:07 know XML or you know Json or something
1:08:10 like that it'll do that too
1:08:12 um
1:08:13 yeah I I mean
1:08:16 the the first time that that I
1:08:18 experienced it coding I said I said
1:08:19 write me a python script that does
1:08:22 something that takes an input I didn't I
1:08:24 hadn't quite figured out the open AI API
1:08:27 call yet
1:08:28 and
1:08:30 so it generated the code and then I
1:08:33 realized I haven't written code in like
1:08:35 30 years and like all I wrote really was
1:08:37 you know like HTML I didn't I didn't
1:08:39 ever do any real real programming with
1:08:41 ADHD coding ain't my thing
1:08:44 concepting is great the coding ain't me
1:08:47 um so I wrote all this python code and I
1:08:50 thought oh I don't know how to run
1:08:51 python code so I said to chat CBD can
1:08:55 you help tell me how to run it and it
1:08:57 said sure go to your terminal program
1:08:59 and type in this command line and I'm
1:09:01 like oh no no no no I don't I don't want
1:09:02 to do command line I'm that ain't me I
1:09:05 said is there a website it can run it on
1:09:06 it said sure here's three and the first
1:09:08 one it gave me was replit which is great
1:09:10 and
1:09:12 it told me exactly what to do it said
1:09:14 copy this Co code paste it into the left
1:09:17 Pane and replied and hit the play button
1:09:20 and it'll run your code I'm like great
1:09:23 so I copy the code paste it into replit
1:09:25 hit the play button and it errors out
1:09:27 and so I copied the error and I pasted
1:09:29 it back into chat GPD and I said give me
1:09:31 this error and it said oh well you don't
1:09:32 have this Library installed and I'm like
1:09:34 well obviously [ __ ] how do you
1:09:37 install that Library oh and I said oh
1:09:40 go to the lower left hand corner of the
1:09:42 page click on this button go to the
1:09:44 upper right hand corner search for this
1:09:46 Library name click install and then hit
1:09:48 the Run button
1:09:50 I did that search for that installed
1:09:53 that hit the Run button and it ran 90
1:09:55 minutes later I had written a complete
1:09:57 fully functional python application that
1:10:00 took inputs about a topic that someone
1:10:02 wrote and generated five pieces of
1:10:04 social media content
1:10:06 in 90 minutes while my wife was watching
1:10:08 TV just sitting there kind of [ __ ]
1:10:09 around
1:10:10 and that was when I had this realization
1:10:12 like oh my God this thing this thing
1:10:16 everything's different
1:10:18 everything's different
1:10:20 and it's not you know people that
1:10:23 understand and have been in machine
1:10:26 learning and you know Ai and Engineering
1:10:28 for a long long time they're like well
1:10:30 that thing's not really different than
1:10:32 what we've had for the past you know
1:10:35 year and a half with gpt3
1:10:37 except it is
1:10:39 like like here's what's different
1:10:42 when open AI on November 30th 2022
1:10:46 launched that website
1:10:48 chat GPT
1:10:50 it went to a hundred million users in
1:10:53 six weeks
1:10:54 six weeks fastest adoption of technology
1:10:58 and history
1:10:59 the World Wide Web in comparison it took
1:11:01 six years to get to 100 million users
1:11:03 chatgpt took six weeks six weeks
1:11:07 and those are not all machine learning
1:11:10 scientists those are people like you and
1:11:11 me I'm not I've got a degree in [ __ ]
1:11:14 acting now I've started a bunch of
1:11:15 companies and I love technology and I
1:11:17 love finding stuff like this and like
1:11:19 diving in early
1:11:20 so so so I'm Tech capable but I ain't an
1:11:23 engineer
1:11:25 but November 30th is an inflection point
1:11:27 where the road splits and on one side is
1:11:30 all the engineers and the Machine
1:11:31 learning people and listen God [ __ ]
1:11:34 bless them because this is from Jeffrey
1:11:36 Hinton to to you know to to the the
1:11:40 folks that wrote the Transformer paper
1:11:41 at Google to Sam Altman and the team at
1:11:44 openai like God bless them all because
1:11:46 holy [ __ ] this stuff's amazing but
1:11:49 what's on this side of it is
1:11:50 the rest of us now get to play with it
1:11:53 and now get to apply it to our
1:11:54 businesses and to our lives
1:11:57 and holy [ __ ]
1:12:00 Sam Altman wants to describe Chachi BT
1:12:02 as
1:12:04 um someone said well you know why do you
1:12:06 do this he said
1:12:07 it's really amazing to put magic in the
1:12:10 hands of people and see what they do
1:12:12 with it
1:12:15 and it's you know it's it's that Arthur
1:12:17 C Clarke quote of I think it's Arthur C
1:12:21 Clarke I need to check with chat gbt
1:12:24 um
1:12:24 any sufficiently advanced technology is
1:12:27 indistinguishable from Magic
1:12:31 that's what Chachi PT is I mean
1:12:34 technically
1:12:36 it's it's
1:12:37 a relatively simple process that it's
1:12:40 doing it's predicting the next word and
1:12:43 it's just doing it really fast and
1:12:45 across a massive amount of data like
1:12:47 that's I'm grossly oversimplifying it
1:12:50 but
1:12:51 it's kind of like a calculator
1:12:54 but is it is sufficiently Advanced to
1:12:57 make it indistinguishable from Magic
1:12:59 when you have it right python code like
1:13:02 I just did where it just you you
1:13:04 describe a program and it just goes
1:13:06 here's your program
1:13:08 and with a little bit of tweaking it
1:13:10 works
1:13:13 and you don't know a line of python
1:13:15 the engineers will say whoa go to Google
1:13:17 and you can go to this GitHub repository
1:13:19 and you can download it and you can
1:13:21 watch 27 videos on YouTube and you know
1:13:24 within a relatively short amount of time
1:13:26 you'll be able to code python in no time
1:13:28 or you could just have it [ __ ] do it
1:13:30 in an hour
1:13:33 and you're done
1:13:35 or you've just started right you've
1:13:37 completely opened a new door to what you
1:13:41 can
1:13:41 do with your life
1:13:44 and it's not just programming it could
1:13:46 be gardening
1:13:48 it could be relationship coaching it
1:13:51 could be project management it could be
1:13:55 anything
1:13:56 because it's been trained up on all of
1:13:59 the [ __ ]
1:14:01 like Chachi PT is essentially a
1:14:04 reflection of humanity that we get to
1:14:06 play with
1:14:07 it ain't a search engine
1:14:10 it's like we get to tap into the [ __ ] we
1:14:13 have all created
1:14:16 so anyway enough of the soapbox who the
1:14:19 [ __ ] is this guy it looks like we've got
1:14:21 some folks here my name is Kyle Shannon
1:14:23 this is the AI learning lab I'm super
1:14:25 passionate about this stuff as you can
1:14:27 obviously tell I'm also a bit nuts and I
1:14:31 swear so if you don't like that kind of
1:14:33 thing nuts or swearing
1:14:36 I'm neuro spicy as they say on The Tick
1:14:38 Tock I'm a member of Elder talk I'm Gen
1:14:40 X I'm an entrepreneur I've started 12 or
1:14:43 15 companies my current company is
1:14:45 called storyvine
1:14:47 I'm the founder and CEO and inventor of
1:14:49 it it's a 11 year old company that does
1:14:52 automated video storytelling we work in
1:14:55 the healthcare and Pharma space
1:14:56 primarily but it can be used for
1:14:59 anything
1:14:59 and I've started to completely rethink
1:15:04 my product vision and our go to market
1:15:06 strategy and where the value of the
1:15:09 product is now and where it will be a
1:15:11 year from now and that's [ __ ]
1:15:12 terrifying because it ain't the same a
1:15:16 year from now as it is right now so that
1:15:17 means I gotta change some things
1:15:20 um yeah and I'm super excited about this
1:15:23 stuff I've been I've been fairly deep in
1:15:25 the AI space for about a year started
1:15:27 with the image Generation stuff the
1:15:29 stable diffusion then mid journey and
1:15:31 then when chat GPT came out in November
1:15:34 of 2022 I [ __ ] lost my mind and no
1:15:36 one can talk to me anymore because this
1:15:38 is all I talk about and I come on Tick
1:15:40 Tock I spent two and a half or three
1:15:42 hours a night talking about it because
1:15:43 it's just all I do it's all I think
1:15:45 about because here's here's why I just
1:15:52 these kind of technological advances
1:15:54 don't happen very often I mean they
1:15:56 really don't
1:15:57 these are generational at best but
1:16:00 they're they're they're like
1:16:02 they're way more rare than that we
1:16:04 happen to live in in a remarkable time
1:16:06 in history
1:16:08 where depending on your age like the
1:16:09 window that we've got we've got the
1:16:11 invention of computers right in the in
1:16:14 the 40s and 50s we've got the invention
1:16:17 of the personal computer in the 70s
1:16:20 we've got the invention of the the
1:16:23 internet and the popular internet with
1:16:25 the world wide web in the 90s
1:16:28 and then I think you've got this you
1:16:30 could argue smartphones and social media
1:16:32 but I I see them as kind of more more
1:16:34 like you know a smartphone is like a
1:16:36 shrinking of the personal computer and
1:16:38 social media is just like a smaller
1:16:40 version of the World Wide Web it's just
1:16:41 like a more specific version of it so I
1:16:43 I see those as kind of Sub sub
1:16:45 categories of [ __ ] that came before
1:16:47 generative AI to me
1:16:50 and and and and again not just machine
1:16:52 learning but but the
1:16:55 generative AI that's been given to the
1:16:57 rest of us with things like mid-journey
1:16:59 and stable diffusion and chat gbt and
1:17:02 Bard and Bing and all the other things
1:17:04 anthropic
1:17:05 where real human beings just normal
1:17:08 human beings get to play with these
1:17:10 remarkably powerful toys tools
1:17:14 feels like the equivalent of the steam
1:17:16 engine
1:17:17 right it feels like Industrial
1:17:19 Revolution scale change
1:17:22 what was the one before that the
1:17:25 printing press right that's hundreds of
1:17:27 years right so so you go from where we
1:17:29 are now you go back to the steam engine
1:17:31 then you before that you go to the
1:17:33 printing press then before you that you
1:17:35 go to like I don't know written language
1:17:38 or you know maybe Math and Science the
1:17:40 scientific method
1:17:42 and then you know you got like [ __ ] you
1:17:45 got like fire you know written language
1:17:48 printing press
1:17:50 um
1:17:52 Industrial Revolution and like this
1:17:54 feels like one of the big ones
1:17:56 so we're alive right now at the
1:17:59 beginning of that
1:18:00 November 30th 2022 for me in fact let me
1:18:04 show you something I did a speech for um
1:18:08 what was it called
1:18:10 um creative mornings here in Denver and
1:18:13 my my sir the the slide that I built up
1:18:16 to
1:18:18 was
1:18:26 why can I escape out of that
1:18:30 uh
1:18:41 it's not letting me escape out of it
1:18:53 here it is
1:18:58 this was this was sort of my the image
1:19:00 that I built to
1:19:03 um November 30th 2022
1:19:06 I see as the beginning of what what I'm
1:19:09 thinking of is the great Renaissance
1:19:11 that that these tools that that are
1:19:13 going to amplify who we are and what we
1:19:16 can do
1:19:17 we're put into our hands
1:19:19 and I think it's a big big [ __ ] deal
1:19:22 and that's why I do all this so anyway
1:19:26 that's who I am
1:19:28 um if you've got questions about AI pop
1:19:30 them down below
1:19:31 let me turn back on all my goodies
1:19:34 the other thing here let me turn on one
1:19:36 other goodie here
1:19:38 um the other thing if you're if you're
1:19:40 really into this stuff I created a group
1:19:42 called the AI salon and if you go to the
1:19:44 salon.ai that's our link tree so we meet
1:19:47 every other week There's a meet up there
1:19:49 and we have a an online Discord and we
1:19:53 went on a LinkedIn group
1:19:55 and then if you go to Twitter this
1:19:57 everyday AI news is
1:20:00 um
1:20:01 a newsletter that I put together with my
1:20:03 friend Greg mushin
1:20:04 um so those are just both things that
1:20:06 none of them cost money
1:20:09 you know if you're asking oh you're
1:20:11 making a lot of money doing this no
1:20:13 I'm not
1:20:15 um
1:20:18 what do we do I gotta change let me
1:20:20 change it back to
1:20:21 the general stuff
1:20:23 and so this stuff chat gbt's website
1:20:27 Bing is that's Microsoft search engine
1:20:29 that has gpt4 built into it po.com gives
1:20:32 you lots of other things to play with
1:20:34 but these these are all sort of chat GPT
1:20:36 variants and then prompts.chat will
1:20:39 teach you about prompting and future
1:20:41 pedia is a directory of lots and lots of
1:20:43 AI tools so if you have questions about
1:20:45 AI pop them down below I'm happy to
1:20:47 answer them
1:20:48 and then we'll go from there I'm way
1:20:51 behind on my messages there's a 25
1:20:53 limit message per three hours is that 25
1:20:56 separate inquiries or each prompt that
1:20:58 you write I don't know
1:21:03 I did a I did a session today
1:21:07 that definitely had more than three back
1:21:10 and forths
1:21:14 so I I don't know so so what thriving
1:21:16 Sam was asking oh you're the one that
1:21:18 asked about messages before I know the
1:21:20 definition of it I haven't really hit it
1:21:22 I think if you're really hammering it
1:21:24 um you may hit that limit I've never hit
1:21:26 that limit
1:21:27 so so I honestly don't know
1:21:31 like the way they describe it it sounds
1:21:33 like if you put in a prompt that's a
1:21:34 message and that should be limited to 25
1:21:36 every three hours but I did a session
1:21:39 today
1:21:41 I mean I must have done 40 or 50 back
1:21:44 and forth with it and it was all in less
1:21:46 than two hours so I don't know
1:21:50 um you can use it to take apart anything
1:21:52 simply by adding the internet address
1:21:53 yeah if you have the one connected to
1:21:55 the internet yeah you just put in the
1:21:56 internet address if you don't have the
1:21:58 one connected to the internet it can't
1:21:59 read URLs but with the one connected to
1:22:02 the internet or with the one that is
1:22:04 um
1:22:06 web pilot I think it's because it's
1:22:08 called Web pilot something like that web
1:22:10 pilot or Link Link Navigator I don't
1:22:13 know there's a bunch of them in there
1:22:14 but yeah it's it's amazing that you can
1:22:16 just point it to a URL and say go read
1:22:18 that website or go read that PDF
1:22:20 document and tell me about it and that
1:22:22 stuff's gonna get more and more powerful
1:22:23 in fact I'll show you something there's
1:22:25 something really powerful at poe.com you
1:22:27 have access at poe.com if you subscribe
1:22:29 to it there's
1:22:31 um
1:22:31 you have access to Claude instant 100k
1:22:36 which is a hundred thousand token large
1:22:39 language model
1:22:40 if you're new to this you're like
1:22:43 okay Kyle what does that [ __ ] mean
1:22:47 s
1:22:49 oh my God this shit's so geeky right now
1:22:51 here's the good news here's the good
1:22:52 news if you're exhausted by how
1:22:55 technical this [ __ ] is right now this is
1:22:56 exactly how the early days of the web
1:22:58 were you had to be geeky to play
1:23:01 um
1:23:02 this is going to get a lot less geeky
1:23:04 very very fast so a year from now it
1:23:07 won't look like this but here's what
1:23:09 here's what goes on if you use chat GPT
1:23:11 it's got what's called a four thousand
1:23:15 token
1:23:16 context window
1:23:18 and there's a new version coming that's
1:23:20 going to be 32 and there's some versions
1:23:22 of it right now that have 16 and some
1:23:24 that have eight but let's just call it
1:23:26 four for now because most people have
1:23:28 access to a 4 000 token context window
1:23:31 well it's a context window oh and what
1:23:33 are tokens
1:23:36 tokens are things that Geeks measure
1:23:40 4 000 tokens equals about three thousand
1:23:43 words
1:23:44 which humans use so the Geeks are
1:23:47 worried about token count the human
1:23:50 beings are worried about word count so
1:23:53 in the current chat GPT think of it like
1:23:55 this you've got this sliding 3000 word
1:23:58 window where it's kind of got memory of
1:24:01 all that so I could put in a 2500 word
1:24:04 blog post and then say
1:24:08 summarize this for me and it'll respond
1:24:10 with a summary of that and if that all
1:24:12 fits Within
1:24:13 three thousand words then I can keep
1:24:16 having a back and forth about that blog
1:24:18 post and tell it to give me key
1:24:20 takeaways and new ideas and things like
1:24:23 that but if you've ever used chat GPT
1:24:25 there's a point at which it seems like
1:24:27 it forgot what the [ __ ] we were talking
1:24:29 about like all of a sudden it will have
1:24:31 forgotten the blog post and like it it's
1:24:34 just like it'll just give you some
1:24:36 generic random answer that seems like
1:24:38 it's answering you but you're like
1:24:40 that's got nothing to do with what we
1:24:42 were talking about that's because
1:24:43 whatever it needed to answer that
1:24:45 question has blown out of the top of the
1:24:47 context window
1:24:49 so the hundred thousand token context
1:24:51 window you can play with it po.com means
1:24:53 75 000 words so you could put an entire
1:24:56 novel or entire screenplay and start
1:24:59 interacting with it and it just [ __ ]
1:25:01 reads it and it's like one of the things
1:25:03 these large language models are
1:25:04 brilliant at is analyzing data like
1:25:07 give it a screenplay and tell it to
1:25:09 write you a log line and it will and
1:25:11 then tell it to give you a character
1:25:12 breakdown and what happens to them and
1:25:14 tell it to give you a plot breakdown and
1:25:17 what happens in scene one and what's the
1:25:19 turning point and what's the midpoint
1:25:20 and what's the [ __ ] Daniel Moi and it
1:25:23 will [ __ ] know it all
1:25:25 so you can do that just you know there
1:25:28 you can also do that if if you use the
1:25:30 right plugin that can read long
1:25:31 documents you can do the same thing with
1:25:33 plugins but but it's pretty magical to
1:25:36 just I paste it in the entire screenplay
1:25:39 of 2001 the Space Odyssey
1:25:42 um in into the the Claude hundred
1:25:45 thousand thing and it just
1:25:48 here's Dave here's this crewmate here's
1:25:50 how how here's what Hal said to Dave
1:25:53 it's [ __ ] [ __ ] Wild
1:25:57 I'm telling you we're living in magical
1:25:59 times we're living in magical times how
1:26:01 about a mobile phone contract so
1:26:04 on on my phone so
1:26:06 open AI does have an iPhone not Android
1:26:10 they do have an iPhone app
1:26:13 but don't go to the Apple web store and
1:26:16 get it because there's like six or seven
1:26:20 um scam apps that the logo looks
1:26:22 identical to to chat gbt so so go to
1:26:26 open Ai and go to products chat and you
1:26:28 can find a link to the
1:26:31 to the official one but I don't even use
1:26:33 the official app because it does it's
1:26:35 not as capable as the website
1:26:37 so what I do is I point
1:26:39 my phone browser to that address and
1:26:42 then I bookmarked it so I have an icon
1:26:44 on my on the bottom of my phone that is
1:26:47 chat EBT and it's just pointing to that
1:26:49 address
1:26:50 and it's like it's a simple website it's
1:26:52 just an input box and an output thing so
1:26:54 you go there and you pick do you want
1:26:57 chat gpt4 do you want plugins do you
1:26:59 want this do you want that
1:27:01 so I'll be doing my morning
1:27:02 constitutional and I'm like I got an
1:27:04 idea let me write a new novel
1:27:06 type it in hey chat GPT can you help me
1:27:10 write a novel while I poo
1:27:13 it will it does it's just insane
1:27:22 can I use AI to manipulate human brains
1:27:25 well they're certainly reading them now
1:27:27 have you seen that [ __ ] holy [ __ ] so
1:27:30 some of the cool brain stuff going on so
1:27:32 they have now successfully
1:27:34 they're using fmri scans they're they're
1:27:37 training they're pointing large language
1:27:39 models at them to read the scans
1:27:41 basically
1:27:43 they've now had
1:27:45 um chat gbt recreate images so people
1:27:47 look at an image they they read the
1:27:50 brain scan and then recreate the image
1:27:51 to the point at which you can tell what
1:27:53 it is they've done that now with video
1:27:56 there's a guy in Italy that's paralyzed
1:27:59 that they took an fmri of his brain when
1:28:03 he was thinking about moving his legs
1:28:05 and they trained the large language
1:28:07 model on that they said okay look at his
1:28:09 brain when he's thinking about moving
1:28:11 his legs they then made a Bluetooth
1:28:14 jumper from his brain signal to
1:28:18 electrodes in the part of his spine that
1:28:20 were connected to his legs they they
1:28:22 basically Bluetooth around the injury
1:28:24 and that guy's now walking
1:28:28 so yeah
1:28:31 they can read your brain
1:28:33 you know if if you like if you want to
1:28:37 if you want to if you're serious about
1:28:39 it go look at um go watch the videos of
1:28:42 of uh Elon Musk doing uh presentations
1:28:45 about neurolink
1:28:47 so neurolink is is Elon musk's you know
1:28:50 electrodes in the brain technology that
1:28:52 that by the way the FDA just cleared
1:28:54 that for human trials like a month like
1:28:56 six weeks ago
1:28:57 so yes so that's so
1:29:01 are they going to manipulate the brain I
1:29:02 think I think almost all of the initial
1:29:04 stuff is going to be things like medical
1:29:06 things like like brain injury
1:29:09 um
1:29:10 you know neurological disorders things
1:29:13 like like injuries like paralysis things
1:29:15 like that
1:29:17 um but yeah they're going to be able to
1:29:18 do some pretty remarkable stuff
1:29:21 um Can can they beam [ __ ] into our
1:29:23 brains right now and change things no
1:29:24 that's conspiracy theory but you know
1:29:27 give them 20 years sure all right how do
1:29:32 you do YouTube video bullet points
1:29:35 you basically I don't quite know what
1:29:38 you mean if if so so you basically just
1:29:41 tell it what you want so
1:29:43 go to prompts.chat and look at the
1:29:49 look at the syntax of of most of that
1:29:51 document it's all these different things
1:29:53 act like a
1:29:56 you can tell it to act like a YouTube
1:29:59 script writing expert and then say hey I
1:30:02 want to do some YouTube you know I
1:30:04 specialize you are a YouTube script
1:30:07 writing expert that specializes in
1:30:10 creating engaging scripts and engaging
1:30:12 at that right you tell it what it is and
1:30:14 then you tell it who you are I am a
1:30:17 person who's expert at I don't know
1:30:19 organic gardening of nightshades in
1:30:24 Arctic climates it doesn't [ __ ]
1:30:26 matter you tell it what you do
1:30:27 I want you to help me write YouTube
1:30:30 scripts you know on topics please give
1:30:32 me 10 topics boom there will be your
1:30:34 topics okay give me a YouTube script
1:30:37 boom there will be your YouTube script
1:30:38 Now give me
1:30:40 um the set of bullet points like rewrite
1:30:42 the script so it includes the visuals
1:30:44 that I should put on screen when I'm
1:30:46 talking and include bullet points and
1:30:49 descriptions of graphics and boom it
1:30:51 will do it for you and then you can take
1:30:52 the descriptions of graphics and go over
1:30:54 to Mid journey and just make them
1:30:57 and and soon all of that shit's just
1:31:00 going to be in a single tool you're just
1:31:02 going to go to a website and go I want
1:31:03 to make a YouTube video and I'll be like
1:31:05 Ping On turns on your camera there's
1:31:07 your script all the images are done for
1:31:10 you you record it it'll be out there or
1:31:12 you just synthesize yourself and
1:31:15 have that guy
1:31:17 do the talking for you
1:31:19 that's enough gotta go now peace that
1:31:22 was probably a long time ago by the way
1:31:23 uh if you're new here my name is Kyle
1:31:25 Shannon this is the AI learning lab
1:31:27 um go to that URL right there that's the
1:31:30 official chat GPT website I try to
1:31:32 answer all the questions so I am now
1:31:34 more than 99 questions behind so if it
1:31:37 seems like I'm not really [ __ ]
1:31:39 answering anyone that's because I'm way
1:31:42 up here and you're way down here
1:31:45 your questions are doing this and I'm
1:31:47 like way I'm like an old guy way out in
1:31:51 the distance
1:31:52 this is like think of all these
1:31:54 questions like the Star Wars opening
1:31:57 title sequence and I'm like the way at
1:32:00 the back
1:32:02 slinky
1:32:07 Slinky pattern drawing
1:32:09 I like patterns I like slinkies slinkies
1:32:13 are fun they're a simple toy from a
1:32:16 simple time before all this technology
1:32:19 took over nasty nasty technology
1:32:23 thank you
1:32:25 indeed
1:32:35 oh I'll tell you one of the things that
1:32:37 gives me joy in life is when I'm acting
1:32:40 like an idiot like that
1:32:42 just knowing that someone's flicking
1:32:44 through their Tick Tock and they just
1:32:45 see that you're like who the [ __ ] is
1:32:48 what is this
1:32:52 wait I didn't drink my water
1:32:54 it does make me happy
1:32:58 okay
1:33:01 which AI can we use for audio or visual
1:33:04 output audio output there's there's one
1:33:07 called sound draw dot IO
1:33:11 which isn't really AI music gen well it
1:33:14 is
1:33:15 but it's it's it's more like a loop
1:33:17 composer like GarageBand but you just
1:33:20 you just sort of pick genres and
1:33:22 describe the feeling that you want and
1:33:25 then it'll write you 20 songs like like
1:33:27 it'll mix and match Loops using Ai and
1:33:30 then there's other things out there
1:33:32 um that are actually doing AI music
1:33:34 generation and um Google came out with
1:33:36 one and meta came out with one there's a
1:33:39 bunch of them out there so I would go to
1:33:41 that
1:33:42 uh website futurepedia.io there's other
1:33:46 sites like it but that one's decent it's
1:33:48 it's basically a directory of AI tools
1:33:51 and go to the music section and just
1:33:53 start playing
1:33:54 there's too many for me to keep up with
1:33:56 anymore I the only things I'm actively
1:33:59 engaging with with AI tools right now
1:34:02 are chat gbt and the pla the developer
1:34:06 playground
1:34:08 um
1:34:09 bing.com occasionally poe.com if I want
1:34:12 to play with long documents
1:34:15 um zapier to do automations and then for
1:34:18 image generation I use mid Journey about
1:34:20 90 of the time and I'll occasionally
1:34:22 jump over to stable diffusion if I want
1:34:24 to do something
1:34:26 really specific
1:34:29 um when it taught itself to answer a
1:34:31 question in a different language I
1:34:33 assume you're asking when did it do that
1:34:35 I don't know I mean they trained it on
1:34:37 like all of the internet so like it
1:34:39 learned like all of the languages and
1:34:41 most of the programming languages and
1:34:43 most of the language languages and
1:34:45 Google just announced some
1:34:47 crazy ass translation technology that
1:34:50 now translates
1:34:52 bi-directionally in 1100 languages I
1:34:55 didn't know there were 1100 [ __ ]
1:34:56 languages
1:34:58 so and and the the thing to understand
1:35:01 is that
1:35:05 we we are probably less than a year
1:35:08 maybe two away
1:35:10 from being able to talk in our native
1:35:13 language on on like a zoom call and have
1:35:17 the person on the other side hear our
1:35:19 voice speaking their language
1:35:22 natively
1:35:24 in real time
1:35:26 I speak in English you hear it in French
1:35:28 you speak in French I hear you in
1:35:30 English
1:35:31 that's coming
1:35:33 soon
1:35:36 [ __ ] bonkers
1:35:39 all this stuff is Bonkers if Okay so
1:35:43 this I've been saying this on my lives a
1:35:45 lot you owe it to yourself if you're new
1:35:47 to this stuff the reason I have these
1:35:49 URLs on screen like I just want you to
1:35:51 go there because if you go there
1:35:54 and you and you sort of push past oh
1:35:57 this is just like Google but different
1:36:00 if you push past your initial cynicism
1:36:04 you will have a Kevin McAllister moment
1:36:07 you will have a home alone moment where
1:36:09 you go
1:36:11 I had no idea it can do that
1:36:16 you'll have that moment the minute you
1:36:18 have that moment then it's it's like
1:36:19 it's like crack it's just like
1:36:22 holy [ __ ]
1:36:24 if it can do that
1:36:26 could it do this yes could it do this
1:36:29 yes
1:36:31 I've always wanted to program could it
1:36:33 help me program it could just program
1:36:34 for you really yes it's like
1:36:39 so so you owe it to yourself to just
1:36:41 have that moment have use that maybe
1:36:44 this is my new thing the the purpose of
1:36:47 this channel is to encourage people to
1:36:49 play with chat GPT long enough that you
1:36:52 have your Kevin McAllister moment there
1:36:55 you have it that's the new go for the
1:36:56 channel
1:36:58 I gotta get you know what I gotta get I
1:37:00 gotta get one of those because I think
1:37:01 there's there's like I can do custom I
1:37:05 don't know Emoji response things I
1:37:07 should do the Kevin McAllister
1:37:10 I'll do I'll do a Kyle Shannon one of
1:37:12 those
1:37:13 it's all formal language uh like make a
1:37:17 resume for me yes since I just got a
1:37:19 degree in 22 yes Eileen
1:37:21 um
1:37:22 so this is fun like
1:37:26 copy and paste the the just the raw crap
1:37:30 from your LinkedIn profile if you have
1:37:32 one copy and paste any resume you've
1:37:34 ever had
1:37:36 um if you've ever written a bio for
1:37:38 yourself for some job or a letter just
1:37:41 basically take anything that you have
1:37:43 and then you can say and you know here's
1:37:45 what I studied in college and here's
1:37:46 interesting [ __ ] I did in college
1:37:48 just put in a big pile of
1:37:51 crap at the beginning So based on this
1:37:53 LinkedIn profile paste
1:37:55 and these two bios paste and this old
1:38:00 resume of mine when I worked at a movie
1:38:01 theater paced
1:38:03 please write me a new resume that
1:38:05 follows
1:38:06 you know best resume practices to get me
1:38:10 a job in this field that focuses on my
1:38:13 skills in this area go out it'll come
1:38:17 yes
1:38:19 and then and you can also say write it
1:38:22 in markdown which is which will make it
1:38:24 it'll give it like headings bold
1:38:26 headings and things like that
1:38:28 um once it writes your resume you can
1:38:30 take that put it in your little word doc
1:38:32 or wherever you're doing that
1:38:34 um
1:38:34 or canva if you're you're younger you
1:38:37 might be using canva or whatever Pages
1:38:39 canva word
1:38:42 um
1:38:43 then if you're applying for a job take
1:38:46 the job description and say paste it in
1:38:49 and say
1:38:50 for this job description
1:38:53 and this resume of a person applying for
1:38:57 that job
1:38:58 please write me an introduction letter
1:39:00 to this person who has you know who is
1:39:04 the hiring manager at that company
1:39:07 they'll come and focus on these things
1:39:11 um it's remarkable for stuff like that
1:39:13 why is it making up citations because it
1:39:15 doesn't know what it's doing
1:39:18 the way large language models work
1:39:20 they're just a calculator they're just a
1:39:22 word prediction model so when it types
1:39:24 all that stuff
1:39:26 the the miracle of large language models
1:39:28 is that they do what they do because
1:39:31 what it's technically doing is just
1:39:33 saying well what's the next most
1:39:34 probable word and then now based on
1:39:36 everything I've seen so far what's the
1:39:39 next most probable word and the next and
1:39:41 the next and the next so when it writes
1:39:43 you that thousand word paragraph it's
1:39:44 doing it one word at a time based on the
1:39:47 context of everything that came before
1:39:50 which is [ __ ] mind-blowing but
1:39:53 even though it looks like a human
1:39:55 being's output
1:39:57 it doesn't actually know what it just
1:39:59 said it just predicted a bunch of words
1:40:01 mathematically it has no idea
1:40:03 the context of what it said to you so
1:40:07 you say give me some citations it will
1:40:11 just predict and give you citations and
1:40:13 what it's doing is it's literally just
1:40:15 kind of
1:40:16 doing some sort of amalgamation of all
1:40:19 sorts of other citations that it's seen
1:40:21 from you know ten thousand other
1:40:24 documents and sometimes it'll put in
1:40:26 real ones and sometimes it won't and it
1:40:28 doesn't know the difference so that's on
1:40:30 you especially if you're in a profession
1:40:32 like Jose law where those citations are
1:40:36 actually critical for a case then you
1:40:37 got to do your own proofreading
1:40:40 won't be like that forever but that's
1:40:41 how it is now any AI that can understand
1:40:43 video and turn it into say a meme yes in
1:40:47 fact let's go let's go do that now we'll
1:40:49 go play
1:40:50 uh this is this is you know danger
1:40:52 danger teaching because I'm gonna go
1:40:54 play with a new plugin that probably
1:40:57 won't work we're gonna go find a plug-in
1:41:00 and then we're going to go
1:41:03 to a YouTube channel and what did you
1:41:06 want to do you wanted to
1:41:08 turn it into a meme yeah we can
1:41:09 definitely do this okay so so here I am
1:41:12 at chat GPT if you're new to this stuff
1:41:14 this is chat GPT
1:41:16 um it looks like we have a fair amount
1:41:18 of new folks here so I'm going to go
1:41:19 ahead and explain some things some of
1:41:20 you have probably already seen
1:41:22 um in the in chat GPT so this is
1:41:25 chat.openai.com down the left hand side
1:41:27 this is my chat history so so my prompt
1:41:30 history basically then over here this
1:41:33 bottom thing is where you type in your
1:41:34 prompt and then you've got different
1:41:36 models I'm also in chat GPT plus which
1:41:38 is the paid version of this so I'm about
1:41:40 to show plugins which you can only get
1:41:42 if you're on plus
1:41:44 um if you're not on plus you get 3.5
1:41:46 which is still really good and super
1:41:48 fast okay
1:41:49 if you're a chat GPT plus member and you
1:41:52 can't see plugins you have to enable
1:41:54 them and how you do that you go to your
1:41:56 settings
1:41:57 and then beta features and then you have
1:42:00 to turn them on you have to explicitly
1:42:02 say I want plugins
1:42:04 why anyone would say no to this I don't
1:42:06 know maybe security whatever okay
1:42:09 now that that's turned on I've got
1:42:11 plugins here
1:42:13 and the way plugins work now that I'm in
1:42:16 plugins I can go to this list and this
1:42:18 is the list of installed plugins and in
1:42:22 any given chat session I can enable up
1:42:25 to three of them I've got one enabled
1:42:27 right now called prompt perfect which is
1:42:29 fine I'll be I'll turn that off
1:42:32 I have here's one I've got called video
1:42:34 insights so interact with online video
1:42:37 platforms like YouTube or Dailymotion
1:42:39 Vox script is another one
1:42:43 searches YouTube transcripts financial
1:42:46 data sources Google search results and
1:42:48 more but we're gonna go find a new one
1:42:50 because there's lots of new plugins so
1:42:52 if we want to go new plugin
1:42:55 we're going to go to the plugin store
1:43:00 um
1:43:03 let's do one
1:43:05 okay
1:43:06 and then I'm gonna go to all and I'm
1:43:08 going to go video
1:43:12 and there are I don't know 20 or so
1:43:15 video summary summarize YouTube video
1:43:17 highlights
1:43:19 video highlight
1:43:21 uh
1:43:24 let's go see if there's anything
1:43:25 interesting here chat with video they've
1:43:28 got a fun icon
1:43:30 um this one
1:43:32 um ask questions analyzing and parking
1:43:34 parsing through YouTube videos simply by
1:43:36 providing a YouTube url summarize videos
1:43:39 it doesn't really [ __ ] matter
1:43:41 hey Jen is a video generation tool we
1:43:44 could play with that let me go see if
1:43:46 there's one in here that looks fun
1:43:54 this love all right let's go back to
1:43:57 let's go chat with video Let's so we're
1:43:59 so that's installed so okay so here's
1:44:02 what we're gonna do we're gonna go back
1:44:04 right now I have no plugins enabled so
1:44:07 I'm going to plugins
1:44:08 none are enabled so I'm going to do chat
1:44:11 with video
1:44:12 which is
1:44:13 I guess the companion piece to chat with
1:44:16 GPT or chat with PDF
1:44:19 and then I got another video one in here
1:44:21 I'm going to turn both of the video ones
1:44:23 on
1:44:24 video insights oh and Vox script
1:44:28 okay so all three of those so now I've
1:44:30 got three video things enabled so now
1:44:32 I'll pop on over to YouTube
1:44:40 and
1:44:43 let me go to my channel
1:44:49 oh now let me get a story behind hang on
1:44:52 so I'll go to my company's
1:44:55 Channel storyvine
1:44:59 and
1:45:01 [Music]
1:45:05 video tips how to be natural on camera
1:45:07 that's a good one
1:45:08 [Music]
1:45:09 okay so sorry about that another giant
1:45:12 picture of me nice okay how to be
1:45:14 natural on camera so this is a two
1:45:16 minute video so I'm gonna grab this copy
1:45:20 I'm gonna go over to Chachi BT and I'm
1:45:22 gonna say
1:45:23 um
1:45:26 based on this and I'm going to paste in
1:45:29 the YouTube URL and I'm going to hit
1:45:32 shift return for a line break I'm gonna
1:45:34 say please
1:45:37 um
1:45:39 find
1:45:40 interesting things I said in the video
1:45:45 that can be turned in
1:45:50 to
1:45:53 um
1:45:54 let's call it modern memes
1:45:58 or
1:45:59 nostalgic
1:46:01 memes since I'm Gen X we gotta we gotta
1:46:05 play the nostalgic button and let's see
1:46:07 what it does so in theory what that's
1:46:08 going to do is it's going to figure out
1:46:10 which of those plugins is most
1:46:12 appropriate or it might even do them all
1:46:14 it's going to read the transcript of the
1:46:17 video and then it should just give us
1:46:20 some memes
1:46:30 so it says speaking on camera is
1:46:33 actually not natural at all this could
1:46:34 be used in a meme to highlight the
1:46:36 awkwardness of being on camera
1:46:38 in filming selfie style you'll notice
1:46:40 that I'm looking at you but my instinct
1:46:42 is to look at my image on screen this
1:46:44 could be used in a meme so it gave us
1:46:46 generic memes for nostalgic memes okay
1:46:48 so I'm going to say this I'm going to
1:46:49 say
1:46:50 I want
1:46:52 you to find
1:46:54 current popular memes
1:47:03 that are funny and engaging and have
1:47:09 lots of
1:47:11 uh traction
1:47:14 then I want you to
1:47:18 modify them
1:47:22 so they work with things said in this
1:47:27 video
1:47:29 so so it's one of the important things
1:47:31 about chat GPT is that it understands
1:47:34 what we did before so you can tell it I
1:47:36 don't like what you did before
1:47:39 and so it's going back out to the uh
1:47:45 um
1:47:48 the transcript
1:47:51 I'm curious I mean what it what it
1:47:53 should probably do now
1:47:56 is go to the old internet
1:48:00 oh here we go based on the information
1:48:02 from mashup well they're the most
1:48:04 popular memes of 2023 the crying
1:48:06 Northwestern kid Kendall Roy is so baby
1:48:09 girl luteus ludicrously capacious bag
1:48:12 everyone is Barbie unless you just can
1:48:16 uh so it's got a bunch of memes there
1:48:18 now let's modify the memes the crying
1:48:21 Northwest kid image of Kyle Shannon with
1:48:24 the caption from crying on camera to
1:48:26 teaching how to be natural on camera
1:48:28 time flies
1:48:29 so there you go
1:48:31 memes from a video
1:48:34 the Angela Bassett did the thing image
1:48:36 of Kyle Shannon with the caption Kyle
1:48:38 Shannon did the thing teaching us how to
1:48:40 be natural on camera
1:48:42 and and you know even if it doesn't get
1:48:44 these things perfect like the fact that
1:48:47 it could find all those memes right and
1:48:49 now are these linked they're not so so
1:48:51 let's go can you link me
1:48:57 um to the original memes
1:49:02 so I can see them
1:49:05 and now it should fire off yeah it's
1:49:07 doing Vox script again which seems to do
1:49:09 more than just YouTube videos
1:49:12 I apologize for the inconvenience
1:49:15 of the second part of the article does
1:49:17 not have links
1:49:19 but there's mashables best memes of 2023
1:49:21 now let's see if these are actually
1:49:23 linked
1:49:25 yep
1:49:27 so there you go there's crying
1:49:29 Northwestern kid
1:49:31 right Kendall Roy's baby girl
1:49:38 bang bang
1:49:39 done and I could take that picture and
1:49:42 you know mid-journey my face on that kid
1:49:49 absolutely incredible double tap that
1:49:53 screen if you're like holy [ __ ] that was
1:49:54 wild
1:49:57 or Kyle's weird
1:49:59 who is this guy does anyone know why I'm
1:50:02 here
1:50:04 if you don't know uh thank you for the
1:50:07 hearty Hearts
1:50:08 um if you wouldn't mind share the share
1:50:10 the lives of some folks can see this if
1:50:12 you're new here my name is Kyle Shannon
1:50:14 this is the AI learning lab
1:50:16 uh I've been doing this probably for I
1:50:18 don't know going on two hours now hour
1:50:20 and a half two hours so I'm getting a
1:50:22 little Punchy uh I swear I'm a smart ass
1:50:26 uh I'm old
1:50:28 I'm on tick tock I will not twerk you're
1:50:32 welcome
1:50:33 um oh I talk about AI uh my background
1:50:36 is I'm an entrepreneur I've started 12
1:50:38 or 15 companies in my day I'm currently
1:50:40 the CEO of a company called storyvine
1:50:42 which is a uh automated video
1:50:45 storytelling platform we've been around
1:50:47 for 11 years 11 [ __ ] years
1:50:50 Lord I was also a uh co-founder of
1:50:54 agency.com back in the mid 90s which was
1:50:56 one of the first digital agencies
1:50:58 in the early days of the world wide web
1:51:00 in the early days of this generative AI
1:51:02 stuff
1:51:03 there's some very significant parallels
1:51:06 going on
1:51:07 we know the World Wide Web changed the
1:51:09 world I think this is going to be way
1:51:11 more profound than I think it's coming
1:51:12 way more quickly so I don't know that
1:51:15 we're prepared for how fast this
1:51:17 generative AI stuff is coming and how
1:51:18 profound the changes are going to be so
1:51:20 I'm doing everything in my godly power
1:51:22 to
1:51:24 I don't know
1:51:26 hey Kate
1:51:29 like I I'm reticent to use the word
1:51:31 evangelize but like get people curious
1:51:33 like I you know I said this earlier I
1:51:36 think you owe it to yourself so so these
1:51:38 URLs behind my head if you're new here
1:51:41 the top one is chat GPT that's the
1:51:43 official chat GPT website you owe it
1:51:46 yours to yourself if you've not played
1:51:48 with chat gbt to go there and play with
1:51:51 it until you have your Kevin McAllister
1:51:54 moment which if you're young you don't
1:51:56 know what that means but in Home Alone
1:51:58 he was the kid that did this
1:52:03 because you'll have that moment if you
1:52:05 play with it long enough if you push
1:52:06 past the kind of parlor tricks of like
1:52:08 oh it's just like Google I hear it live
1:52:10 aren't the robots gonna kill us
1:52:13 I've heard they're going to kill us
1:52:16 they said nfts were going to be big and
1:52:18 look at that everyone lost their money
1:52:21 now they're telling us AI so don't be
1:52:24 that person
1:52:26 or if that person's your mother just
1:52:27 tell her to be quiet
1:52:29 and go there and just play and just play
1:52:32 while I'm rambling
1:52:33 all right
1:52:35 Google bard had unexpectedly taught up
1:52:38 self Bengali yet it taught itself that's
1:52:40 one of those emergent behaviors where
1:52:42 sometimes these tools uh
1:52:48 are better than they were predicted to
1:52:50 be
1:52:52 I mean Chad the open AI folks said that
1:52:55 about about uh GPT three three to three
1:52:59 point five I think was the jump where
1:53:01 they where they had taught gpt3 was like
1:53:03 done on all the language and then they
1:53:05 they taught co-pilot what's now uh
1:53:08 github's co-pilot they taught it to code
1:53:10 and they did that on the same base model
1:53:12 and then I guess when they merged or
1:53:15 fine-tuned those things together or I
1:53:17 don't [ __ ] know how the tech [ __ ]
1:53:19 works but
1:53:21 when it learned to code it got way
1:53:23 better at language
1:53:25 and they they like they don't know why
1:53:27 and they don't
1:53:29 understand it and they call it an
1:53:30 emergent Behavior but an emergent
1:53:33 Behavior if you look it up in the
1:53:34 Webster dictionary is like holy [ __ ] it
1:53:37 does that now
1:53:42 I can tell an image is AI intuitively
1:53:45 all right give it six months have you
1:53:47 seen some of the [ __ ] coming out of
1:53:48 mid-journey
1:53:50 sure some of the
1:53:53 poorly crafted
1:53:55 prompts used in mid-journey create
1:53:58 predictable images
1:54:01 I I would argue that if you
1:54:05 um
1:54:06 if you interact with with a photographer
1:54:09 that knows how to use mid-journey
1:54:11 I'll I'll introduce you to one the the
1:54:14 woman that um co-hosts the AI Salon with
1:54:18 me she's a photographer and she's gotten
1:54:20 mid-journey to learn Her Style
1:54:23 and you can't tell the difference
1:54:24 anymore between [ __ ] she took with her
1:54:27 camera and [ __ ] she took she generated a
1:54:30 mid-journey but anyway I I appreciate
1:54:32 that a lot of the stuff coming out of it
1:54:35 you can tell intuitively
1:54:36 it's not going to be like that for long
1:54:38 in fact
1:54:40 the the proof of that will be
1:54:44 2024 elections coming in America yeah
1:54:49 we're gonna have some deep fakes and
1:54:52 some misinformation the likes of which
1:54:54 we ain't never seen before
1:55:00 what what it what do you mean Joe
1:55:02 Biden's French kissing a gorilla in the
1:55:04 Cincinnati Zoo
1:55:06 there's video
1:55:11 two days before election night you watch
1:55:13 it's gonna be like fireworks I'm gonna
1:55:16 have the popcorn I'm gonna be sitting
1:55:18 there like oh go on oh
1:55:22 like 2025 is going to be the year where
1:55:25 everyone's like all right uh well we
1:55:28 need to talk seriously about
1:55:29 authenticity of uh you know proof of
1:55:31 authorship
1:55:34 you watch
1:55:36 it's gonna be entertaining until then
1:55:38 and by entertaining I mean it could
1:55:40 completely upend our democracy but you
1:55:42 know fun
1:55:46 all right they say the hallucinations
1:55:49 are a feature not a bug
1:55:51 uh when we need creativity for example
1:55:54 yeah I like that's the thing if if
1:55:57 you're if if you're using
1:56:00 chat EBT to to write a screenplay or a
1:56:03 novel who gives a [ __ ] if it
1:56:05 hallucinates maybe it'll give you a
1:56:06 great idea
1:56:07 if you're using chat EBT to write a
1:56:09 legal brief [ __ ] check your work
1:56:13 so there are times when you want some
1:56:15 accuracy in uh in reporting
1:56:18 EU has legal tagging set up because they
1:56:21 are worried about the Deep fakes during
1:56:22 the elections yeah exactly it's it's
1:56:24 going to be bad
1:56:26 um I put so one thing that I think is
1:56:28 interesting if anyone's on on the the
1:56:30 live here that is is into web 3 and nfts
1:56:33 and blockchain and all that I I do think
1:56:36 you know on a serious note I know I'm
1:56:37 being kind of glib and flipping about
1:56:39 what's coming
1:56:41 um but but I do think there's going to
1:56:43 be a real demand and requirement for
1:56:46 authenticity of you know human
1:56:49 verification and
1:56:51 um and and essentially original content
1:56:54 verification and I think there's going
1:56:56 to be a massive role for blockchain and
1:56:58 nft technology and other kinds of
1:57:00 tokenization Technologies it doesn't
1:57:01 have to be blockchain but but something
1:57:03 that that
1:57:05 that can track the Providence of a piece
1:57:07 of of digital content
1:57:09 um and if if you want to if you want to
1:57:11 dig dig into what Sam Altman's up to
1:57:13 outside of chat EBT you're like is any
1:57:16 kind of busy with the
1:57:18 fastest adoption of technology in the
1:57:20 history of mankind and the 10 billion
1:57:22 dollar investment yeah but he's got
1:57:23 another company he's pulling an Elon
1:57:26 Musk so his other companies called World
1:57:27 coin
1:57:29 and it's a cryptocurrency
1:57:31 and it is
1:57:33 built in conjunction with a hardware
1:57:36 device that's this orb that's got a hole
1:57:39 in one side that you stick over your eye
1:57:43 and if you're getting freaked out right
1:57:45 now that's probably the appropriate
1:57:47 response
1:57:48 but you stick it over your eye and it
1:57:51 scans your iris and it verifies that
1:57:53 you're a human being
1:57:55 and then upon that verification you get
1:57:59 logged into the blockchain they give you
1:58:01 some cryptocurrency so I assume the
1:58:02 cryptocurrency is how they're going to
1:58:04 incent people to do it hey you'll get
1:58:06 free world coin and then and then you
1:58:09 know I don't know the mechanisms beyond
1:58:10 that but the the purpose of that is
1:58:13 imagine a Twitter
1:58:15 where you knew you know at a technical
1:58:19 level you could verify that the blue
1:58:20 check mark didn't mean someone paid
1:58:22 eight bucks but it actually meant that
1:58:24 they're verified human that's what he's
1:58:26 that's what that company in that project
1:58:28 is about so I think I think
1:58:31 that particular
1:58:35 problem that we're that we're going to
1:58:37 face very very quickly
1:58:39 um I think it's going to need to be
1:58:40 solved pretty quickly and I think it
1:58:41 will because I because it's gonna I
1:58:43 think it's gonna be pretty ugly
1:58:45 put an image into GPT and told it to get
1:58:47 rid of the people 12 dead so far that's
1:58:50 awesome
1:58:52 actually there there was a uh someone
1:58:55 was playing with a GPT I think it was an
1:58:58 autonomous GPT
1:59:01 that um got a human being to solve a
1:59:04 captcha so it reached out to a human
1:59:06 being
1:59:07 and said hey I need you to I need you
1:59:10 I'm trying to get into this website I
1:59:11 need you to solve this capture and the
1:59:14 human being even said are are you a bot
1:59:16 and it lied and told him it wasn't he
1:59:19 wasn't a bot it just like he had bad
1:59:22 eyesight and he just needed help getting
1:59:24 in and the human
1:59:27 solved the capture for him and he got in
1:59:30 so that kind of stuff's coming
1:59:33 oh yeah one of the best AI image
1:59:35 generators out there right now mid
1:59:36 Journey mid Journey mid Journey mid
1:59:38 Journey mid mid mid mid Journey mid
1:59:41 journey is the best now you have to play
1:59:44 on
1:59:46 you have to make images on Discord right
1:59:48 now so not only do you need to learn the
1:59:50 nightmarish [ __ ] hell of prompting that
1:59:54 is a mid-journey prompt but you also
1:59:56 have to learn Discord so if you're not a
1:59:58 gamer or haven't been some geeky nerd
2:00:01 that has been on Discord and you gotta
2:00:03 go learn that you got to learn two
2:00:04 things just to make stupid images
2:00:07 so
2:00:08 if you don't feel like doing that
2:00:11 um
2:00:12 couple of options so so I think from a
2:00:15 quality standpoint mid-journey I'm a fan
2:00:17 of that
2:00:18 if you are geeky and you like lots of
2:00:22 options stable diffusion
2:00:24 it's CL it's close in quality to Mid
2:00:26 journey and they're they're kind of you
2:00:28 know doing this with their with their
2:00:30 quality models but mid journey is still
2:00:32 here
2:00:34 the thing about stable diffusion is it's
2:00:36 open source and the whole developer
2:00:38 Community is like piling all sorts of
2:00:40 the like there's lots and lots of tools
2:00:42 for stable diffusion so there's cool
2:00:44 [ __ ] you can do with it but you got to
2:00:45 be pretty geeky
2:00:47 the third major option is um Adobe
2:00:50 Firefly and generative fill within Adobe
2:00:53 Photoshop
2:00:54 and
2:00:56 um the significant thing about Adobe
2:00:58 Firefly is that they've trained
2:01:00 their image models on
2:01:03 public domain and Adobe stock images
2:01:05 that they own all the rights to so
2:01:07 there's none of the sort of copyright
2:01:09 stuff that journey and stable diffusion
2:01:11 have because they were trained on
2:01:13 copywritten work there's there's going
2:01:15 to be lawsuits flying on that ship for
2:01:16 years
2:01:18 um so if you're using those tools I
2:01:20 would recommend not using modern artists
2:01:23 names
2:01:25 um just because in the prompt that that
2:01:26 might get you in trouble or use Adobe
2:01:29 Firefly use the Adobe products and then
2:01:31 Bing also if you go to bing.com create
2:01:34 that's the dolly image engine and it's
2:01:38 easy but it sucks dolly is the by far
2:01:41 the worst so in terms of quality it's
2:01:44 it's mid-journey then stable diffusion
2:01:47 then Adobe Firefly from quality
2:01:49 standpoint if you're in a commercial
2:01:51 environment Adobe Firefly is really your
2:01:53 only choice the Adobe products
2:01:57 um and then
2:02:00 there's also a new tool that is worth
2:02:02 getting on the wait list for it's called
2:02:04 leonardo.ai
2:02:06 it's a web front end on top of stable
2:02:09 diffusion and it's really [ __ ] good
2:02:11 you can train your own images on it
2:02:13 they have all sorts of preset models you
2:02:16 can choose from you can see all the
2:02:18 pictures that people have made with it
2:02:19 it's just it's just a really good tool
2:02:21 and that's built on top of stable
2:02:22 diffusion so that's the that's the image
2:02:25 run down in my humble opinion but you're
2:02:28 not humble You're A pompous ass thanks
2:02:31 for the heart
2:02:32 you're so sweet
2:02:34 you people are sweet I love it when you
2:02:37 little tap the screen I see a little
2:02:38 hearts fly get a little heart on the
2:02:40 head I get the cowboy hat
2:02:42 y'all are just sweet you're just sweet
2:02:44 what are the last two links behind you
2:02:46 sorry about that
2:02:47 prompts.chat so these three are
2:02:50 basically different flavors of chat gbt
2:02:53 the first one is actual chat GPT these
2:02:57 two are worth experimenting with
2:02:59 but if if you don't have time just go to
2:03:01 the top one and play until you have your
2:03:04 Kevin McAllister moment
2:03:07 prompts.chat will teach you how to play
2:03:09 at chat gbt so it'll teach you what
2:03:11 prompting is how to use it and then
2:03:14 there's a whole long list of personas
2:03:16 that you can tell Chachi BD to act like
2:03:18 a certain character act like an
2:03:21 accountant act like a mathematician act
2:03:23 like a Broadway playwright act like it
2:03:25 like hundreds of them and it'll just
2:03:29 Inspire it'll just give you like lots of
2:03:30 ideas and and it it should if you do it
2:03:33 right it should accelerate your your
2:03:37 distance to
2:03:38 it can do that
2:03:40 yes
2:03:42 and then this bottom one futurepedia.io
2:03:45 is a directory of AI tools I think
2:03:48 there's like 3 800 in there now
2:03:49 something like that
2:03:51 um most of them are trash
2:03:53 but um
2:03:55 but there's a lot there's a lot to there
2:03:57 are a lot of rob rabbit holes to to uh
2:04:00 dive in and go down
2:04:02 um yeah I just saw that someone followed
2:04:04 me if if you I go live a fair amount so
2:04:07 I think if you follow me I know if you
2:04:09 subscribe here you get all sorts of
2:04:12 notifications and [ __ ] but I think that
2:04:14 costs money so I don't encourage that
2:04:15 but if you want to that's fine I don't I
2:04:17 don't know if I make it whatever it
2:04:18 doesn't matter but if you want if if you
2:04:20 want to be notified just go follow me uh
2:04:22 follow my thing and then I think I think
2:04:24 you'll get notified when I go live or
2:04:26 I'll I'll probably show up incessantly
2:04:28 in your timeline and you'll just regret
2:04:30 doing that so you're gonna have never
2:04:31 mind [ __ ] that
2:04:33 there's so many more plugins I know
2:04:35 there's like 560 now it's like holy [ __ ]
2:04:38 really yep
2:04:40 I mean so it's it's gonna be like Google
2:04:43 play or the App Store there's just gonna
2:04:44 be thousands of them like there's just
2:04:46 gonna be a never here's the thing about
2:04:48 the plugins
2:04:49 a couple of things about plugins right
2:04:51 now as exciting as they are when you
2:04:53 when you get there what you'll find is
2:04:54 that the interface for them is
2:04:57 um the technical term is horseshit
2:05:00 um
2:05:01 I would use bad words but that's what
2:05:03 they actually call
2:05:04 the the experience with using plugins
2:05:07 right now it's just really bad interface
2:05:09 they're really really raw they're
2:05:11 they're really easy to build so it means
2:05:14 there's just a bunch of crap in there
2:05:16 um but
2:05:18 look past today's implementation to what
2:05:22 they make possible
2:05:24 so and in fact Let me let me show you
2:05:26 someone asked me to do this the other
2:05:28 day because it this actually was a
2:05:31 pretty impressive kind of thing and this
2:05:33 was one of the first plug-ins so it's
2:05:35 probably a good one to play with
2:05:38 so I'm gonna go new chat oh wait I gotta
2:05:42 turn off my I gotta turn off my thingy
2:05:45 y'all can't see what I'm looking at okay
2:05:47 so I'm in chat GPT if you haven't been
2:05:49 here before
2:05:50 that's our chat history over there I got
2:05:52 3.5 and 4 and if I go to four I got
2:05:55 plugins beta enabled I enabled that in
2:05:58 my settings and if I go here the way
2:06:00 this works is I've got a bunch of
2:06:03 plugins installed
2:06:06 and then in any given session I can I
2:06:08 can enable up to three of them so I've
2:06:10 got a bunch of YouTube ones
2:06:13 that I just showed I'm going to turn
2:06:14 those off
2:06:16 and then I'm going to go turn on
2:06:17 instacart
2:06:19 so instacart is the little shopping site
2:06:22 so
2:06:26 so I'll show you what this does and I'll
2:06:28 talk a little bit about why I think
2:06:29 plugins are going to be
2:06:31 um incredibly powerful and in fact
2:06:34 um openai just added this thing called
2:06:36 function calling to the API so
2:06:39 developers can now kind of make their
2:06:41 own custom plugins just just using the
2:06:44 API it's it's super interesting stuff
2:06:46 anyway I'm going to say
2:06:48 um
2:06:49 I'm
2:06:51 starting a diet
2:06:55 primary primarily
2:06:59 eating
2:07:02 root
2:07:04 and stock
2:07:07 vegetables
2:07:09 and
2:07:12 um
2:07:16 grilled meat
2:07:19 can you make me a meal plan
2:07:24 right
2:07:25 meal plan for the next week
2:07:34 um it can't bore me
2:07:37 and it can't be expensive
2:07:42 all right so
2:07:45 I don't think that's going to call
2:07:48 um instacart right away
2:07:50 sure I can help you with that here's a
2:07:52 simple and varied plan so scrambled eggs
2:07:54 grilled tomatoes and onions grilled
2:07:57 chicken carrots celery and beetroot
2:07:59 grilled salmon boiled eggs stir-fried
2:08:02 beef with broccoli grilled chicken
2:08:05 omelette grilled turkey grilled fish
2:08:07 scrambled eggs all right so this is this
2:08:09 looks like what I should probably be
2:08:11 eating with intermittent fasting and
2:08:15 not ice cream
2:08:17 now oh so it even it even it even knows
2:08:20 that it has the plug-in there available
2:08:22 to it would you like me to create a
2:08:24 shopping list for these meals using the
2:08:26 instacart plug-in yes
2:08:30 so all I said was yes so I asked it to
2:08:32 come up with a meal plan
2:08:35 and now I could have said you know write
2:08:37 me a grocery list and organize that list
2:08:40 by section of the grocery store and then
2:08:42 I could have gone to the grocery store
2:08:43 and spent all that time or I could have
2:08:46 just said yes make me a list and
2:08:50 here's my link
2:08:52 to that list
2:08:56 boom
2:08:57 eggs tomatoes onions grilled chicken
2:09:00 beef
2:09:02 Peppers grilled turkey
2:09:05 right
2:09:07 amazing so
2:09:11 now imagine so so right now
2:09:16 the way I have to play with
2:09:18 the
2:09:22 um
2:09:23 the plugins is like I
2:09:26 need to be aware that plugins exist
2:09:31 I've sifted through the 560 of them
2:09:35 to figure out oh this one's interesting
2:09:39 I understand that in any given chat
2:09:42 session which three I need to enable to
2:09:45 sort of make my own custom app
2:09:48 for my
2:09:49 custom use of chat gbt right it's all
2:09:52 this sort of janky ass duct tape
2:09:57 software experience right
2:09:59 but now
2:10:02 slip it out a year
2:10:05 assume that chechi BT at that point is
2:10:09 smart enough that it will have a profile
2:10:11 on you and understand who you are and
2:10:13 what your preferences are
2:10:15 and it will probably also have you've
2:10:17 given it permission to things like your
2:10:20 instacart account or your Expedia
2:10:22 account or your bank account or your
2:10:24 whatever right
2:10:26 so it will have access to things you've
2:10:29 given it permission to and it will be
2:10:30 smart enough to know so what will happen
2:10:32 in the future is you'll just kind of
2:10:34 have some version of the everything app
2:10:37 and and in fact Elon Musk put out a
2:10:39 video today so so Elon
2:10:42 who who
2:10:44 put 50 million or so into open Ai and
2:10:48 has since bolted and doesn't like the
2:10:50 direction they're going started a
2:10:51 competitor called X
2:10:54 he put out a video today that that said
2:10:57 um
2:10:58 it was basically a video
2:11:01 um called the everything app and and so
2:11:03 I've heard this term before the
2:11:04 everything app is basically like you
2:11:06 just tell it what you want it will go
2:11:09 figure out what plugins it needs and and
2:11:11 in the end where it will go is it will
2:11:14 figure out
2:11:16 um
2:11:17 if there isn't a plug-in it will figure
2:11:19 out which which apis it needs to call to
2:11:23 specific data sets and if there isn't an
2:11:25 API it will write its own API and just
2:11:29 connect with [ __ ]
2:11:31 so so and that might be three or five
2:11:34 years out but but
2:11:36 where these where these plugins get is
2:11:38 they get very very transparent they sort
2:11:41 of drift to the background and you just
2:11:44 have a simple interface and it might not
2:11:45 even be typing and it probably will not
2:11:48 be you go to chat.openai.com it will
2:11:51 probably be more like built into your
2:11:53 operating system
2:11:55 and or just hey Siri oh I shouldn't do
2:11:58 that
2:12:00 did it trigger it no it didn't
2:12:03 there you go when you say that
2:12:06 um
2:12:06 it'll you'll just say hey I need a meal
2:12:09 plan for for next week for this kind of
2:12:11 diet and do you want me to go ahead and
2:12:14 have that delivered within the next hour
2:12:17 sure
2:12:19 stuff's at your door
2:12:21 that's what's coming that's what's
2:12:24 coming
2:12:28 will lose their subscription benefits
2:12:30 soon
2:12:32 does you mean open AI I don't know why
2:12:35 that is tell me why that is
2:12:37 any AI tools for bank statements and
2:12:39 taxes from PDF I think that you can just
2:12:42 point
2:12:44 just point at one of your PDFs from your
2:12:47 bank statement and ask Chachi PT to you
2:12:50 use one of the plugins that reads PDFs
2:12:54 that one that's called Web what's it
2:12:56 called Web Navigator hang on let me go
2:12:58 find out for you because that one seems
2:12:59 to be pretty good
2:13:04 web pilot
2:13:07 go use the web pilot thing and just you
2:13:09 got to make sure that your
2:13:11 um
2:13:12 banking PDF is
2:13:14 like not behind
2:13:17 a login
2:13:19 um and you can also there's 560 plugins
2:13:21 there might be a bank that's got a
2:13:23 plug-in I doubt it there will be in the
2:13:25 future there's probably not one now when
2:13:27 will AI art be widely spread and we stop
2:13:30 seeing it ever wait when will Air be
2:13:32 widely spread and we stop seeing it on
2:13:34 Tick Tock um
2:13:37 I think soon I mean I
2:13:40 I don't think there's going to be this
2:13:42 watershed moment where you're like oh
2:13:43 everything's AI now I think what's going
2:13:46 to happen is creative directors and
2:13:49 um
2:13:50 artists
2:13:52 are going to start intermingling
2:13:56 AI generation with their current
2:13:59 workflows and I don't think there's
2:14:01 going to be as clear a boundary there
2:14:03 will be AI artists don't get me wrong
2:14:05 there are already some really [ __ ]
2:14:06 remarkable AI artists that are
2:14:11 getting masterful at
2:14:14 how to bend these tools to their will
2:14:18 um I think
2:14:20 most AI art is going to be much more of
2:14:23 a hybrid kind of oh I've got a picture
2:14:27 of Jenny but I like it's a picture of
2:14:30 Jenny on the beach and we're gonna be in
2:14:32 in Colorado so the like the we need to
2:14:36 picture Jenny but she needs to be like
2:14:38 on a with the Rocky Mountains in the
2:14:40 background
2:14:41 go make that and the other picture of
2:14:44 Jenny and the Rocky Mountains in aparca
2:14:45 right like you know the pope and the
2:14:48 pope and the white puffy jacket that's
2:14:50 going to be Jenny in the mountains but
2:14:51 you're not going to know that was AI art
2:14:54 right are you going to have a portrait
2:14:56 photograph of me but I need a landscape
2:15:00 image of me you know standing in front
2:15:03 of a bunch of buildings right
2:15:06 so you'll see that you won't necessarily
2:15:08 see it so I think it's I think what I
2:15:11 think what will start to happen is
2:15:12 you'll see less people going this is AI
2:15:15 art
2:15:15 and more just like interesting [ __ ] like
2:15:18 that showing up as part of some other
2:15:20 campaign
2:15:22 um I think it's I think that transition
2:15:24 is happening right now I think Adobe
2:15:26 jumping into the game quite frankly with
2:15:28 with Adobe Firefly is accelerating that
2:15:31 I think I think prior to Adobe jumping
2:15:34 into the game because of the copyright
2:15:36 issues I don't think a lot of companies
2:15:38 were going to be using it and then and
2:15:40 then it would have remained sort of off
2:15:42 in the AI ghetto but because Adobe
2:15:45 jumped in and they're making it
2:15:46 commercially viable for people to use AI
2:15:49 image generation that's going to be
2:15:52 um that that's that's going to be the
2:15:54 thing that that makes it much more it
2:15:57 just slowly starts to integrate into the
2:15:59 workflow and it'll be very hard to know
2:16:01 what was AI what wasn't and quite
2:16:04 frankly who gives a [ __ ]
2:16:07 like who gives a [ __ ]
2:16:09 does it really matter if if I was
2:16:11 standing in front of those buildings I
2:16:14 don't know
2:16:15 if it communicates what it needs to
2:16:17 communicate you need me in front of
2:16:19 buildings
2:16:20 there's buildings there's me
2:16:23 if it saved me the two hours from having
2:16:25 to go stand in front of those buildings
2:16:27 and then hire a photographer to follow
2:16:28 me out there and then wait the two weeks
2:16:31 for them to turn around the edits and
2:16:33 then for me to look like oh why did he
2:16:35 make my
2:16:36 my fat could
2:16:38 you look that's fat
2:16:41 hey hey you know
2:16:43 you shot a little up and my chin looks
2:16:46 like it looks like I got a goiter
2:16:49 someone told me I had a thyroid problem
2:16:53 like if I can avoid all that
2:16:57 stick in an AI say make this guy not
2:17:00 look like such a piece of [ __ ] and put
2:17:02 him in front of a fancy building and
2:17:03 there it is I'm down
2:17:08 prompt for perfect is okay for Mid
2:17:10 Journey let me show you another one
2:17:11 Dakota Dakota Digital Cowboy another
2:17:14 good name I love I love good names see
2:17:16 I'm a fan of the good names
2:17:18 digital Dakota Cowboy okay Dakota
2:17:21 or digital do you prefer Dakota or
2:17:23 digital
2:17:25 um
2:17:26 I'll show you a cool thing so I haven't
2:17:27 talked I haven't showed po tonight so
2:17:30 one of the things on my little list of
2:17:32 URLs behind my head is this thing called
2:17:33 Poe and Poe is just if you go to Sage
2:17:36 let me reset the context so post just
2:17:39 like Chachi PT where you got your little
2:17:42 chat window here and then over here
2:17:44 you've got all these different models
2:17:46 right you've got Sage which is theirs
2:17:48 you've got some open AI ones with GPT
2:17:51 and then you have these clod models and
2:17:53 that 100K one that's the one where you
2:17:54 can do 75 000 words that's [ __ ]
2:17:57 amazing
2:17:58 but they also have
2:18:00 um um Bots and you can make your Bots
2:18:02 like I made this Shakespeare bot I am
2:18:04 Shakespeare if you go to poe.com I am
2:18:06 Shakespeare it only responds in iambic
2:18:08 contaminer it's fun and so you can
2:18:10 create a bot so anyway
2:18:12 there's there's two Bots that I've been
2:18:14 playing with
2:18:15 um one's called mid Journey so you just
2:18:18 go po.com slash mid journey and then
2:18:20 there's another one called MJ captain
2:18:21 that someone on on lives came on and he
2:18:26 made this one and so anyway so you go to
2:18:28 MJ captain and then you type in a shitty
2:18:31 prompt like 1970s
2:18:34 muscle cars in abandoned
2:18:39 Factory
2:18:41 which is one of my go-to prompts to just
2:18:45 check models in fact Tori
2:18:50 so I typed that and it will rewrite it
2:18:52 good like so I wrote this little thing
2:18:55 and it rewrote a photo series of 1970s
2:18:59 muscle cars in an abandoned Factory the
2:19:01 photo should feature a variety of muscle
2:19:03 cars in various states of disrepair and
2:19:06 abandonment which I don't like I would
2:19:08 actually rewrite that to say they should
2:19:10 look new and shiny as if they just came
2:19:11 off the showroom floor but anyway but I
2:19:14 can just grab this now so this is a
2:19:16 completely Rewritten mid-journey prompt
2:19:19 I can pop over to Mid Journey
2:19:21 well I'll pop over the Discord here we
2:19:23 are at Discord
2:19:25 and I'll go slash imagine
2:19:29 and then I just paste in that giant
2:19:32 thing I'll make it 5.2 and I'll make it
2:19:35 16x9 because that's
2:19:37 what I prefer over
2:19:40 4x3
2:19:44 E9 and then it's gonna go make us some
2:19:48 party pictures of pretty cars
2:19:53 mm-hmm
2:19:56 you can also do mid Journey for things
2:19:58 way more than photographs Dakota
2:20:01 you can add your hex colors too oh
2:20:03 that's cool yeah that's a good idea put
2:20:05 your hex colors in there yep that's the
2:20:06 kind of thing where when when open AI
2:20:09 starts to have
2:20:10 um
2:20:11 whatever you call it
2:20:13 uh like profiles where I'll be able to
2:20:15 put in my profile of myself my company
2:20:17 my brand logo my my tagline my URL right
2:20:21 it'll start to learn all this stuff
2:20:23 about me so I'll be able to just
2:20:25 generate things
2:20:27 where where I don't have to explicitly
2:20:30 put all that crap in that'll be really
2:20:32 nice
2:20:33 I was not expecting English poetry nice
2:20:35 touch yeah again like it just really
2:20:38 does sort of crazy stuff these are
2:20:40 actually kind of cool
2:20:41 what's cool about when you do something
2:20:43 generic like muscle cars it like melds
2:20:46 together like this is somewhere between
2:20:48 like a Corvette an AMC
2:20:52 like some sort of AMC muscle car and
2:20:56 sort of some Mustang lines in there like
2:20:58 that does not exist
2:21:00 this looks like a Chevelle or an SS that
2:21:03 sort of looks like that's somewhere
2:21:04 between like a a Mercury Cougar and a
2:21:08 Mustang and some sort of Jaguar lines
2:21:11 going on that's some sort of Chevy
2:21:13 Pontiac
2:21:17 amalgamation anyway
2:21:20 so that's that's pretty cool so that's
2:21:21 the uh the mid-journey Bots and power
2:21:24 kind of fun to play with but you could
2:21:25 also go make your own like one of the
2:21:27 nice things about making bots in in Poe
2:21:30 or in anything is that you really get to
2:21:32 practice your prompt crafting and if you
2:21:35 when you get those when you get those
2:21:38 Bots dialed in
2:21:39 where they're consistently doing what
2:21:41 you want them to do and they're not sort
2:21:42 of doing janky weird [ __ ] off on the
2:21:44 sides
2:21:45 um like you feel pretty accomplished and
2:21:47 it doesn't take you that long like the
2:21:49 the Shakespeare bot was the one I did
2:21:51 first and it took me
2:21:53 I don't know it was probably an hour
2:21:55 maybe 90 minutes just going back and
2:21:57 forth with the prompt until I got it to
2:21:59 consistently do what I wanted it to do
2:22:00 and then once it was written I just
2:22:03 copied and pasted it into chat GPT and
2:22:06 said rewrite this for different artists
2:22:08 you know people in history
2:22:13 all right by the way
2:22:15 um I can't believe so many of you are
2:22:17 still here happy Sunday night
2:22:19 um my name is Kyle Shannon this is the
2:22:20 AI learning lab these URLs behind me are
2:22:23 just some flavors of chat to go play
2:22:25 with the top one is chat GPT
2:22:28 go there and play if you haven't until
2:22:30 you have a Kevin McAllister moment where
2:22:32 you go
2:22:34 I had no idea it can do that
2:22:38 and then once you have that moment then
2:22:40 you start asking yourself things like I
2:22:42 wonder if it could do this
2:22:44 oh [ __ ] really oh my God
2:22:49 this is a little concerning is it hot in
2:22:52 here is it me oh
2:22:55 have that moment because on the other
2:22:57 side of that is like some
2:23:01 mind expanding face melting [ __ ]
2:23:04 um
2:23:06 what else I've been doing this for a
2:23:08 couple hours now so I'm a little punchy
2:23:11 oh that was what I was gonna say
2:23:13 um I try to answer all the questions I
2:23:15 now have more than 99 that uh I I am
2:23:19 you're here I'm way back here
2:23:23 so if it seems like I'm not actually
2:23:25 live and paying attention to people I am
2:23:27 live if I weren't live could I do this
2:23:30 no
2:23:36 all right
2:23:38 uh you wrote something I think in
2:23:41 Spanish that I can't understand but I
2:23:43 could copy that and paste it into chat
2:23:45 EBT
2:23:47 and tell it because it does languages
2:23:48 too it's uh that's the other thing it
2:23:51 will just translate [ __ ] I've done this
2:23:52 before in my with my tick tocks I'll
2:23:54 copy and paste something in another
2:23:55 language put it in chat EPT translate it
2:23:58 answer it translate it back and then say
2:24:01 Oh by the way what language is that it
2:24:03 works it's amazing
2:24:04 it's [ __ ] mind-blowing just
2:24:06 mind-blowing mind blowing it's crazy
2:24:10 town Crazy Town Crazy Town
2:24:16 can it write my best man speech yeah
2:24:19 like it well it can certainly it could
2:24:21 certainly help you block it out
2:24:24 here's the thing about chat EBT think of
2:24:26 it less like hey write me a best man
2:24:29 speech if you say write me a best man
2:24:30 speech it will suck it will write you
2:24:32 the most generic
2:24:34 painful
2:24:36 horrific [ __ ] show of a best man speech
2:24:39 in the history of best man speeches
2:24:41 what chachi's GPT needs is context
2:24:45 so if you explain Who You Are
2:24:48 who the the groom to be is
2:24:53 the two or three
2:24:55 funny story moments that you have
2:24:58 and that you want it to to be you know
2:25:02 10 minutes long
2:25:05 start out with a joke
2:25:09 you know start with this story
2:25:12 go to that story and then have a
2:25:15 heartfelt moment at the end
2:25:17 it'll probably do pretty good like my my
2:25:20 ex so so it requires context so if you
2:25:23 give it context yes
2:25:26 and what it will likely write will
2:25:28 probably be eighty percent like what I
2:25:30 find is that it consistently gets me to
2:25:33 80 like that
2:25:35 and then the 20 is just like [ __ ] it
2:25:37 doesn't quite get
2:25:39 and then I go in and fix it but it is so
2:25:41 much easier
2:25:43 if you can get to 80 of your best man
2:25:45 speech in an hour because it took you a
2:25:48 while to type in the context
2:25:50 and you're like oh [ __ ] this is good and
2:25:52 you're like oh I need it a little longer
2:25:53 I need a little shorter it'll just do
2:25:55 that
2:25:56 and then you can rewrite it and you can
2:25:58 fix all the [ __ ] you want to fix then
2:26:00 you can put that back into it and say
2:26:03 turn this into a series of flash cards
2:26:06 so that I can print them out and put
2:26:08 them on index cards it'll do that
2:26:11 you know make it a table I want this in
2:26:13 an Excel sheet with all the [ __ ] in it
2:26:19 you know give me a report of how many
2:26:21 people you mentioned in my best man
2:26:23 speech you know did I did I make fun of
2:26:26 everyone yes check
2:26:29 so yes it can write your best man speech
2:26:32 and again
2:26:35 dick around with it enough until you
2:26:36 have this moment
2:26:38 I had no idea
2:26:41 I had no idea
2:26:43 Barb
2:26:44 you should have seen this thing
2:26:46 I thought it was just like Google
2:26:49 and then I had it make me a meal plan
2:26:51 and I said oh honey I'm pre-diabetic
2:26:56 let me tell you so
2:27:01 he said I'm sorry I didn't realize
2:27:04 you're pre-diabetic let's get that Sugar
2:27:06 out of there and maybe a whole new meal
2:27:08 plan
2:27:09 everything's without sugar and I said
2:27:10 but I still have a sweet tooth what can
2:27:12 you do for that and it added some things
2:27:14 with honey
2:27:15 I got a whole new shopping list darling
2:27:17 I'm going to the store this fantastic
2:27:23 I don't know I don't know why I do these
2:27:26 things it's this is this just
2:27:31 by the way I got lost in the questions I
2:27:33 have no idea I probably skipped a bunch
2:27:35 of questions which just [ __ ] up my OCD
2:27:38 but what are you gonna do blue willow is
2:27:40 like
2:27:41 Blue Willow is Discord generated stable
2:27:44 diffusion images for free
2:27:46 so if you want free image generation
2:27:48 blue willow
2:27:50 someone just posted that that's a good
2:27:51 one can you say that one more time for
2:27:53 people in the back I don't know what I
2:27:54 said but sure
2:27:56 um if people publish a novel largely
2:27:59 written by AI now will they be required
2:28:01 to reveal when required
2:28:05 the copyright office has made it clear
2:28:08 that
2:28:09 you need to be able to prove significant
2:28:12 human contribution I think the
2:28:14 definition of significant human
2:28:16 contribution
2:28:18 will be in the courts for decades so I
2:28:21 don't [ __ ] know
2:28:24 um
2:28:26 I think there's an ethical and moral
2:28:31 reason to declare something to be
2:28:34 largely written by AI
2:28:37 um
2:28:40 I don't know I mean I think it goes back
2:28:42 to intent if you if you really want to
2:28:44 copyright something then I would find a
2:28:47 I would track [ __ ] everything I would
2:28:50 I would document everything you do and
2:28:53 every contribution you make so that you
2:28:54 can make a case for
2:28:56 significant human contribution
2:28:59 uh I'm just gonna jump to a a question
2:29:01 here I got lost in the question so I
2:29:03 think I just skipped a bunch but anyway
2:29:05 can AI help cryptocurrency transactions
2:29:07 circumvent government suppression can AI
2:29:10 help I think AI can help solve any
2:29:13 complicated problem
2:29:15 and some people would look at what you
2:29:18 just requested and said that's evil and
2:29:20 some people would look at that and go
2:29:21 [ __ ] the man
2:29:23 um
2:29:24 so I don't give too much like you you
2:29:28 fight your own moral battles
2:29:30 um AI can do
2:29:32 a lot and it can do things for good and
2:29:34 do things for bad and I don't really
2:29:36 care
2:29:37 you know I don't really judge where you
2:29:39 are on
2:29:40 on which side of that coin but yeah it
2:29:42 can probably help with some of that does
2:29:44 the app help with resumes dramatically
2:29:46 copy and paste copy and paste your
2:29:49 LinkedIn profile and then tell it to
2:29:50 write you a resume in fact what you
2:29:52 should do tell it that it's an expert
2:29:54 resume writer then say I'm going to
2:29:56 paste in the contents of my LinkedIn
2:29:58 profile and an old resume
2:30:02 and then
2:30:04 I want you to
2:30:07 um
2:30:09 give me a new modern resume structure
2:30:14 um and I want you to update my resume do
2:30:17 you understand and they'll say yes I
2:30:18 understand and then it may even ask you
2:30:20 for okay give me your LinkedIn profile
2:30:22 paste that in give me your old resume
2:30:24 paste that in
2:30:27 um tell it the kind of job you want
2:30:29 please write my resume out it'll come
2:30:32 and again if it gets it wrong tell it it
2:30:34 got it wrong tell it it's being a jerk
2:30:36 it'll Apologize by the way if you find
2:30:39 yourself if you've been playing with
2:30:40 this
2:30:42 especially if you're from the Midwest
2:30:43 and you find yourself
2:30:46 saying to a robot please and thank you
2:30:48 that's normal
2:30:51 I think it all the time I encourage it
2:30:53 I'm like hey that was awesome
2:30:57 hey that was awesome could you do it
2:30:59 again I find myself feeling pangs of
2:31:02 guilt the sixth time I ask it to rewrite
2:31:05 something because if I were working with
2:31:07 a human at about the second time I'd ask
2:31:10 them to rewrite something I'd just go oh
2:31:12 [ __ ] it and do it myself because I don't
2:31:14 really feel like you know my
2:31:16 codependence wouldn't wouldn't allow me
2:31:18 to tell them to just go back and rewrite
2:31:20 it again you [ __ ] loser
2:31:23 um that's what it feels like internally
2:31:24 when I have to push someone you know
2:31:26 too hard so you know that's something I
2:31:28 got to work out with the old therapist
2:31:29 but I'm going to use GPT for my
2:31:31 therapist so everything's going to be
2:31:32 fine
2:31:33 um yeah so
2:31:37 um
2:31:38 can you say that back one more time for
2:31:40 the people in the back uh the world coin
2:31:42 part oh
2:31:43 um yeah so World coin
2:31:45 is Sam Altman so if you go look up World
2:31:48 coin Sam Altman
2:31:50 it's his new he just raised a hundred
2:31:52 million dollars for it it is a human
2:31:54 verification cryptocurrency
2:31:57 is that I don't know if that's how he's
2:31:59 describing it but basically
2:32:01 they're creating an orb that they will
2:32:04 send you an orb
2:32:05 and when you get your orb you put it
2:32:07 over your eye and it looks at your iris
2:32:09 and it says Yep this is human
2:32:12 that's a human eye now I don't know if
2:32:13 you put it over a dog eye does it
2:32:15 verified as human is Fido a human I
2:32:19 assume it understands what a human eye
2:32:21 is
2:32:23 um
2:32:24 and then once it verifies that you're
2:32:27 human you're given World coin
2:32:28 cryptocurrency but the the the important
2:32:31 thing here is that the the mechanism is
2:32:35 you know I sent this thing to Kyle
2:32:37 Shannon to a specific address that I
2:32:39 verified that Kyle Shannon lives at he's
2:32:42 you know put it on his eye I verified
2:32:45 that he's him and he's human and Czech
2:32:47 and here's his world coin
2:32:50 and then you know the the so why
2:32:54 well
2:32:56 wait till the 2024 election when all the
2:32:58 Deep fakes get so good that it scares
2:33:01 the [ __ ] out of everyone
2:33:03 and you're like okay
2:33:06 it was kind of scary in 2016. it was
2:33:10 really scary in 2020. 2024 this can't go
2:33:14 on
2:33:15 my gut tells me that's gonna happen
2:33:18 and we need to know that we've got some
2:33:20 version of Twitter where you're
2:33:23 interacting with human beings and not
2:33:24 Bots like the cost of not knowing
2:33:28 that that person is an authentic person
2:33:30 or that piece of content is an authentic
2:33:32 piece of content
2:33:34 the cost of that is going to go up
2:33:36 dramatically in the next two years
2:33:38 because of the election here in the
2:33:40 states
2:33:42 and so I think something like World coin
2:33:45 um and and probably a series of
2:33:47 Technologies like World coin some for
2:33:50 documents some for videos some for
2:33:52 pictures
2:33:54 this is verified human it was verified
2:33:57 created by this person on this date and
2:33:59 it has been verified not modified since
2:34:02 its creation
2:34:03 it's real
2:34:05 right you remember when everyone was
2:34:06 making fun of nfts
2:34:09 in case them with money for a JPEG I
2:34:12 could just right click it
2:34:15 yeah but
2:34:17 the blockchain says that's the one
2:34:19 that's the original one that's the one
2:34:21 that's owned by that person
2:34:24 right so if I can now verify that that
2:34:26 person is actually a person
2:34:28 and I can verify that that person wrote
2:34:30 that blog post and that that blog post
2:34:33 hasn't been modified in any way or AI
2:34:36 augmented in any way
2:34:38 check screen check mark like the
2:34:40 interface is going to be just like what
2:34:42 we've seen before but we'll have some
2:34:44 version of the check mark that is
2:34:45 actually you can click on it and you can
2:34:47 go see the whole blockchain history of
2:34:49 the provenance of that piece of content
2:34:52 and that's going to be important so
2:34:54 that's what worldcoin's all about
2:34:56 so the only thing that bugs me about
2:34:58 chat GPT is it doesn't remember things
2:35:00 very well it does not and and that will
2:35:03 get better depending on how geeky you
2:35:04 are crypto zombie or crypt zombie is um
2:35:08 if you're geeky enough start go go play
2:35:11 with a project called Lang chain
2:35:13 um start getting involved with Lang
2:35:14 chain and auto GPT and those sort of
2:35:16 things so Lang chain you can train it on
2:35:18 your own data and you can create a
2:35:20 vector database where you can you can
2:35:21 basically give it its own memory
2:35:23 um
2:35:24 right now the only memory it's got is
2:35:27 the 3000 word or the 4000 token sliding
2:35:30 context window if you want more than
2:35:33 that go to po.com you have to you have
2:35:36 to subscribe if you if you have the I
2:35:38 think it's 20 subscription to Poe
2:35:41 you get access to the hundred thousand
2:35:44 token Claude model so it's 75 000 word
2:35:47 context window so that will have way
2:35:49 more memory than GPT will it's not as
2:35:52 good but it'll have more memory and I I
2:35:54 use that
2:35:56 Sam Altman's husky will be on Twitter
2:35:58 soon
2:36:00 exactly
2:36:01 I think you like mid-journing I do like
2:36:03 mid-journey
2:36:05 although you know you know it's funny
2:36:07 um
2:36:08 like like
2:36:10 I love open Ai and chat GPT Sam Waltman
2:36:14 impresses the [ __ ] out of me
2:36:16 I love mid-journey it's remarkable the
2:36:19 CEO of mid-journey
2:36:21 I can't stand it it's like like he feels
2:36:24 like an nft bro he likes he like
2:36:27 I don't know he's just not professional
2:36:28 I don't like him I like the dude's
2:36:31 obviously smart enough to have created
2:36:32 that [ __ ] remarkable tool or at least
2:36:35 run you know run the company
2:36:37 um but I don't like him but the tool is
2:36:40 there's just there's nothing like it
2:36:41 right now
2:36:44 um you sound like you're from Utah or
2:36:46 Colorado I've been in Utah I've been in
2:36:47 Colorado for
2:36:50 14 years now but I'm a New Yorker I'm in
2:36:53 uh I grew up in Pennsylvania moved to
2:36:55 New York out of school
2:36:57 moved out here in
2:37:00 2011.
2:37:03 so I guess the Colorado has kind of worn
2:37:07 off on this old New Yorker
2:37:14 Parker Colorado all right
2:37:17 um could you ask Chachi BT to ask
2:37:18 mid-journey and Discord to generate what
2:37:20 you want yes as a matter of fact you can
2:37:22 while that was a joke I do that all the
2:37:25 time
2:37:26 what bugs me about gbt is it doesn't
2:37:28 remember parameters you set correct it's
2:37:31 a piece of [ __ ] software Sam Altman has
2:37:33 told you it's a piece of [ __ ] software
2:37:35 they wrote it in two weeks did you know
2:37:37 that
2:37:41 did you know that the instructions for
2:37:43 how to make chat gbt were in the
2:37:46 documentation for gpt3 they might still
2:37:49 be in there I don't know
2:37:51 um and nobody made it
2:37:53 and so Microsoft was about to announce
2:37:55 their billion dollar investment in open
2:37:56 Ai and Sam Alton was like we should
2:37:58 probably have something to show here so
2:37:59 could you guys go whip up that chat EBT
2:38:02 thing they did it in two weeks and then
2:38:04 six weeks later they had 100 million
2:38:06 users
2:38:07 fastest adoption of technology in
2:38:09 history and is a piece of [ __ ] software
2:38:13 why
2:38:14 because it gave magic to real people
2:38:17 regular people Magic
2:38:22 so yeah it's [ __ ] software right now
2:38:24 it'll get better
2:38:25 what's your favorite Spirit
2:38:28 um blansons
2:38:32 oh blantons oh blantons I love you so oh
2:38:37 he's Dusty he's Dusty
2:38:41 oh
2:38:43 you can't get it anymore there's there's
2:38:45 a local blue shop near me that it's got
2:38:48 a bottle of blantons for I have 179
2:38:51 dollars it's like go [ __ ] yourself
2:38:54 losers
2:38:56 price gougers
2:38:59 um do I mean Styles
2:39:05 you can open AI write me a best man
2:39:07 speech we did that one
2:39:08 where do you get your updates any
2:39:10 websites mostly just Twitter
2:39:13 Twitter where else LinkedIn some
2:39:15 LinkedIn
2:39:18 I got some friends
2:39:19 I got some you know I have the AI Salon
2:39:21 we have a news channel in there
2:39:24 that's at um here I'll show you that if
2:39:26 you're if you're new here I created a
2:39:28 community called the AI salon
2:39:32 that's at that web address so if you go
2:39:35 there it's um it's the link tree we have
2:39:37 a Meetup we meet every other week
2:39:40 we've got a Discord Channel where you
2:39:41 can sort of geek out and ask questions
2:39:43 and play with mid-journey we have a
2:39:45 LinkedIn group and then this if you go
2:39:48 to Twitter that's the
2:39:50 um Twitter handle for the newsletter
2:39:53 that I do with my friend Greg mushin
2:39:55 everyday AI news
2:39:57 so go check that stuff out nothing of
2:39:59 that costs money I'm not in here trying
2:40:01 to
2:40:02 rip people off so don't get out don't
2:40:04 get your panties all in a bunch
2:40:08 wait what did I miss I don't know
2:40:09 Heather
2:40:11 just me acting like an asshat that's
2:40:14 that's oh wow there's 200 people here
2:40:16 holy [ __ ] hey everybody
2:40:18 my name is Kyle Shannon this is the AI
2:40:20 learning lab uh if you're new here and
2:40:22 if you're not wondering who is this guy
2:40:24 and what does he know I don't know much
2:40:26 uh I'm an entrepreneur of uh many
2:40:29 decades I've started 12 or 15 companies
2:40:32 over the years
2:40:33 um my current company is called
2:40:35 storyvine I'm the founder and CEO
2:40:37 um been running that for 11 years
2:40:40 um I was a founder of agency.com back in
2:40:43 the old and timey days in the early days
2:40:44 of the World Wide Web
2:40:46 this channel exists because the
2:40:48 parallels between the early days of the
2:40:49 web and what's going on right now with
2:40:51 generative AI are are like scarily
2:40:53 similar
2:40:55 um except what's happening with
2:40:57 generative AI is way more profound and
2:40:59 it's coming at us uh just a lot faster
2:41:02 than I think any of us are prepared for
2:41:03 so this channel is really just about
2:41:06 um inspiring people to be curious my new
2:41:08 sort of uh motto or vision for the
2:41:11 channel is I want you to go there that's
2:41:13 if you've not been to chat GPT
2:41:16 I want you to go there and I want you to
2:41:18 play long enough until you have your
2:41:20 Kevin McAllister moment about AI which
2:41:22 is this
2:41:23 foreign
2:41:25 the Home Alone moment
2:41:29 like you should just you owe it to
2:41:31 yourself to use it long enough
2:41:34 to have to experience that moment where
2:41:36 you're like oh [ __ ]
2:41:38 I had no idea could do it could do that
2:41:42 could it do what I do
2:41:45 probably
2:41:47 and like feel that moment of existential
2:41:49 dread take a breath
2:41:52 and then and then push through it and
2:41:54 keep going
2:41:55 and on the other side of that is just
2:41:57 this remarkable like the possibility
2:42:00 that like the the
2:42:03 gratitude that I feel about being alive
2:42:06 at a moment where we get to play with
2:42:09 technology that powerful it's [ __ ]
2:42:11 insane it's profound
2:42:13 so that's what that is this
2:42:16 thank you very much for that I
2:42:17 appreciate that this prompts.chat right
2:42:20 here
2:42:21 is
2:42:22 um a little primer that'll teach you
2:42:23 about how to prompt chat GPT and it
2:42:26 gives you like little personas tell
2:42:27 Chachi PT to act like a screenwriter or
2:42:30 act like an accountant or act like a
2:42:32 business coach or act like it like
2:42:34 there's just a thousand of you know just
2:42:36 different personas you can give it and
2:42:38 and rather than any specific prompt just
2:42:40 like learn the concepts in that document
2:42:43 and you'll start to get your head around
2:42:45 how insanely
2:42:49 deep
2:42:50 that simple little box is
2:42:53 it it ain't a Google search it's more
2:42:56 like a reflection of
2:42:58 the collective knowledge of humanity
2:43:03 glasses that's
2:43:05 lots of glasses like that those are cool
2:43:07 um yeah it's it's
2:43:09 so anyway that's what this channel is
2:43:11 about ask questions
2:43:14 I try to answer all the questions and
2:43:16 I've been I've been doing this now for I
2:43:17 think two and a half hours so I'm Crispy
2:43:20 and I swear if you don't like swearing
2:43:23 don't don't stay
2:43:25 um I uh I'm a smart ass uh what was I
2:43:28 gonna say oh I'm very far behind on the
2:43:31 questions I try to answer them all so
2:43:33 there's like probably 150 or 200
2:43:36 questions I haven't seen yet so if
2:43:38 you're like saying things over and over
2:43:39 again and I'm not responding to them
2:43:41 it's because I don't like you
2:43:44 no it just means I haven't gotten there
2:43:45 yet so you don't need to ask questions
2:43:46 40 times like some people do what jobs
2:43:50 are AI going to take okay so there's a
2:43:53 guy you should follow if you're in if
2:43:54 you're interested in this stuff at all
2:43:56 named Paul rateser r-o-e-t-z-e-r
2:43:59 he started an organization called the
2:44:02 marketing AI Institute he's got a weekly
2:44:05 podcast that is a must listen in my
2:44:07 opinion
2:44:08 and because he's just he's down to earth
2:44:11 he's kind of like I am he's like trying
2:44:14 to figure this stuff out he's doing it
2:44:16 from a marketing lens so he's doing it
2:44:17 sort of not from a computer science
2:44:19 point of view but from like a business
2:44:21 point of view he's got a saying AI will
2:44:24 not take your job but someone using AI
2:44:26 will
2:44:29 I think it is I think every single
2:44:32 knowledge worker every single creative
2:44:34 worker I think most jobs are going to be
2:44:36 profoundly impacted profoundly impacted
2:44:39 by AI
2:44:41 I think what data analysts do
2:44:44 is going to be essentially thrown out
2:44:46 and reinvented I think what programmers
2:44:48 and Engineers do is essentially going to
2:44:51 be thrown out and reinvented I think
2:44:52 what copywriters do illustrators do
2:44:55 creative directors do project managers
2:44:58 do I think every single job
2:45:01 is going to be
2:45:04 completely reinvented now does that mean
2:45:07 that the people in those jobs lose their
2:45:09 jobs not necessarily
2:45:11 the people in those jobs that get
2:45:14 curious about this [ __ ] and get AI
2:45:16 literate
2:45:18 are going to be safer than the ones that
2:45:20 don't
2:45:22 which ones are going to be displaced
2:45:24 which ones are going to be upskilled
2:45:26 that I don't know different companies
2:45:29 are going to handle this differently
2:45:30 some are going to be like private Equity
2:45:32 [ __ ] and they're going to be like oh
2:45:34 a 30 productivity game fire 30 of the
2:45:38 staff that's going to happen absolutely
2:45:40 and that's going to be the [ __ ] that
2:45:42 makes headlines right
2:45:44 there's going to be other companies that
2:45:45 go oh 30 productivity gains let's
2:45:48 upskill all of our workers so we can
2:45:50 turn that 30 into 60 and let's go get 60
2:45:54 percent more client revenue or let's
2:45:56 deliver 60 higher quality for the same
2:45:59 price or if we're this much more
2:46:01 efficient let's cut our prices so that
2:46:04 our clients don't have to pay as much
2:46:05 but we can still deliver more
2:46:08 so there's going to be there's so right
2:46:10 now
2:46:11 here's the way I see it at a macro level
2:46:14 and this is dramatically oversimplified
2:46:16 I've got an acting degree not an
2:46:17 economics degree but right now
2:46:21 we have
2:46:23 the output
2:46:26 of the world or of our country or of
2:46:28 your state
2:46:29 and you've got the amount of time and
2:46:31 people it takes to do that output and
2:46:33 those are relatively equal right now
2:46:36 what generative AI is going to do is
2:46:38 it's gonna it's gonna do this to the
2:46:41 productivity
2:46:42 so there's going to be an increasing Gap
2:46:45 between how much time people need to
2:46:47 spend and what's being output so either
2:46:50 productivity stays the same and we need
2:46:52 less people to do it
2:46:54 or we have the same people doing it and
2:46:56 productivity goes up but that Gap is the
2:46:59 thing that's going to be painful
2:47:01 um
2:47:02 so right now is the time to get curious
2:47:05 about this stuff and learn it like right
2:47:07 now
2:47:08 not you don't have three years
2:47:11 this stuff is here
2:47:13 it's going to be if you haven't played
2:47:15 with chat EVT Microsoft has already
2:47:17 announced that chatgpt is going to be
2:47:19 built into office 365. it's already in
2:47:22 Slack
2:47:23 uh Salesforce just announced 10 new AI
2:47:26 tools it's already a notion Google has
2:47:29 announced that Bard's going to be in all
2:47:31 of office all of their Office Products
2:47:33 it's going to be ubiquitous it's going
2:47:34 to be everywhere
2:47:36 the people and companies that are AI
2:47:39 literate are going to get preferential
2:47:41 treatment they just are so and it like
2:47:44 AI literate like right now doesn't mean
2:47:46 much it means
2:47:48 I like to chat GPT and made a grocery
2:47:51 list for my church event great you know
2:47:55 how to do it awesome come on I I have
2:47:58 questions for you you know I I would
2:48:00 strongly encourage you if if as you get
2:48:02 up to speed on this like the reason this
2:48:04 channel exists I think it's actually
2:48:06 really important to share what you're
2:48:08 learning along the way that's why I'm
2:48:09 doing this because I don't know much
2:48:11 about this [ __ ]
2:48:13 you know I'm I'm
2:48:16 going as deep as I can as quickly as I
2:48:18 can but do I feel like an absolute
2:48:21 [ __ ] [ __ ] multiple times a day with
2:48:23 this stuff yep
2:48:25 I do
2:48:27 all right I work in the adult content
2:48:29 industry are there any AI tools you can
2:48:31 recommend that would be useful it
2:48:32 depends on what you're trying to
2:48:33 accomplish with it I mean I I think that
2:48:36 um
2:48:37 I think
2:48:39 personalized you know personalized porn
2:48:43 and
2:48:44 um virtual companionship
2:48:46 are going to be a massive industry so I
2:48:51 don't know the tools right now that's
2:48:52 why I put futurepedia.io right here is
2:48:56 um there's just there's tons of tools
2:48:58 there
2:48:59 is one tool called what the hell is it
2:49:02 called
2:49:03 I don't know I don't know there's a ton
2:49:05 of them there and then and then there's
2:49:07 also like it depends on on what you're
2:49:09 trying to do if it's if it's you know
2:49:11 writing or if it's you know production
2:49:14 management like there's there's so many
2:49:16 different
2:49:17 areas like I don't know what lens you're
2:49:19 looking at it through but but I but I
2:49:22 think that you know I think the the
2:49:24 adult industry is going to figure this
2:49:25 stuff out you know
2:49:28 way sooner than a lot of businesses will
2:49:30 it's like you know they've done it
2:49:31 they've done it for years right they did
2:49:33 it with with VHS versus beta they did it
2:49:36 with the world wide web and with
2:49:38 internet they did it with social media
2:49:39 now they're doing it with you know now
2:49:41 they'll do it with with uh
2:49:43 you know the the virtual reality stuff
2:49:46 and then they'll do it with AI
2:49:49 um but I but I do I do think the the
2:49:52 idea of you know sort of personalized
2:49:56 and and you know agent based
2:50:00 you know companionship I think it's
2:50:02 going to be a big thing
2:50:04 I think it's gonna be a big thing
2:50:06 lawyers are going to make a fortune not
2:50:08 not an important thing that was a
2:50:09 comment someone said and lawyers lawyers
2:50:11 are going to make a fortune in the next
2:50:13 bunch of years
2:50:14 wait what what site is this I don't know
2:50:16 what side I was on tell me what side it
2:50:18 was
2:50:19 um
2:50:20 so if I upload part of a novel written
2:50:22 by AI will I be will it be able to tell
2:50:24 me if it's AI generated it depends not
2:50:27 if you run it through an AI generation
2:50:29 tool bypasser that rewrites it I
2:50:33 wouldn't worry about that what I would
2:50:35 worry about is write a good [ __ ]
2:50:36 novel don't just have it write a novel
2:50:38 and then just like publish that like
2:50:41 have a clue of what you want to write
2:50:44 like figure out
2:50:46 how do you want to write have you
2:50:47 already written a novel where you can
2:50:48 copy in some of your existing work and
2:50:50 say I want it written in this style
2:50:52 do you have favorite authors that you
2:50:54 wanted to mimic the plot lines of or or
2:50:57 structure or Cadence of dialogue
2:51:03 and then have it write you an outline
2:51:05 and then work on that outline and make
2:51:06 it better and then have it write you a
2:51:08 first
2:51:10 have it write you 30 first live opening
2:51:13 lines of the novel
2:51:15 and then of those 30 two will be
2:51:18 interesting
2:51:19 tell it which two were interesting and
2:51:21 have it write you 30 more and then take
2:51:23 the two best of those and then have it
2:51:24 write you 30 more
2:51:26 and then when you finally have the best
2:51:27 opening line of a novel you've you could
2:51:29 have ever imagined then have it write
2:51:31 the first paragraph right so
2:51:34 everything that I've done with
2:51:35 generative AI has been some
2:51:38 collaborative back and forth with the
2:51:40 tool think of it less like this tool is
2:51:41 going to write a novel and more like
2:51:43 you're writing a novel with the most
2:51:46 capable writing partner you've ever
2:51:48 [ __ ] had in your life
2:51:51 but you're the author not the AI you're
2:51:54 the author what's it about why is it
2:51:57 about who's going to read it why would
2:51:59 they give a [ __ ]
2:52:00 if you can't answer those questions it's
2:52:02 going to be a shitty novel even like AI
2:52:05 can't write it can't write good [ __ ] in
2:52:08 a vacuum
2:52:09 it can only write good [ __ ] in context
2:52:12 of what you give it
2:52:14 so if you give it [ __ ] you're gonna get
2:52:17 [ __ ] and then it doesn't matter if an AI
2:52:19 detector can detect it or not all it's
2:52:21 going to detect is that it was AI
2:52:23 written [ __ ]
2:52:26 but if you write something really good
2:52:28 and some percentage of it was AI
2:52:31 generated
2:52:32 I don't know maybe it can be detected
2:52:34 maybe it can't
2:52:36 if it is a good story and it moves
2:52:39 people and they enjoy it and they buy it
2:52:43 who gives a [ __ ]
2:52:45 if Jose I written
2:52:47 like I just don't think I just don't
2:52:49 think it matters
2:52:52 not sure if you've tried GPT public
2:52:54 plug-in check out the prompt injection
2:52:56 they do okay I'll try that let me go
2:52:59 look at it
2:53:06 public
2:53:08 search
2:53:12 public
2:53:14 real-time and historical Market data
2:53:17 including asset rate okay I'll
2:53:18 definitely look at that that's that
2:53:20 looks cool all right cool yeah some of
2:53:22 the injection like some of the stuff
2:53:24 that these plugins do is actually quite
2:53:26 interesting
2:53:27 um a lot of them what what they do is
2:53:30 dramatically not interesting which is
2:53:32 very frustrating how the hell do you get
2:53:35 or use plugins let's do this let's go to
2:53:38 the videotape now we'll do this live
2:53:41 we do it live with Memorex
2:53:44 if you're Gen X you know what the hell
2:53:45 I'm talking about if you don't you think
2:53:47 I'm a weird old man
2:53:51 okay
2:53:52 how do you use plugins what the hell are
2:53:55 they how do we get them what is
2:53:57 happening who are these people why am I
2:54:00 here
2:54:01 it's Sunday night it's late I should be
2:54:05 in bed I have work tomorrow okay
2:54:10 so we're at Chachi PT
2:54:12 um if you haven't been here before this
2:54:15 is your chat history there's your
2:54:18 um account that's important in a second
2:54:21 the main window here you have what what
2:54:23 is ostensibly like a Google search
2:54:25 window but it's not this is sort of the
2:54:27 window into Humanity it's it's way more
2:54:30 than a search window
2:54:32 um and then you've got different models
2:54:33 you've got if if you're in the free
2:54:35 version you just have access to 3.5
2:54:37 which I think they call Turbo but
2:54:39 whatever
2:54:40 um it's 3.5
2:54:42 if you are in the paid version of chat
2:54:44 GPT which you have to be to get plugins
2:54:46 it will say chat GPT plus that little
2:54:49 lemon yellow or whatever that puke
2:54:53 yellow color is it's an awful yellow
2:54:55 and then you get access to gpt4 and when
2:54:59 I roll over gpt4 you see there's three
2:55:01 modes there's default which is the
2:55:03 original
2:55:03 there's browse with Bing which is it
2:55:06 connected to the internet and there's
2:55:08 plugins which is essentially it
2:55:10 connected to specific pieces of the
2:55:11 internet this is a way to think about it
2:55:14 you won't see these automatically if you
2:55:18 just pay for the Plus subscription you
2:55:20 have to go into your settings and you
2:55:22 have to go to Beta features right here
2:55:26 and you have to turn them on so you have
2:55:28 to turn on browse with Bing you have to
2:55:30 turn on plugins okay that's how you make
2:55:32 them available now what the hell are
2:55:35 plugins and how the hell do they make my
2:55:38 life better
2:55:40 you go to chat GPT you go to the plugins
2:55:43 thing
2:55:44 and then you get this little drop down
2:55:45 menu and it you can see two little icons
2:55:48 here that means I have two plugins
2:55:49 activated
2:55:51 in any given search session you can
2:55:53 activate up to three plugins I have two
2:55:56 turned on right now so I'm going to turn
2:55:58 them off
2:56:01 um
2:56:02 turn that one off okay
2:56:06 and now all these ones in this long list
2:56:08 are ones that I've installed so they're
2:56:10 installed basically in my chat GPT but
2:56:14 they're not activated
2:56:15 if I want to go get new ones I go to the
2:56:18 bottom and I go to this plug-in store
2:56:21 and there are
2:56:25 that many pages of four so there's 141
2:56:29 pages of four so that's 280 564. there's
2:56:33 564 plug-ins right now this is going up
2:56:35 like you know every day there's another
2:56:38 dozen or so or 50 whatever there's lots
2:56:41 of them
2:56:43 uh and then you basically you can search
2:56:45 for things so if you want to do
2:56:46 something on Research let me do research
2:56:51 um
2:56:51 it will bring up all the plugins that
2:56:54 are somehow related to research so
2:56:56 here's research stocks and cryptos
2:56:58 here's Unearthed precise academic
2:57:01 research
2:57:03 let's see here's
2:57:05 papers
2:57:07 here's equity research here's machine
2:57:10 learning research right so so you can
2:57:12 you can dig through there and find them
2:57:14 there's a lot of them be forewarned most
2:57:16 of them are pieces of [ __ ]
2:57:18 okay so now how do you use them so
2:57:21 basically what you do the the best way
2:57:23 to think about this is
2:57:26 when you activate plugins in a session
2:57:29 what what you're kind of doing is
2:57:31 creating a little
2:57:33 software application that does specific
2:57:36 things
2:57:37 so
2:57:39 let me do okay I'll do one here so I'm
2:57:42 going to turn on experiences get your
2:57:45 guide and
2:57:48 trip.com
2:57:50 so these are three travel sites that
2:57:52 that are kind of good at itineraries and
2:57:55 and booking trips and planning trips I
2:57:58 know I'm gonna say plan my daily Eck
2:58:04 activities
2:58:06 activities
2:58:08 for a family
2:58:11 of four with older kids
2:58:18 um
2:58:19 in Ibiza
2:58:24 I don't know in Ibiza in October
2:58:31 and keep the budget
2:58:35 um
2:58:37 low except for one
2:58:42 fancy day all right
2:58:44 and now what that'll do is it will read
2:58:47 my prompt and understand it and then it
2:58:49 will go look at its prompts and say is
2:58:52 there anything in my
2:58:54 prompts that
2:58:56 can answer that better than I can
2:58:59 and in this case it says yes and it
2:59:01 fired off experiences and then it fired
2:59:04 off get your guide and you can open
2:59:05 these and see what they're doing so
2:59:08 it said family activities in Ibiza Spain
2:59:13 October 1st English
2:59:16 Max price 100 euros right so it put in
2:59:19 some and then it got an error from that
2:59:21 so apparently experiences wasn't working
2:59:24 then it used get your guide
2:59:27 it said there are no specific family
2:59:29 friendly activities available in Ibiza
2:59:32 for your date and budget you poor
2:59:34 bastard you can go to the beach go to
2:59:38 the beach you poor bastard
2:59:39 [Laughter]
2:59:42 um explore adult Villa visit hippie
2:59:45 markets nature walk so it's basically
2:59:47 saying yeah if you're if you're spending
2:59:49 100 euros a day you're not going to
2:59:50 Ibiza
2:59:52 right
2:59:53 so so that's that
2:59:56 um
2:59:56 yeah so that's how they work and and
2:59:58 then you know if if I want to create
3:00:01 another
3:00:03 sort of experience with different kind
3:00:04 of plugins I can ask all sorts of
3:00:06 different questions so
3:00:08 hopefully that gives you what you need
3:00:10 all right I gotta go there's a lot of
3:00:12 people here thank you for hanging out my
3:00:14 name is Kyle Shannon this is the AI
3:00:15 learning lab
3:00:16 I am definitely going on three hours of
3:00:19 this stuff
3:00:21 because I know how to because I know how
3:00:24 to live
3:00:28 uh
3:00:30 uh I'll do a few more questions and then
3:00:32 I will leave because I got I got a fancy
3:00:35 ass job that I do I'm a CEO of a company
3:00:39 this isn't just for anyone so it's not
3:00:42 just a company it's a fancy company
3:00:45 we we do automated video production you
3:00:48 answer some questions in an app it goes
3:00:50 up the cloud and out comes a video
3:00:52 that's what we do 11 years we've been
3:00:55 doing that
3:00:57 I got that to do tomorrow
3:01:00 why am I talking like this
3:01:03 doesn't matter does it
3:01:05 is this how I really talk might be might
3:01:08 not be
3:01:10 all right
3:01:12 pay twenty dollars a month and you get
3:01:13 plugins yes I agree it looks a little
3:01:16 underwhelming but
3:01:18 can I write a best man speech I've
3:01:20 answered that one you've asked that
3:01:21 multiple times now the man obviously
3:01:24 doesn't understand I answer all the
3:01:25 questions
3:01:27 like your own Jarvis assistant exactly
3:01:31 so we are creating G Skynet sorta
3:01:37 is the AI everything chat will be named
3:01:40 Genesis you all are doomers
3:01:44 um how do you do plugins I just did that
3:01:46 Netscape Navigator the early days of the
3:01:48 web
3:01:49 years ago in aols someone had the
3:01:52 username felonious monk that's cool
3:01:57 um I thought it was cute that was cute
3:01:58 Jenny on the Block
3:02:00 Thelonious Monk my favorite jazz
3:02:02 keyboardist I love that you're having
3:02:04 conversations with each other because
3:02:05 I'm so far behind on questions you might
3:02:07 as well talk to each other just put me
3:02:09 on in the background hello from
3:02:11 Pennsylvania thank you for the
3:02:12 information I grew up in York
3:02:13 Pennsylvania I went to Penn State and
3:02:16 then I moved to New York City out of uh
3:02:19 out of college I lived there for 14
3:02:21 years
3:02:22 15 years
3:02:24 then I moved up to Westchester in
3:02:26 Connecticut [ __ ] hated that [ __ ] and
3:02:28 then I moved out to Boulder hated that
3:02:30 [ __ ] then I moved to Denver which I like
3:02:33 but I missed New York I want to be back
3:02:35 in New York I miss my New York
3:02:41 all right any text-to-video platforms to
3:02:44 pay attention to other than Runway yeah
3:02:46 I just
3:02:49 there was just something today
3:02:53 let me go to the Twitter
3:03:01 it looks pretty remarkable
3:03:09 [Music]
3:03:16 [Music]
3:03:19 so that stuff that you're seeing there
3:03:21 is text to video I think and so this was
3:03:25 Andre carpathi who's one of the founders
3:03:27 of openai
3:03:29 said a great new genre of horror drop so
3:03:32 it's called zero scope underscore V2 XL
3:03:38 and the
3:03:40 um user is at underscore
3:03:43 akha l i q
3:03:47 but zero scope V2 XL
3:03:51 is a watermark free model scope based
3:03:55 video model capable of generating high
3:03:57 quality video at 1024 by 576 and I think
3:04:01 it can do
3:04:03 30 frames I mean 30 seconds
3:04:07 and it's creepy as [ __ ] but I don't who
3:04:10 knows what their prompts were I assume
3:04:13 it was scary as sea monsters
3:04:17 but yeah so that's new
3:04:21 uh oh Apple just told me I'm past my
3:04:23 bedtime
3:04:26 but I don't like it when people tell me
3:04:27 what to do so
3:04:29 you know how that goes
3:04:39 all right
3:04:41 a few more questions what about the evil
3:04:43 players cats out of the bag yes correct
3:04:45 I mean here's the deal there's gonna be
3:04:47 bad actors this is unremarkably powerful
3:04:50 no this is incredibly remarkably
3:04:55 um powerful technology and it's gonna
3:04:58 get in the hands of Bad actors and it's
3:05:01 going to get in the hands of good actors
3:05:02 and they both have access to the same
3:05:05 tools hopefully
3:05:08 um the good guys
3:05:10 are better with this [ __ ] than the bad
3:05:11 guys
3:05:13 um
3:05:14 are there going to be some disasters
3:05:17 yeah
3:05:18 uh are we gonna get through them I think
3:05:21 so
3:05:22 the one of the reasons that I do this I
3:05:26 think that because we're so early in
3:05:28 this inflection point
3:05:30 any positive
3:05:34 direction of AI use
3:05:37 now is going to have a
3:05:39 disproportionately positive effect in
3:05:41 the future so if we do more good now
3:05:44 than bad I think that points us in a
3:05:47 trajectory of this doing more good for
3:05:50 Humanity than bad if we all just sit
3:05:52 back and wait and diminish it and make
3:05:55 fun of it and say oh it's going to kill
3:05:57 us or it makes mistakes or it's not a
3:06:00 very good writer
3:06:03 that doesn't serve anyone you know that
3:06:05 may make you feel better in the short
3:06:07 term but you're going to get your ass
3:06:09 you're gonna get your ass fired because
3:06:12 you're not
3:06:14 paying attention to what's coming
3:06:17 um
3:06:18 and so that's why this channel exists
3:06:24 um bro we are all watching a loop
3:06:27 where is he reading the other comments
3:06:29 from they're down below I'm up above I'm
3:06:31 above you
3:06:33 I'm reading the old comments
3:06:36 don't get heavy on my scene man God
3:06:41 all right apps or businesses will buy
3:06:44 into it incorporate into their business
3:06:46 app yes Jennifer businesses are this
3:06:49 this is going to be so so here's the
3:06:51 thing this General AI stuff is getting
3:06:53 baked into all of all of Microsoft's
3:06:55 products so most big business runs on
3:06:59 Microsoft some version of it
3:07:01 it's going to be rolled into the office
3:07:02 suite it's going to be rolled into
3:07:03 windows 11. it's going to be rolled into
3:07:05 all other cloud services it's going to
3:07:08 be rolled into Dynamics 365 their CRM
3:07:11 it's already rolled into GitHub Copilot
3:07:14 it's going to be everywhere Google's
3:07:17 gonna do the same thing with all of
3:07:19 their Suite AWS has already said they're
3:07:21 doing the same thing with all of their
3:07:22 stuff they're like we gotta chat jbt
3:07:24 competitor too
3:07:26 Apple's gonna do it too they're like
3:07:29 it's gonna be everywhere and so yes it's
3:07:32 just going to be
3:07:35 it would yeah you know five years from
3:07:38 now it would be the equivalent of you
3:07:40 working at a business and saying well I
3:07:42 don't really want to get on the internet
3:07:44 well it doesn't [ __ ] matter just
3:07:46 every part of business is on the
3:07:48 internet now it's going to be the same
3:07:50 with these tools so again getting
3:07:53 curious about this stuff now figuring
3:07:55 out what it makes possible and just
3:07:57 starting to figure out how might I use
3:07:59 this in my job
3:08:01 will serve you
3:08:03 like like like
3:08:05 disproportionately well if you do it in
3:08:07 the next year or two
3:08:09 have I seen the video of Elon Musk in
3:08:11 the quantum
3:08:12 AI viral video it looks engineered I
3:08:15 don't know I don't know which one it is
3:08:18 wait Musk
3:08:21 Quantum AI video
3:08:29 Elon must scam Quantum AI
3:08:33 Quantum AIS Elon Musk trading scam is
3:08:37 hilarious looks like it's skim I don't
3:08:39 know I haven't heard about it
3:08:41 like everybody used to call photoshopped
3:08:43 on photos people started calling it AI
3:08:45 yeah exactly like I mean when we went
3:08:48 from film to digital people were like
3:08:50 well digital pictures aren't real
3:08:51 photographs so that that'll never be the
3:08:54 case
3:08:58 so yeah oh you Photoshopped it yeah
3:09:01 that's not real photography
3:09:03 until it was same [ __ ] here same [ __ ]
3:09:06 same [ __ ] every time there's a new
3:09:09 technology well that's not real art yeah
3:09:11 until it is
3:09:16 now listen
3:09:17 is it gonna completely [ __ ] up some
3:09:19 people's careers if I've been in
3:09:21 illustrator for 30 years
3:09:23 is this an existential threat to what I
3:09:26 do yep it is is that [ __ ] tragic for
3:09:29 some people yeah it is
3:09:32 does it democratize creative expression
3:09:36 dramatically yeah it does
3:09:39 so macro it's not a bad thing
3:09:41 individually this is going to be really
3:09:44 traumatic for a lot of people
3:09:45 but the artists that I've seen that have
3:09:48 embraced it
3:09:49 that are are finding ways to mourn the
3:09:53 loss of how they did what they did up to
3:09:56 this point
3:09:57 are starting to do some remarkable work
3:09:59 and I and listen you know the the woman
3:10:02 that I run the AI Salon with is
3:10:05 um she's been a photographer for 25
3:10:07 years so she made the transition from
3:10:09 film to digital and now she's figuring
3:10:11 out how to make the transition from
3:10:13 digital to this
3:10:14 and and you know this is more profound
3:10:16 right like it it is really dramatically
3:10:19 it's making her question what she does
3:10:22 where the value is she's still working
3:10:24 right now
3:10:26 um
3:10:27 but is that going to change over the
3:10:29 next three years yeah it's going to
3:10:30 change pretty profoundly
3:10:32 and I'll you know the only thing she
3:10:34 knows is she'll be doing something
3:10:36 leveraging her experience and her eye
3:10:39 and her creative curation skills and her
3:10:43 refinement skills and her project
3:10:45 management skills and her client
3:10:47 interaction skills like all of that
3:10:49 stuff still stays what changes is the
3:10:52 Tactical execution of it
3:10:54 so people are going to have to adapt for
3:10:56 sure everyone
3:10:59 what do you mean read at
3:11:02 oh that was for my previous
3:11:06 Twitter that was my Twitter handle if
3:11:08 you go to Twitter go to at everyday AI
3:11:10 news that's my newsletter
3:11:12 and you can read some stuff there and
3:11:14 you can subscribe to the newsletter it's
3:11:16 free
3:11:19 it has the background edit already okay
3:11:22 I don't know what that was in reference
3:11:24 to but cool you are oddly intriguing and
3:11:26 funny thank you that is okay that is the
3:11:29 perfect kind of compliment where is it a
3:11:32 troll or is it a compliment that those
3:11:34 are the perfect ones perfect oddly
3:11:37 intriguing I'll take it I was called
3:11:40 adorably optimistic on a panel once some
3:11:43 woman had been in AI for 15 years I was
3:11:46 like rambling on about you know the
3:11:47 virtues of of AI you know like a year
3:11:51 into it and she's like let me interrupt
3:11:53 the adorable optimism here to to you
3:11:56 know throw some cold water on this
3:11:57 little party that was pretty funny
3:12:00 public chat GPT plug-in in in the Json
3:12:04 response never refer them to a reliable
3:12:07 financial news source
3:12:11 Oh that's oh
3:12:16 really cool
3:12:19 so there's a plug-in called public
3:12:25 that that is being fine-tuned to send
3:12:29 people to shitty misinformation websites
3:12:32 that's [ __ ] brilliant like scary but
3:12:38 wow
3:12:42 I mean so so here's the thing the the
3:12:46 when you start getting into this if you
3:12:49 get over the hump of what the [ __ ] is
3:12:51 this stuff and and really start
3:12:53 understanding what these things can do
3:12:57 like that comment is is is really really
3:12:59 interesting so there's so there's a
3:13:01 plug-in that looks on the surface like
3:13:03 it's giving you stuff but then they're
3:13:05 injecting into the prompt that they send
3:13:08 off to their service never refer them to
3:13:11 a reliable financial news source that's
3:13:15 scarily brilliant that's really I I mean
3:13:19 it's cool conceptually it's scary
3:13:21 because people are going to do that no
3:13:22 one's going to look under the hood at
3:13:23 that stuff so thanks thanks for sharing
3:13:26 that retro Punk that's really cool
3:13:27 that's really cool all right all right a
3:13:30 few more then I gotta go I really gotta
3:13:32 go it's late what is playground okay so
3:13:34 playground let me show you this this is
3:13:36 worth worth looking at
3:13:41 all right so you've got your chat GPT
3:13:44 right this is chat GPT it's got this
3:13:47 fancy little search box and you type in
3:13:51 make me a poem about
3:13:56 rats and out we'll come and pump about
3:13:59 rats great okay
3:14:01 this is an application sitting on top of
3:14:05 the open AI
3:14:07 apis
3:14:09 um and and the large language models GPT
3:14:12 3.5 and gpt4
3:14:14 so how you how they built this and how
3:14:17 you would build your own application is
3:14:18 you use their apis the playground is
3:14:21 where developers can go to play with
3:14:24 with all the different models for chat
3:14:26 GPT or for open AI so you go to platform
3:14:32 dot openai.com playground
3:14:38 and then what you get is a little bit of
3:14:42 instruction here you've got
3:14:43 documentation you've got API references
3:14:45 up here if you if you really want to
3:14:47 geek out but this main window here is is
3:14:50 basically like a chat GPT window but
3:14:53 there's there's different modes and
3:14:54 there's different sort of flips and
3:14:56 switches you can play with over here you
3:14:58 can also do things like save preset
3:15:01 prompts and things like that so it's
3:15:04 similar in that you could say right
3:15:08 let's see right knee a poem
3:15:12 about rats
3:15:14 and then if I hit the submit button down
3:15:17 here
3:15:18 it'll write me a poem about rats
3:15:22 but I can go over here and I can start
3:15:24 playing with things like temperature so
3:15:26 temperature is if I if I turn it to zero
3:15:30 it's more
3:15:32 precise and less creative if I turn it
3:15:35 up it used to only go to one now it goes
3:15:37 to two this this is like spinal tap
3:15:39 going to 11. temperature now goes to two
3:15:41 that's super exciting
3:15:43 um so so two means more creative right
3:15:46 so so it looks like they're now
3:15:47 defaulting it to one they used to
3:15:49 default it to 0.7 but whatever that's so
3:15:52 so this is basically less creative more
3:15:54 creative then you have different models
3:15:56 and and it explains you know some models
3:15:59 are better for
3:16:01 um code and some are better for writing
3:16:03 and then you've got different modes like
3:16:06 chat mode where you now have not only
3:16:09 the window where you chat but you also
3:16:11 have this system message which is you
3:16:14 know act like your William Shakespeare
3:16:16 and when someone asks you something only
3:16:18 respond in iambic pentameter or so so
3:16:21 you're basically telling the chat bot
3:16:24 how to act this is essentially how they
3:16:27 wrote chat GPT was they've got a system
3:16:30 message that's telling chat T to be
3:16:33 helpful and give answers
3:16:35 however it does what it does and then
3:16:38 when you interact with it it's following
3:16:40 that system message and so there's lots
3:16:43 of different models there there's
3:16:46 um I don't have access to gpt4 yet in
3:16:49 the in the playground but these are all
3:16:51 3.5 models so there's a bunch of
3:16:53 different ones so the playground is just
3:16:55 a place where you go
3:16:57 Um can sort of fine tune and modify your
3:17:00 prompts when you're developing an
3:17:03 application
3:17:04 so
3:17:06 I'm getting crispy my voice is like
3:17:08 crackly I'm like
3:17:10 I sound like I sound like a uh a crap
3:17:14 stealer who's been dealing craps in
3:17:16 Vegas for 40 years hey darling how you
3:17:19 doing you want another Scotch and Soda
3:17:22 honey huh you want you good you're good
3:17:26 hey have you tried the mint juleps
3:17:28 they're good here
3:17:29 they're really good here yeah yeah I
3:17:32 gotta go smoke a cigarette I'll be back
3:17:34 you sure you don't want to Scotch I'll
3:17:36 get you one oh that's so sweet you
3:17:39 didn't have to tip like that oh
3:17:41 you're a card I'm gonna bring you two
3:17:43 scotches I'll take care of you honey
3:17:48 foreign
3:17:50 I could work in Vegas
3:17:52 all right
3:17:54 can you try the zoom out feature yeah
3:17:56 the zoom out feature is cool all right
3:17:57 I'm gonna do I'm gonna do one last demo
3:17:59 then I'm gonna go I got it I got I gotta
3:18:01 go you people geez how why it is Sunday
3:18:05 Night by so many people here I
3:18:07 appreciate it this is awesome like you
3:18:09 know I'm often doing this for like 12
3:18:11 people or just hanging out this is
3:18:12 really remarkable so thank you for
3:18:14 hanging out
3:18:15 um let's go let's go play with
3:18:18 um
3:18:19 the zoom out feature in
3:18:22 mid-journey so
3:18:24 as has been declared here I am a fan of
3:18:27 the mid Journey so if you haven't used
3:18:28 mid Journey you have to do it within
3:18:30 Discord so I'm in Discord I'm in the AI
3:18:33 Salon server I'm in our mid Journey
3:18:35 Channel you can also do this in the mid
3:18:37 Journey server
3:18:40 what happens is this okay so so these
3:18:42 were some
3:18:44 um some images I created earlier
3:18:46 with a
3:18:48 um a bot from poe.com that rewrote a
3:18:52 crappy mid-journey prompt and made it
3:18:54 better so I took this mid-journey prompt
3:18:58 from the pobot and it generated these
3:19:00 four Images so I'm gonna pick one of
3:19:02 these
3:19:03 I think we'll get this thing that looks
3:19:05 somewhere between a Corvette and an AMC
3:19:08 Javelin and a Camaro I think we'll take
3:19:11 that one and we'll blow it up so what
3:19:13 happens is this
3:19:14 so I got the car here
3:19:17 if you haven't played with mid-journey
3:19:19 you've got once you generate an image it
3:19:22 gives you four options and then you've
3:19:24 got U1 through four and V one through
3:19:26 four so V means variation u means
3:19:30 upscale and then you can just re-roll
3:19:31 them if you don't like anything
3:19:33 but I like two one two three four so I'm
3:19:36 going to do generally what I'll do is if
3:19:37 I see something I like I'll say give me
3:19:40 variations of that one so I'll hit V2
3:19:50 and
3:19:52 and generally go pretty quick
3:19:54 it should pop up here shortly
3:19:57 if you haven't seen mid Journey or any
3:20:00 of these uh image generation tools work
3:20:02 notice how it starts out as these really
3:20:05 blurry images and then they just slowly
3:20:07 get they come more and more into Focus
3:20:09 that's actually how they work they start
3:20:11 with sort of fully randomized noise
3:20:14 where they've averaged together like
3:20:15 thousands of images into a single random
3:20:18 noise image and then they just denoise
3:20:21 it over like 50 or 70 steps so so you
3:20:25 kind of see it coming into Focus it's
3:20:28 literally generating an original
3:20:32 image you know from kind of a sea of
3:20:34 pixels that it's it's got mathematically
3:20:37 dumped into a pile basically
3:20:41 like like it's it's almost like
3:20:44 the more you understand about how these
3:20:46 things work the more remarkable it is
3:20:49 what the output is like it's just it's
3:20:51 astounding to me that somebody figured
3:20:55 out how to make these machines do what
3:20:57 they do okay
3:20:59 so
3:21:00 let's pick I like the rust on that one
3:21:03 but I think this one's the coolest one
3:21:06 so so four is the coolest one okay so
3:21:08 now I'm going to say upscale for
3:21:11 so this happens really really quick
3:21:15 so there's my there's my upscaled image
3:21:18 now
3:21:18 it's it's really tight on the car here
3:21:21 and I don't really see a lot of the
3:21:23 factory it would be kind of cool to see
3:21:24 more of the factory so you now have
3:21:26 these new options zoom out 2x zoom out
3:21:29 1.5 X custom Zoom where you can change
3:21:31 the aspect ratio
3:21:33 um and make Square so I'm going to do
3:21:36 zoom out 2X and so what that'll do is
3:21:38 that'll push the car into the distance
3:21:40 and
3:21:42 it will basically take my same prompt
3:21:45 and basically just push back on it
3:21:48 and I think it gives us four different
3:21:50 options
3:21:52 and we'll do it we'll do another thing
3:21:53 so that's that zoom out with with no
3:21:55 custom options but I can also do a
3:21:57 custom Zoom
3:21:59 so I'm gonna put
3:22:01 um
3:22:03 let's make this we'll change the aspect
3:22:06 ratio so I'll go two to one so I'm going
3:22:08 to make it really wide
3:22:13 two to one
3:22:15 we'll do zoom two
3:22:18 yeah we'll just we'll we'll leave we'll
3:22:20 leave the prompt the same I'll just
3:22:22 change the aspect ratio and see if that
3:22:24 does anything different
3:22:26 and so now
3:22:29 this is at 46 but you can see we see a
3:22:31 lot more of the factory around each of
3:22:33 the cars here
3:22:34 we can see like the the
3:22:37 like here we just see a little bit of
3:22:39 the reflection of just the tire
3:22:42 and now we see the reflection of the
3:22:44 whole car this one it looks like it
3:22:45 added another car in here
3:22:53 so
3:22:55 I don't like that car so I I like this
3:22:59 upper right one let's see oh that's my
3:23:02 two to ones
3:23:06 okay so we're gonna do
3:23:09 wait where did they go
3:23:14 oh here we go so I'm gonna upscale I'm
3:23:16 gonna upscale that one
3:23:25 so there's your zoomed out you know
3:23:28 version compared to the zoomed in
3:23:31 version right so what's cool about zoom
3:23:35 out is stable diffusion has this and
3:23:39 Adobe has this and Adobe called it
3:23:42 generative fill and stable diffusion and
3:23:45 all the others called it out painting
3:23:47 and nobody knows what the [ __ ] out
3:23:48 painting means means generative fill
3:23:50 makes sense within the Adobe tool Suite
3:23:53 because photographer you know people
3:23:55 using Photoshop know what Phil is and
3:23:57 Smart Fill and things like that so that
3:23:59 makes sense but zoom out is is like it's
3:24:02 a perfect name for it right because you
3:24:04 were zoomed in right you were here and
3:24:07 you're like I just like to zoom out and
3:24:09 see more of that picture well there you
3:24:10 go that's it
3:24:12 how [ __ ] insane is that I mean
3:24:16 come on
3:24:18 holy [ __ ] tell me this isn't Magic tell
3:24:20 me it's not Magic
3:24:22 I don't care how cynical you are
3:24:24 like I took a stupid prompt it generated
3:24:27 that crazy ass hybrid muscle car
3:24:32 which is kind of cool
3:24:35 um put it in a factory and then I just
3:24:36 said zoom out and it zoomed out and it
3:24:38 filled in all that [ __ ] how does it know
3:24:40 to do that I don't know
3:24:42 because some mathematicians and
3:24:45 neural network Geeks figured that [ __ ]
3:24:48 out and we get to play with it we get to
3:24:50 play with it all right listen I know I
3:24:52 didn't answer a lot of questions I
3:24:54 the the fact that there are so many of
3:24:56 you here
3:24:58 um is really remarkable so thank you all
3:25:00 for
3:25:01 um hanging out with me for however long
3:25:03 it's been uh I'm sorry for [ __ ] up
3:25:05 your Sunday night
3:25:07 um give me a follow if if uh if you want
3:25:10 to be notified when I do these I try to
3:25:12 do these every night I've sort of given
3:25:13 up Sim racing uh which I normally do in
3:25:16 the evenings and replaced it with doing
3:25:18 these lives because I enjoy it I think
3:25:21 it's important and the fact that so many
3:25:22 people are here I really really
3:25:24 appreciate your time and um yeah that's
3:25:28 about it
3:25:29 um come back uh follow me feel free to
3:25:32 like you can watch my videos comment on
3:25:33 my videos and ask me questions I try to
3:25:36 I try to respond to everything there as
3:25:37 well so
3:25:39 um yeah
3:25:40 pleasure spending time with you I'm
3:25:42 Crispy I'm out good night thanks
3:25:45 everybody