aiLL 6/19/23 - The Generative AI Tsunami Is Coming

Description

Chapters

Transcript

0:01 foreign
0:17 let's try this again we will see if it's
0:21 gonna work I thought it was Comcast but
0:24 it turns out it was my son's downloading
0:27 [ __ ]
0:28 ah
0:31 Father's day was Sunday they get their
0:33 Revenge today you see how that works
0:39 hey everybody
0:41 um do me a favor as you join here tap
0:43 the screen the like screen that'll let
0:46 people know we're here or you can share
0:48 the live that would help also
0:50 uh this is the AI learning lab I'll get
0:53 chatting here in a second funny man
0:57 it's all comedy all the time
0:59 it was my shitty kids that were screwing
1:02 up the bandwidth here it was not Comcast
1:05 I prematurely blamed Xfinity for shitty
1:09 service and it was just shitty children
1:13 they did yell at me though for resetting
1:15 the router without warning them so
1:18 I think they've got a pretty good
1:19 argument that I was a shitty father in
1:21 that in that point uh hey hey Sonia how
1:23 are you good to see you
1:25 good to see you welcome everybody
1:27 welcome welcome welcome welcome
1:29 well we'll let this thing get wound up
1:31 here I appreciate the Taps on the screen
1:33 that lets people know we're here uh how
1:37 old are your kids I have 23 year old
1:39 twin boys
1:42 and they are lovely human beings
1:50 my name's Kyle Shannon it's the AI
1:53 learning lab if you are new to
1:55 generative AI
1:58 um go here that's chat.openingi.com just
2:01 point your browser there that's chat CPT
2:03 if you've not played with it yet go play
2:06 with it go start playing while I ramble
2:08 on if you've got questions about AI pop
2:11 them down below uh we have fun here we
2:14 we uh we we uh we dig in deep into the
2:17 philosophy of this AI stuff the future
2:20 of it the the glory of it the crap that
2:25 doesn't work of it which is a lot
2:27 we talk about it hallucinating I like to
2:31 call chat GPT mansplaining as a service
2:34 because it will confidently answer you
2:36 even if it's wrong
2:38 like a good man like a good man in your
2:41 life
2:45 oh yeah
2:47 um uh my background I'm an entrepreneur
2:50 um I've been I've been on the planet for
2:53 a while as evidenced by this
2:58 um but on a slightly more serious note
3:00 the uh I was there for the early days of
3:03 the World Wide Web I started one of the
3:04 early digital agencies in an online
3:06 magazine back in those days
3:08 and
3:10 um felt pretty lucky to be part of a
3:14 technological change that you know
3:17 transformed everything and I didn't
3:18 think I'd see another one of those in my
3:20 life
3:21 and here we are
3:23 um November 30th 2022 when chat GPT
3:26 launched from those wacky kids over at
3:29 open AI um that changed the game in my
3:32 opinion and I think this stuff is coming
3:35 way faster than any of us are prepared
3:38 for I think the changes are going to be
3:39 profound
3:41 um and I think I think they're coming so
3:44 with that feel free to pop any questions
3:47 down below I'll do what I can to answer
3:49 them I will probably fall behind fair
3:52 warning I will probably cuss if you're
3:55 not a fan of the cussing
3:57 I'd get out because when I get excited
3:59 the shitty words come out
4:01 do you think software engineer I assume
4:05 you meant to finish that sentence Oh
4:06 wait do you think software engineering
4:08 needs will decrease because of AI I
4:11 think they'll change dramatically
4:13 I think that we're still going to need
4:15 software Engineers there's going to be a
4:17 lot more
4:18 software written
4:21 um
4:22 but
4:23 it's probably going to look more like
4:26 managing AI systems
4:29 and being more like a product person
4:33 than actually doing the individual
4:34 coding I think I think some amount of
4:36 coding will get done but it'll probably
4:39 be more like you know connecting modules
4:41 together and things like that because
4:43 the the generative AI tools are going to
4:46 do a lot of the a lot of the base work
4:48 right so if if you're an engineer and
4:50 have haven't used GitHub co-pilot the
4:53 kind of pair programming tool within
4:55 within GitHub I'd strongly encourage you
4:58 to go do that so I do think software
5:00 Engineers are going to be
5:02 in demand but they're going to need to
5:04 be AI literate software engineers
5:08 because what's what's wild about these
5:11 generative AI tools is that
5:15 so for like how long have I been dealing
5:17 with software since probably
5:20 I don't know
5:21 late 70s early 80s in some form or
5:24 another
5:25 and the thing that I've always known
5:27 about software is you kind of think
5:29 about what you wanted to do and then you
5:31 start thinking about the logic and and
5:34 how do we what are the data elements
5:36 that we need to code and how do we code
5:39 them and what's the logic for how we
5:41 glue them all together
5:43 um generative AI just does all that like
5:46 all the logic thinking it's really good
5:48 you can just ask it to do [ __ ] it will
5:50 figure out the form
5:52 and it can look at like there's a
5:54 there's a thing coming from chat GPT
5:56 called code interpreter where you can
5:59 upload big piles of data like a giant
6:01 spreadsheet it will just learn the
6:04 spreadsheet and then
6:06 do full-on data analysis for you like
6:09 it's incredible at that [ __ ] so so I
6:11 think I think the engineering world is
6:14 is going to transform in pretty profound
6:17 ways
6:18 um one of the things Jason kalkanus is
6:21 what's it called the all-in Pod with the
6:23 like the four billionaire dudes talking
6:25 they've been talking a fair amount about
6:27 this that what they've been hearing is
6:29 that tech companies in San Francisco
6:32 that are going all in on AI they're
6:34 seeing 30 to 40 percent productivity
6:36 increases for companies that are going
6:38 all in on it so that's kind of profound
6:40 stuff when you see that significant of
6:42 productivity increase just because of
6:44 switching to something like generative
6:46 AI I'm having mid-journey issues I am
6:49 creating a series of images for a book
6:51 and it's not going well sorry about that
6:53 that's that's
6:55 that's tough like so the thing about
6:58 generative AI I would say the single
7:00 most frustrating thing about generative
7:02 AI is that um getting it to be
7:05 predictable is really really hard
7:08 um let me share a couple of things with
7:11 you that may help
7:14 all right I flipped my camera now I
7:16 gotta turn off my little thingy and then
7:18 we're gonna go over here we're gonna do
7:20 some playing okay so the first place I'm
7:22 gonna send you
7:25 well let's see yeah yeah yeah okay so go
7:29 to open art
7:33 wait open art
7:36 dot is it AI
7:39 yeah
7:40 go to openart.ai slash prompt book
7:45 openart.ai promptbook now this this is
7:49 for stable diffusion
7:51 but
7:53 what this but it's good it's good for
7:55 mid-journey as well because this is more
7:57 like a Concepts primer than than it is
8:00 you know a specific to to prompting so
8:04 if you flip through here it will talk
8:06 about
8:08 um prompt engineering and and what it is
8:10 and like what the components of a prompt
8:13 are and and you know really expanding on
8:16 the different elements
8:18 um it'll then give you some examples
8:19 then it goes into modifiers and
8:21 modifiers are things like
8:23 what are the different terms that these
8:27 image generation tools know like it
8:29 knows all the photography terms right
8:31 like tilt shift and long exposure and
8:34 color splash and satellite monochrome
8:38 um so so that's just photography
8:40 different kinds of lenses
8:42 um different kinds of lighting art
8:43 mediums this is probably where where you
8:46 know you could potentially have some
8:48 advantages if you're doing stuff for a
8:50 children's book maybe you want it to
8:52 look like a watercolor maybe you want it
8:54 to look like particular characters it
8:57 also knows artists right it also knows
9:00 Styles let me see I think there's Styles
9:02 coming up here
9:04 um like illustration styles
9:08 um you know Vector children's book right
9:11 things like that so so this
9:15 it's like a 96 page uh document if if
9:20 you are feeling a little lost this this
9:22 could help you so that's one thing you
9:24 can do another thing you can do is I'm
9:27 going to pop you over to poe.com so this
9:30 is one of the sites I recommend going to
9:32 play with so when you come to poe.com
9:34 it's basically just a normal
9:36 um like chat cpt-like experience
9:40 and you know you can you can type in
9:43 stuff and it will
9:45 generate content but there's also
9:49 um
9:50 Bots that people have created so there's
9:53 one called MJ Captain so poe.com MJ
9:57 Captain which stands for mid-journey
9:59 Captain
10:00 um there's another one called po.com
10:02 mid-journey so let me just I'm gonna go
10:04 to that one poe.com slash
10:09 mid Journey so I think this is just
10:12 something that someone put together
10:15 but what happens is I can do
10:18 children's
10:22 book illustration
10:26 of a happy elephant
10:30 um
10:31 walking through a field of flowers
10:39 or whatever so so when I type that in
10:41 what this bot's gonna do is actually
10:44 rewrite that with all sorts of
10:47 um
10:50 of uh
10:52 what it was like embellishments and
10:54 things like that now this just did
10:57 um this did it as a color photograph I'm
10:59 gonna say I don't want a photo I want
11:04 a watercolor
11:08 illustration
11:10 let's see if let's see if it does better
11:12 okay I apologize for misunderstanding
11:15 here's the here's the prompt watercolor
11:18 illustration of a happy elephant walking
11:20 through a field of flowers the elephant
11:22 is depicted with a big smile rosy cheeks
11:24 as if enjoying the scenery blah blah
11:26 blah then then
11:28 it's got it's weird it put directors in
11:31 here and it puts some
11:33 photographers in here so I'm gonna kill
11:36 the photographers but let's let's pop
11:38 over to
11:40 um mid-journey so if you haven't seen
11:42 mid-journey now you're in for a treat
11:44 okay so to create a mid-journey image
11:48 you actually have to go to this thing
11:49 called Discord if you don't know what
11:50 Discord is it's basically kind of like
11:52 slack but for geeks these things down
11:55 the left are different servers so think
11:57 different communities I'm now in the AI
12:00 Salon community so this is the the uh AI
12:03 Salon that I started and we have a
12:05 mid-journey channel so here's one of the
12:07 channels these uh these things here are
12:10 channels of different conversations
12:11 basically
12:13 and then to create an image here you
12:15 type slash imagine
12:17 and I'm just going to paste in that
12:20 prompt and I'm going to kill the
12:21 cinematographers and photographers
12:25 you could replace one of the things that
12:27 could help is you could replace the
12:30 artist that they give you if if you find
12:32 specific illustrators of children's
12:35 books that you like you can pop them in
12:37 there
12:38 but I'm just going to go ahead and hit
12:39 return and now so it takes that giant
12:42 prompt that the chat bot created
12:45 and it will it should start making us
12:48 some images so we'll see how these do
12:51 um so I don't know if that helped but
12:53 hopefully that'll at least point you in
12:55 in some directions we'll we'll sit here
12:57 and watch as that thing goes
12:59 I'll try to figure out some other
13:00 questions here
13:02 and
13:06 I'm not sure if that's looking like an
13:08 elephant maybe oh yeah it is okay
13:12 have you tried the prompt library for
13:14 help oh yeah that's the oh that's good
13:16 Sonia I'll go show that okay so wait so
13:19 while that's while that's launching so
13:20 what Sonia's talking about
13:23 so we're gonna go to the mid Journey
13:26 um
13:28 to the mid-journey server and then
13:30 within the mid-journey server there is a
13:33 prompt Library
13:37 let's see where is it uh
13:42 [Music]
13:44 is it in here
13:47 it was a stable diffusion
13:49 prompts image prompts
13:52 so
13:54 so there's image prompts multi-prompts
13:56 patterns textures environments
13:58 characters you can also Within These
14:01 communities you can also ask for help so
14:04 there's General conversations and show
14:06 and tells so you can ask people for help
14:11 uh
14:14 there's lots of places to do things see
14:16 there's these are voice conversations
14:18 going on so so lots you can do there
14:20 let's go back and see
14:22 how did that illustration come out
14:26 [Music]
14:27 well they're definitely children's book
14:30 like
14:34 like this this lower right one's pretty
14:37 good
14:38 that's not bad although the the little
14:41 the pointy trunk there's there's some
14:43 genetic issues going on in the drippy
14:45 ears
14:46 this elephant's got some issues
14:49 um that one looks that one looks pretty
14:51 cool so anyway that's you know Without
14:54 Really Trying you know I got this kind
14:55 of example now if the challenge that
14:58 you're having is getting consistent
15:00 characters across different renders then
15:03 you're gonna need to do some things like
15:06 um use the blend tool within mid-journey
15:08 uh or upload an image and then create
15:11 other images based on that image so that
15:14 might be a way that you need to go as
15:15 well
15:17 so
15:18 hopefully that helped
15:21 let me get my little background up here
15:23 looks like we have a few folks here
15:26 welcome everybody my name is Kyle
15:28 Shannon this is the AI learning lab if
15:30 you have questions about AI pop pop them
15:32 down below I think coding is going to be
15:34 needed more than ever I think that's
15:35 right I think how coding gets done is
15:38 going to change though
15:41 um so so again I I go back to I think we
15:44 need Engineers but we definitely need
15:45 Engineers who who are
15:48 um AI generative AI literate because how
15:51 you think about programming logic
15:55 really does change with generative AI
15:57 mid-journey is ignoring my prompts now
15:59 that stinks
16:01 assume the worst hope assume the worst
16:03 hope for the best prepare for everything
16:05 I think that's right in general I was
16:09 able to get 20 images then it just
16:11 stopped giving me back what I'm asking
16:13 for
16:14 hmm that's weird and you're in
16:17 you're in the official mid-journey
16:19 Discord doing that
16:23 um I would go ask questions and I I
16:25 would share your prompt with people
16:26 there might be something so so let me
16:29 show you one other thing
16:32 you could have inadvertently
16:35 um
16:37 hang on you could have inadvertently
16:44 um
16:45 change to setting so if you're in
16:48 mid-journey Go slash settings
16:55 and then you'll see all of these
16:58 um settings so you've got mid Journey
17:01 one two three four five and five point
17:04 one
17:05 um four might be better for
17:07 illustrations I think it's pretty good 5
17:09 and 5.1 seem to be optimized for
17:11 photography
17:12 I I would play around with different
17:14 versions try version four
17:17 um niji is a very specific Japanese
17:20 illustration style but that could be
17:22 something to play with and then you've
17:24 got this stylized low medium and high
17:26 and very high there might be something
17:28 there so I would say go go check your
17:30 settings and also look at all of the
17:32 different options that that you can use
17:34 aspect ratio things like that so yeah
17:38 it's really frustrating when when you
17:40 get these tools not doing what you want
17:42 them to do it's maddening because you
17:44 when you get it right you're like oh I'm
17:46 a genius and then all of a sudden it's
17:49 like nothing's working it's really
17:51 frustrating so I uh I feel for you there
17:54 brother
17:57 um
17:58 um all right and it looks like you're
18:00 helping mid-journey's giving me back
18:02 weird images that don't match my prompt
18:03 at all that's very strange
18:05 I'm even using image uploads yeah that's
18:07 weird there must be something weird and
18:09 I I would say if you haven't done it
18:11 start your prompt from scratch as well
18:13 like retype it maybe there's some weird
18:15 character that you keep copying and
18:17 pasting a weird character and it's just
18:19 freaking it out
18:21 don't forget with these tools they're
18:23 we're very new like a lot of these
18:25 generative AI things mid Journey's been
18:27 around for what two years it's it's been
18:29 sort of working well for
18:32 you know maybe six months maybe eight
18:36 months
18:37 um these things are very raw and very
18:40 janky and there's there's a lot of
18:42 people I I don't know what the latest
18:43 numbers are in mid-journey but it's a
18:45 lot so it's also possible
18:48 um
18:49 oh you know it's worth sorry sorry
18:51 having add moments um just like like a
18:54 bunch of ideas are popping in my head
18:57 um try quitting your browser and
18:58 restarting also try running oh well no
19:01 you're you're in Discord never mind
19:03 strike that I'm wondering if there's a
19:05 way you can reset the cache and Discord
19:07 quit Discord and restart it that could
19:10 help maybe you've got some weird cash
19:11 [ __ ] going on I don't know
19:13 strange strange strange okay
19:20 adding you on Discord what practical
19:24 uses have you found for jet chat gbt
19:28 um lots and lots of them so I've created
19:31 a number of content generators oh it
19:33 looks like we've got a bunch of new
19:34 people here hey everybody my name is
19:35 Kyle Shannon this is the AI learning lab
19:37 these URLs behind my head I'll sort of
19:40 scooch out of the way here
19:41 um these are here just kind of as a
19:43 public service for the community
19:46 um chat.openai.com that's the official
19:49 chati PT website so if you haven't
19:50 played with it go start playing with it
19:52 just go do it now trust me
19:55 um bing.com is also chat GPT it's gpt4
19:59 but it's connected to the internet so
20:01 when you go to bing.com it'll look like
20:03 a Google search but it's it's Microsoft
20:05 search click on the chat button and
20:08 thank you for that that's a lovely uh
20:10 what is that a swan origami swan uh but
20:13 bing.com is gpt4 connected to the
20:16 internet for free and then poe.com is
20:19 another chat CPT tool that'll let you
20:21 play with six different models
20:23 um chat gbt like models
20:26 um prompts.chat will teach you about
20:27 prompting and future PD is a big
20:29 directory of tools so that's what all
20:31 that stuff is
20:33 um I'm an entrepreneur is my background
20:35 I am I like geeking out on this
20:38 generative AI stuff we'll answer
20:40 questions we'll do what I can
20:43 um so what practical uses have I found
20:45 for chat GPT
20:47 um
20:48 anything related to writing it's it's
20:51 incredibly good at so that might be
20:52 things like
20:54 um maybe you want to come up with a
20:56 marketing plan for your business maybe
20:58 you have a marketing plan and you need
20:59 some ideas for copy for a particular
21:02 campaign maybe you need creative briefs
21:05 you know for to come up with ideas for
21:08 the campaign
21:10 um it's really good at coding so if
21:14 you're doing anything code it can you
21:17 can copy and paste code in it it can
21:18 debug it for you figure out what's wrong
21:20 with it it can document code
21:22 um it can do creative writing
21:26 um you can do what I like to call parlor
21:28 tricks like have it write songs and
21:29 poems and you can just start to play so
21:32 what I what I would encourage you to do
21:34 is is if you haven't played with it just
21:36 go play with it right just start asking
21:38 it anything and and let it be kind of
21:41 um
21:42 a toy and then what I would do is how
21:46 how I get to practical uses is rather
21:48 than sitting in front of the tool that
21:50 can essentially do anything
21:52 and say well what could I use this for
21:54 flip it around and I would say look in
21:57 your in your life either professional or
21:59 personal where you've got a specific
22:01 problem you want to solve
22:03 like maybe you want to get in shape and
22:06 you could say hey I want I want to see
22:08 if chat GPT can help me come up with a
22:10 meal plan for the week that help helps
22:12 me eat healthier yes it can it could
22:16 come up with an exercise routine you can
22:18 tell it you know where are you
22:20 um you know I haven't worked out in a
22:21 while I'm just getting started again you
22:23 know I need something easy to to be an
22:25 on-ramp to getting back in shape it'll
22:27 give you a whole plan for that
22:29 um so so you know just start playing and
22:32 start taking specific problems in your
22:35 life and using it to solve them and and
22:38 you'll get to practical uses really
22:39 really quickly
22:42 um Ctrl F to search for Library looks
22:45 like you're helping each other that's
22:46 cool what do I think quantum computers
22:48 do you think quantum computers will be
22:50 integrated with Advanced AI to speed up
22:53 data processing I assume so
22:57 um the the stat that I heard about
22:58 quantum computers is that they're
23:00 doubling their number of qubits every
23:02 year now and and when they add a qubit
23:05 it's an exponential increase in power or
23:08 some crazy [ __ ] like that so I know that
23:10 they just launched they just created a
23:12 500 qubit chip
23:14 someone did so so those things are
23:16 pretty powerful I don't know how far
23:18 they are from viable commercial
23:20 applications right like I don't know
23:22 where and their Continuum they are but I
23:26 assume it's going to do things like
23:27 dramatically speed up the training side
23:30 of AI it might not be so much on the
23:32 real-time side my guess would be it's on
23:34 the training side but you can probably
23:36 train these large language models much
23:39 more quickly much more
23:42 um at a far greater scale and
23:45 potentially using completely new models
23:47 that will make the current Transformer
23:49 model look like Child's Play I don't I I
23:51 don't know enough about the world of
23:53 quantum Computing other than it's weird
23:55 as [ __ ] and as powerful as as it's just
24:00 nuts powerful but it's all I think it's
24:02 all still research right now but I don't
24:04 know
24:06 but basically what Sam Altman said is
24:10 that the more compute power the more
24:12 data and the more compute power you
24:13 throw at these things the better they
24:15 get so
24:16 it will be interesting to see
24:19 um how do I search specific bots on Poe
24:22 let's go look at that
24:26 um
24:27 [Music]
24:28 so we'll pop over to poe.com uh what did
24:32 I do
24:34 what did I do oh wrong window
24:39 okay so here we are at poe.com so so
24:42 you've got you've got your
24:45 um Chachi BT models at the top and then
24:47 you have Bots below that there's this
24:48 explore Bots button so if you hit
24:50 explore Bots now I can go
24:52 I can go over here and they have a
24:55 search button
24:56 tell me they don't have a search
24:59 they may not have a search
25:04 so you can look for history Bots new
25:06 Bots advice Bots game Bots travel Bots
25:09 music Bots
25:11 but it does not look like they have
25:13 search which
25:16 AI Bots
25:18 professional programming Bots oh that's
25:21 cool HTML machine
25:23 let's go play with that one
25:25 um make me a
25:30 um
25:33 website
25:36 about sauerkraut
25:45 there you go
25:47 to test it
25:49 can you write
25:52 the HTML for me
25:58 sure there's your HTML uh it doesn't
26:02 look like you can search for the Bots
26:03 which is weird
26:08 explore Bots you can create a bot create
26:11 a Bots fun
26:12 it's really easy to make your own go to
26:14 create a bot I've done a bunch of those
26:19 the fact that they don't have a search
26:21 engine is almost as unforgivable as the
26:24 fact that open AI doesn't have a search
26:26 for your chat history
26:28 which is just maddening but you know
26:31 what are you gonna do we're early if you
26:33 want to play with these these these if
26:36 you want to play with the most powerful
26:38 technology in the history of mankind
26:40 you know you're gonna have to you're
26:43 gonna have to suck it up and recognize
26:45 that we're early and this stuff kind of
26:47 sucks right now
26:48 all right
26:51 um do you think quantum computers
26:52 answered that today I use Bing thanks to
26:55 you cool how did it go was it good
26:57 anything interesting
26:59 um could you explain exactly what a bot
27:02 is yeah so this is I hate the term
27:05 because people have been talking about
27:07 AI Bots for like a decade and they've
27:09 sucked so historically what a bot is
27:12 is it's it's a um
27:15 it's basically just a giant logic tree
27:17 where where they where you kind of
27:22 you you create a giant tree of if the
27:25 person says this then answer this and if
27:27 they say that then answer that and then
27:29 they got a little bit smarter where you
27:30 didn't have to ask it exactly but
27:33 they've just been shitty for for decades
27:37 Chachi BT is a bot in a sense and and
27:41 basically the bot is you type something
27:43 in and it responds with something right
27:46 that's that's it at its most basic it
27:49 kind of looks like a chat like you're
27:50 having you know like a chat conversation
27:53 on your phone but you're interacting
27:55 with a machine
27:56 historically like the chat Bots that you
27:59 see on customer support sites those are
28:01 those
28:02 um hand programmed you know trees and
28:06 they and they suck because you you can't
28:09 possibly anticipate all of the different
28:12 things someone might ask and so you know
28:14 invariably you're the one that types in
28:16 something that the bot doesn't
28:17 understand and it just either doesn't
28:19 answer you or gives you the wrong answer
28:21 so what what happens with these large
28:23 language models is
28:27 you're you're giving it a specific
28:28 prompt to behave in a certain way it's
28:31 at the at the base sitting underneath it
28:34 is this thing called the large language
28:35 model it's it's basically
28:38 um this thing's been trained on all of
28:40 the internet so it kind of knows all of
28:41 the internet so why chat GPT is so
28:44 powerful is that it's been trained on
28:48 um so much data I'm gonna I'm gonna go
28:51 play and show you some things here
28:54 um
28:57 wait hold please hold wait for it wait
29:01 carry on people carry on rather
29:04 Cheerio pip pip okay
29:08 so
29:10 so in a typical chatbot let's go look at
29:13 one of the PO chat bots so I'll go to
29:15 one that I created
29:17 I created let's see let's do a fun one
29:20 Shakespeare is an obvious one so because
29:23 Shakespeare replies in iambic pentameter
29:29 um
29:30 come on
29:31 what's going on okay
29:33 so this is if you go to poe.com I am
29:36 Shakespeare I can type in
29:39 um
29:40 tell me about
29:43 um
29:46 the smell of wild flowers in the
29:52 springtime
29:55 spring time
29:58 and so you know
30:00 I type that it responded and in this
30:03 case it responded like Shakespeare would
30:06 the springtime yields a soothing site
30:09 indeed when wealth of wildflowers fair
30:13 in Meadows spread with Hues of red and
30:15 blue and gold they plead their sweetest
30:19 sense of Loft breezes shed so
30:22 so in this case the bot is very very
30:25 specific to answer like Shakespeare you
30:28 can also have one that
30:29 writes mid-journey prompts or writes you
30:32 HTML or answers in a specific way
30:35 but the the generic chat GPT you can ask
30:39 it almost anything so so you can just
30:41 say
30:43 um explain
30:46 quantum
30:48 mechanics
30:50 as if I were 5 and use
30:55 um Candy as the metaphor
31:05 and and so it just does that right so
31:07 let's see sure imagine you have a
31:09 magical candy box filled with different
31:11 kinds of candies these candies are very
31:13 special because they can do really
31:15 strange things in the world of quantum
31:17 mechanics candies can be in two places
31:19 at once it's like having a candy that
31:21 can be in your hand and in your pocket
31:23 at the same time it's called
31:24 superposition right so
31:27 so unlike a typical chat bot where
31:31 someone's got a program in all these
31:32 answers anticipating all of the
31:34 questions the large language models just
31:37 know [ __ ] right and and they know
31:39 remarkable amounts of [ __ ] like
31:42 um write me
31:44 a C sharp
31:47 uh
31:49 clone
31:51 of Flappy
31:53 Birds game
31:56 that I can run
31:59 in unity this is one that I did for a 10
32:03 year old and and it just writes you code
32:06 it understands what Flappy Birds is it
32:09 understands what C sharp is it knows
32:11 Unity it writes the code and then it
32:14 tells you how to install the code
32:16 so
32:18 so the chat bot metaphor is really just
32:21 you type in some [ __ ] and it returns an
32:24 answer
32:25 and the bot part is you're interacting
32:28 with a machine not a person
32:32 all right do you think quantum computers
32:34 I got that one already up I just lost my
32:37 place welcome everybody my name is Kyle
32:39 Shannon this is the AI learning lab let
32:41 me
32:42 flip the camera back around so
32:46 I've got this and I got this and I got
32:48 this and then I got this and I got that
32:50 uh if you're new here
32:53 um that top URL there is the official
32:55 chat gbt website I'm not affiliated with
32:57 it in any way in case you're cynical and
32:59 bitter and think I'm just here to scam
33:01 you out of your money giving you free
33:03 advice
33:07 um why do you think normalizing flows
33:10 are becoming so popular why are they
33:12 preferable to to diffusion
33:15 um I don't know I'm not digging into
33:18 normalizing flows that's something I
33:20 need to look up that's new to me
33:23 um
33:24 I would assume just judging by the name
33:27 that if there's if there's something
33:29 about predictability in in output
33:33 um that might be why but but I'm totally
33:35 guessing I don't know what they are but
33:37 I'll definitely look those up
33:39 um how could chat GPT be integrated with
33:41 the Apple Vision Pro to change our lives
33:44 so
33:45 a couple of things about Apple Vision
33:48 Pro you'll notice that they didn't use
33:50 the phrase AI once
33:53 um they didn't use the term metaverse
33:56 once which probably pissed off
33:57 Zuckerberg Tim Cook's probably like
34:02 um
34:04 what they did make very very clear
34:06 though is that so within the goggles
34:08 themselves you've got an M2 Chip so you
34:11 got a an incredibly powerful computer in
34:13 your goggles
34:15 and then there's this thing called the
34:17 R1 chip which I guess stands for reality
34:20 or whatever it stands for
34:22 but that's basically an AI processing
34:25 chip it's taking you know all of the
34:26 input from all these sensors and
34:28 processing it in real time they made it
34:30 very very clear that they're doing most
34:31 of their machine learning processing on
34:33 device that's very important for privacy
34:38 so right now if you use chat GPT you're
34:40 sending your data to open AI
34:43 they're letting you use it for free in
34:45 exchange for training on your data you
34:47 can now ask them to not train on your
34:49 data but you have to do that explicitly
34:51 so Apple's going to be really good at
34:53 the Privacy thing
34:55 so that's one thing
34:57 um a lot of what they're doing is using
35:01 AI to do the processing and the
35:06 taking of the images of the eyes and
35:08 projecting them automatically when
35:10 someone walks into your field of view
35:12 interrupting your full immersion
35:14 experience with their image like a lot
35:17 of what they were showing is full AI
35:21 what Apple's gonna do really well and
35:24 and they haven't announced this yet they
35:26 were very careful not to mix I was
35:29 disappointed in this
35:31 but I understand why they did it they
35:33 were very careful not to mix
35:36 anything like Siri Plus or you know
35:40 their AI products with these these
35:43 glasses
35:44 they were very very careful to say this
35:47 spatial Computing is a new thing we're
35:48 doing it's all about spatial Computing
35:50 right
35:51 under the the subtext of it all is like
35:54 this [ __ ] is all running with AI
35:56 so when when Apple decides to come out
35:59 with AI
36:01 they're gonna have you know some version
36:04 of chat GPT it will probably be Siri
36:06 Plus or the relaunch of Siri or some new
36:08 thing like Siri where it'll be a tool
36:11 that you can interact with you'll
36:12 probably be able to do it with your
36:13 voice it'll probably be close to if not
36:16 real time right with no delay
36:19 where I think Apple's gonna be really
36:21 good is with agents with autonomous
36:25 agents where you'll have your own little
36:27 version of you that you can send out to
36:30 go do tasks for you so you know hey
36:33 agent or hey mini me or you know however
36:36 the [ __ ] they decide to trigger it hey
36:38 mini me
36:40 um the wife and I want to go on a
36:42 vacation two months from now you know
36:45 somewhere in the Caribbean can you go
36:47 find me some deals and and get you know
36:49 give me some options and it will go off
36:51 it'll find that stuff it'll come back
36:52 and say here's some things I found what
36:54 do you like oh I like this one I don't
36:56 like that one go ahead and book that and
36:58 then the next day hey mini me uh we
37:01 decided we don't want to go on that trip
37:02 anymore we're just gonna go on a road
37:04 trip but we want to rent a nice car can
37:06 you cancel that trip and go so I think
37:08 Apple's gonna be really good at having
37:10 these autonomous agents that go interact
37:12 with all of your data and all of your
37:14 services that you're already hooked up
37:16 to in the Apple ecosystem and then of
37:19 course that will also be seamlessly
37:22 incorporated into the space computing
37:24 platform so I think Apple's gonna be
37:28 [ __ ]
37:31 unreal at this generative AI stuff
37:34 um but they're keeping the cards cards
37:37 close to their vest like they like they
37:39 are you know like they do
37:41 that those goggles were in development
37:43 for nine years five thousand patents is
37:47 what I heard there's more than 5 000
37:49 patents in those goggles and it didn't
37:51 surprise me those those were incredibly
37:53 impressive and it was an incredibly
37:55 impressive piece of uh
37:57 piece of Hardware I love this stuff too
38:00 I'm going through the kaggle book right
38:01 now and learning about Gans very fun
38:04 yeah Gans are really interesting
38:06 um Tick Tock a lot of the the filtery
38:10 [ __ ] they do is is using Gans
38:13 um you said how I don't know what that
38:15 was in response to
38:17 do you think there will be a day where
38:18 AI will replace authors screenwriters
38:21 and maybe journalism
38:22 um journalism
38:24 well so I don't know if it will replace
38:27 authors screenwriters the this the
38:30 screenwriters guilds on strike right now
38:31 and one of the things that is is they're
38:33 striking about is they don't want
38:35 producers to use chat GPT to replace
38:37 them
38:38 um
38:39 I think that I think their position
38:41 right now is a little short-sighted but
38:42 I understand it because these tools can
38:45 write screenplay they can write they can
38:48 they can at least block out screenplays
38:51 very very quickly they can do the same
38:54 with
38:55 with novels when you get to journalism
38:57 one of the challenges with
39:00 um with chat EPT in its current form is
39:02 it does a lot of what's called
39:03 hallucinating even if you're connected
39:05 to the internet it will give you
39:07 citations
39:09 um for facts that it just makes [ __ ] up
39:11 like it's not actually aware of what
39:13 it's telling you it looks like it is and
39:15 it will confidently give you answers but
39:17 it's often wrong so I think journalism's
39:19 got a bit to go
39:21 um but I think
39:23 every single
39:26 um
39:28 every single knowledge work position is
39:33 going to be fundamentally altered by
39:34 these tools because these tools are
39:36 going to be rolled into Office 365
39:38 they're going to be rolled into
39:40 Google Google's office suite
39:44 um and and yeah they're gonna they're
39:46 gonna transform everything so I think
39:48 writers
39:50 writers who learn the AI stuff are going
39:54 to be in a in a better position than
39:57 writers who resist it because writers in
40:00 particular so so there's two things
40:02 these tools are so good at writing that
40:04 writers are absolutely existentially
40:07 threatened
40:09 justifiably by these tools but also
40:12 writers are really good with words
40:14 they're writers and in order to get good
40:18 quality content out of AI tools right
40:21 now you got to be good with words you
40:23 got to be able to think in Concepts and
40:25 think critically so I think writers that
40:27 learn AI are going to have a distinct
40:28 advantage over non-writers
40:31 but yeah I think I think it's coming
40:34 free AI video editor go to so the reason
40:37 I have these uh things up on the uh up
40:41 on behind my head my big fat head
40:44 futurepedia.io is a directory of AI
40:47 tools there's
40:50 hundreds probably thousands in there at
40:52 this point so just go into the video
40:54 section there and start start learning I
40:56 don't I stop following all the Tactical
40:59 execution tools because there's just too
41:01 many I mean it's like it's like five or
41:03 six tools a day it's impossible to keep
41:05 up with this [ __ ] like it's it's it's
41:07 enough to just keep up with what's going
41:10 on with the core large language model
41:11 stuff which is a lot that's a lot much
41:14 less all the tools that are being built
41:16 on top of this [ __ ]
41:20 ah all right I mean
41:23 I already had AI generate the next book
41:25 in The Game of Thrones series yeah
41:28 exactly well and there's a here I'll
41:30 show you a cool thing if you haven't
41:32 seen this
41:34 um this is pretty trippy
41:37 um
41:39 so if you go to poe.com
41:43 um
41:44 where are we here okay
41:47 so you you've got Sage GPS or GPT for
41:53 Claude plus Claud instant 100K Claude
41:55 instant and chat EPT all of these are
41:57 chat GPT models the the three that say
42:01 subscribed you have to pay for it right
42:03 so so this one isn't free but if you pay
42:05 the subscription for Poe and again I'm
42:07 not affiliated I don't give a [ __ ] if
42:09 you do it or not but you get access to
42:11 this thing called Claude instant 100K
42:13 100K means you can paste in up to 75 000
42:18 words
42:20 into the prompt and it won't break chat
42:23 EBT by comparison does about three
42:26 thousand words and they're upping that
42:27 to I think 12 000 words here shortly but
42:31 right now it does three thousand words
42:32 so this is 75 000 words you can pop in
42:35 the prompt so you could take the
42:37 manuscript of a novel
42:39 paste the whole thing in there and go
42:41 write me the first chapter of the sequel
42:43 or or give me give me three mini
42:46 outlines of
42:48 possible SQL plot lines
42:51 um and it will it'll give you that
42:54 um like that's nuts I I put in um I put
42:58 in the screenplay I just pasted the
43:00 whole screenplay of 2001 A Space Odyssey
43:03 and started asking the questions about
43:04 it and it it read the whole thing in a
43:07 second or two or you know five seconds
43:08 and then started answering questions
43:10 about the screenplays it's remarkable
43:13 make a bot to search for the Bots oh
43:15 that's a good idea yeah that would be
43:17 probably pretty straightforward assuming
43:20 you could actually
43:23 search their directory you might need to
43:25 write a plug-in for that
43:27 a bot is a big tree yeah AI makes its
43:32 own questions and its own answers
43:34 yeah it was interesting but I'm new to
43:36 it getting creative every day
43:40 cool the whole point of this channel is
43:43 get curious about AI
43:46 um
43:47 my background if it looks like we've got
43:49 some new folks here my name is Kyle
43:50 Shannon this is the AI learning lab I am
43:53 not a machine learning specialist I'm
43:55 not a computer scientist I'm an
43:56 entrepreneur I've started 12 or 15
43:58 companies in my day
44:00 um I was there for the early days of the
44:02 World Wide Web started in one of the
44:04 first digital agencies and got to
44:06 experience
44:08 um a transformative technology from its
44:10 birth the world wide web and
44:13 um
44:15 I didn't think I'd see another one of
44:16 those in my life and Along Comes
44:18 generative AI so the the birth of chat
44:21 GPT for me is a seminal moment where
44:24 there's an inflection point where we
44:25 shift from scientists and researchers to
44:28 regular people having access to these
44:31 tools and and just to put put a little
44:34 bow on how quickly this technology is
44:37 being adopted I know there's a
44:38 tremendous amount of hype about Ai and
44:41 it's really [ __ ] exhausting
44:44 um
44:45 Chachi BT went from zero to a hundred
44:48 million users in six weeks
44:51 it was the fastest adoption of
44:53 technology in history
44:54 it it
44:57 only took him two weeks to write chat
44:59 GPT there's a whole story there
45:01 it got to 100 million users in six weeks
45:04 the World Wide Web when it from when it
45:06 was started to when it got to 100
45:08 million users was six years
45:11 and we know how transformative that was
45:13 so so that's part of what this channel
45:16 is about is just
45:19 I don't know screaming at the top of my
45:21 lungs like pay attention to this [ __ ]
45:24 it's coming it's here it's coming faster
45:26 than you realize it knows other
45:29 languages too it knows all the languages
45:31 like
45:32 what was I do oh I was on a uh was I on
45:35 a live
45:38 I was on a I was on a live or oh no uh
45:41 in in my in my Tick Tock Channel
45:44 um in a number of my videos people are
45:47 putting comments in in their languages
45:49 and so what I'm doing is I go in I copy
45:52 it out I paste it into chat GPT I say
45:55 first tell me what language this is and
45:57 then you know put it in English and it
46:00 writes it in English and then I say you
46:02 know respond here
46:04 um and I give him my response and I say
46:06 translate it back and then I copy and
46:07 paste that back so
46:09 um yeah the language translation stuff
46:11 is mind-blowing it's very very good at
46:13 that
46:14 someone asked who would win Call of Duty
46:16 tournament over the weekend and it
46:18 predicted it correctly yeah if it's got
46:20 the right data like one of the things I
46:21 I assume they they popped a bunch of
46:23 data in there and based on that data it
46:26 did some analysis it's really good at
46:28 analyzing data so one thing here here's
46:31 the trick if you haven't played with
46:32 Chachi BT
46:34 um
46:36 if you give it a very generic answer
46:39 it's going to give if you give it a very
46:40 generic prompt question it's going to
46:42 give you a very very generic answer and
46:44 so a lot of people get turned off with
46:46 chat gbt because they're like oh it gave
46:48 me like what it wrote was boring it
46:50 wasn't all that interesting
46:52 give it an example
46:54 and it mimics really well so if you've
46:57 got like you know the opening to a novel
46:59 that you wrote or a song that you wrote
47:01 or a business email that you wrote that
47:04 you've got your own particular voice you
47:06 can say based on this email
47:09 shift return to do a line break paste
47:12 your email in
47:13 and then say write me a new email to
47:16 this person talking about this subject
47:19 using that as an example and just watch
47:22 what it does
47:23 the shit's amazing
47:25 it really is amazing
47:27 all right does it work in other
47:29 languages too it does they understand
47:32 context
47:33 they understand context is the big thing
47:37 I don't know what that's in reference to
47:38 but context is really important so so
47:40 context is basically here's the thing
47:42 with context
47:45 tell chat gbt more than you think is
47:48 reasonable like it's almost like this
47:50 Google has taught us to put to use as
47:53 few words as possible Right put make
47:55 your your
47:57 prompt for Google as concise as possible
48:00 and then they'll figure it out from
48:01 there Chachi PT is kind of the opposite
48:03 if you give it a shitty little concise
48:05 prompt it'll just give you generic [ __ ]
48:08 answers
48:08 but if you think of it more like a like
48:11 a
48:13 a collaborative partner that like an
48:16 employee that you work with that just
48:17 happens to be remote and you start
48:19 talking to it like that
48:20 hey chat EBT I want you to act like an
48:24 expert marketer and I'm struggling with
48:27 my marketing right now and you know
48:29 here's what my business is about and
48:32 here's what we've tried in the past and
48:34 it didn't work and like all the [ __ ]
48:36 that you would tell a person if you were
48:38 getting them up to speed in your
48:39 business tell it that and then say can
48:41 you help me come up with some ideas I
48:43 want to brainstorm ideas for you know
48:45 how we can do something different than
48:47 we've done before that that you think
48:49 will work based on on what you know and
48:52 it will start interacting with you just
48:54 in a whole different level so that
48:57 context is super super important
49:00 okay
49:01 um if you've got questions about AI pop
49:03 them in the comments below I am
49:05 painfully behind on questions but I do
49:07 try to answer them all so I'll get to it
49:09 eventually or not
49:13 um
49:18 all right
49:20 I'm back to where I was okay
49:22 I read that you can have chat EPT
49:24 compress your prompt to use less tokens
49:27 can you show an example
49:30 um
49:32 yeah I haven't really done that
49:36 um
49:38 I'm trying to think of a good example
49:39 that that would make sense
49:46 well okay so so here let me try
49:49 something this is I'm kind of
49:52 um making it up as I go here because I
49:54 don't
49:55 quite know what what you want to know
49:57 but so so there's a couple of important
50:00 things here so I'm in I'm in chat EPT
50:02 the official website chat.openai.com
50:07 um
50:09 so when you
50:13 um
50:17 chat EBT has what's called a context
50:20 window of 4 000 tokens
50:23 and the context window think of it kind
50:25 of like this sliding window that the
50:29 older stuff in that window sort of drops
50:31 off the top like it'll forget stuff at
50:33 the top so if you and and it translates
50:35 to about three thousand words so if I
50:37 put in like 2500 words here and then I
50:39 start having a conversation with it it
50:42 will start to forget stuff at the
50:43 beginning of the conversation
50:45 so you can tighten things up so let me
50:47 here's here's an example that I I just
50:49 thought of that that might give you an
50:51 idea of of how you could use this so I'm
50:54 going to literally I went to my this is
50:55 my LinkedIn profile
50:57 so I'm going to my LinkedIn profile and
50:59 I'm going to literally just copy you
51:01 know whatever I can see here there's not
51:03 a ton of stuff here
51:04 but
51:05 I'm just going to copy that I'm going to
51:07 go back to chat GPT and then I'm going
51:09 to say
51:11 um here's
51:13 a LinkedIn
51:15 profile
51:18 colon and I'm going to hit shift return
51:19 to make some line breaks I'm just
51:21 pasting in all my crap
51:23 and then I'm going to say please
51:26 um
51:27 consolidate
51:30 um
51:30 the relevant
51:34 data in
51:36 this profile so I have a summary
51:44 of
51:46 this person's experience in
51:50 as
51:52 few tokens as possible
51:58 and so it read all of that stuff and
52:01 then
52:06 it basically wrote
52:09 that small paragraph so so it took from
52:12 from all of this stuff right which is
52:14 there's a lot of wasted [ __ ] in
52:16 there
52:17 and it got rid of all that and then just
52:20 Consolidated Consolidated it down into
52:22 this this little thing so if I'm having
52:24 a conversation going back and forth and
52:26 back and forth and back and forth
52:29 um
52:29 and it and it starts to forget the
52:32 things above you could potentially go
52:34 grab some of the things above that are
52:36 relevant tell it to compress it into a
52:38 smaller chunk and then kind of insert
52:40 that later in the conversation so that
52:42 it doesn't forget that history the other
52:45 thing you can do is I mentioned this
52:46 before you can you can go to poe.com and
52:51 use the Claude instant 100 000 token
52:55 um model right so so this has got a
52:58 context window of 75 000 words so this
53:01 won't forget stuff uh nearly as as much
53:04 so hopefully that gives you what you are
53:07 asking for I I am kind of making it up a
53:10 little bit here because I haven't done a
53:11 ton of that but it is frustrating when
53:13 you're
53:14 things are going really well in in your
53:17 in your chat with the thing and all of a
53:19 sudden it kind of forgets what the hell
53:21 where the conversation started and it'll
53:23 just kind of lose its mind and you're
53:25 like ah [ __ ] and if in my case because
53:28 I've got 80d I'm like I don't feel like
53:30 going and dealing with that but that
53:32 thing that I just did would help
53:34 with that
53:36 all right
53:39 what does it prefer
53:41 to call itself besides its programmer
53:45 given name I don't know ask it
53:48 um Let me let me go well I'm gonna go
53:50 ask three different things I'm going to
53:51 ask Sage what is your name
53:56 my name is Sage
53:58 let's go to gpt4 what is your name
54:05 I am gpt4 that's [ __ ] boring
54:09 I'm gonna ask Claude what is your name
54:14 my name is Claude
54:15 um when the when the New York Times
54:17 Reporter was um playing with bing.com
54:20 and he was having like a four-hour
54:23 conversation with with uh gpt4 at
54:27 bing.com when it first came out and he
54:29 got it to admit that it's it's like it's
54:31 working name
54:33 um internal name was Sydney
54:35 um and then at some point Sydney snapped
54:37 and told Kevin Roos that it loved her it
54:40 loved him and he should leave his wife
54:42 for GPT for for Sydney that made the
54:46 cover of the New York Times
54:49 um so I don't know like you you can you
54:50 can do some jailbreaking kind of things
54:52 where you get it to ignore it's it's
54:54 sort of pre-programmed guide posts
54:57 um and and probably get it to tell you
54:59 its internal name
55:01 um I wonder if it can accurately predict
55:03 winning teams in sports I haven't tried
55:04 again I think I you know I I don't know
55:08 that it can predict winning teams but I
55:10 think what it probably can do really
55:11 really well is look at historical data
55:14 and analyze the [ __ ] out of it
55:17 and when if you haven't seen
55:19 um if if you're interested in that kind
55:21 of thing go to YouTube and and do a
55:24 search for code interpreter chat GPT
55:27 code interpreter chatbt just go watch
55:30 some videos on that and watch some data
55:32 analytics guys lose their [ __ ] mind
55:34 when they see what this thing does
55:36 and that's coming like that's like it's
55:38 out there some people have access to it
55:40 now I don't and I'm bitter about it but
55:42 you know what are you gonna do all right
55:44 do you know
55:45 um about how to get started with open
55:47 ai's whisper API for text to speech
55:49 we've Incorporated it in Story by my
55:52 company storyvine we use Whispers
55:55 um we're not using the API we're using
55:57 like a local install of whisper to do uh
56:00 transcriptions of video
56:02 um
56:03 I will show you this I'll show you how
56:05 you can play with it how you call it is
56:09 um I'm sure it's in the documentation
56:14 but so instead of going to chat GPT
56:16 which is this thing we're going to go to
56:23 um platform.openai.com
56:26 slash playgrounds
56:29 and that will take us to the
56:32 um the the playground where where
56:34 developers can play with
56:38 um
56:39 the the GPT models and so
56:42 um you know you can you can do chat
56:44 interfaces or you can do just normal
56:48 um
56:49 um you know sort of API calls and things
56:51 like that so basically this works the
56:53 same way like
56:54 um like write me and
56:58 Eminem
57:01 Eminem
57:02 whatever poem
57:05 about uh red wagons
57:11 in the turn
57:14 and it will it will write so this is
57:16 where like if if you're working with an
57:18 API this is where you sort of dial in
57:21 your prompts
57:22 so in the upper right hand corner of
57:24 this window there's a little teeny tiny
57:27 microphone
57:28 and that's that's access to whisper so
57:31 if you click on that
57:33 it says upload audio from your computer
57:35 you can upload videos or audio
57:37 and or you can just start recording and
57:40 it will automatically do the
57:41 transcription for you
57:43 um so if you go to the documentation
57:50 so now I'm in the documentation
57:52 let's see
57:55 speech to text overview
57:58 so here's your whole
58:00 instructions for how to use whisper AI
58:04 in the documentation
58:06 so
58:07 platform.openai.com playground will get
58:11 you to this place where you can play
58:13 with it by clicking that little
58:15 doohickey right there and then go to
58:19 documentation
58:20 and that will give you the documentation
58:22 for the API so hopefully
58:25 that gives you the answer you need
58:29 I've got I didn't Implement that our
58:32 Engineers did so I don't
58:35 have the specifics on exactly what to do
58:38 but that should at least get you pointed
58:39 in the right direction why is it
58:41 important to regulate AI do you think
58:43 the government will enforce it well
58:50 so
58:53 it's important to regulate it because
58:55 this stuff's really powerful and and
58:57 there are
58:59 tools out there called autonomous agents
59:03 there's one called Auto GPT and there's
59:04 another one called Baby AGI
59:07 and what those things do is rather than
59:09 give it an explicit like write me a poem
59:12 or write me a document you give it a
59:14 goal and the goal could be
59:17 um go find all the
59:20 um
59:22 intelligence about my top 10 competitors
59:25 and it will start writing its own
59:27 instructions for how to accomplish that
59:29 goal it'll say okay let's use Google and
59:32 go find out who are the top 10
59:33 competitors of of this company and then
59:36 let's let's go understand what they're
59:38 trying to do strategically and it it
59:40 literally starts writing its own
59:42 um rule prompts and and triggering tools
59:47 on my behalf which is great if I ask it
59:50 something
59:52 non-nefarious but what if I said hey go
59:56 find all the vulnerabilities in
59:59 computers running this version of
1:00:01 Windows exploit that vulnerability and
1:00:05 you know send me back links with access
1:00:07 to all of those machines
1:00:09 that's the kind of thing where you could
1:00:10 you can very startly very quickly start
1:00:13 to get into trouble
1:00:15 um and and there's just there's just
1:00:17 sort of piles and piles layers and
1:00:19 layers of places we can we can get in
1:00:23 trouble
1:00:24 um you know to the to the point at which
1:00:26 you know some of the the if if you play
1:00:29 this thing out far enough and these
1:00:31 machines get smart enough
1:00:33 if they start to prioritize their
1:00:36 existence over human existence
1:00:40 they could probably figure out ways to
1:00:42 take us out right
1:00:45 um so that's why you want government
1:00:46 regulating it now what's the problem
1:00:49 at least in America our government
1:00:51 doesn't know how [ __ ] Wi-Fi works
1:00:53 much less how these tools work
1:00:57 um so we need to very very quickly get
1:01:00 smart people in Washington so Sam Altman
1:01:04 from openai has been pretty aggressive
1:01:05 about saying you need to regulate us
1:01:07 here's how to do it
1:01:10 um Jeffrey Hinton who's whose nickname
1:01:12 is the Godfather of AI he invented a lot
1:01:14 of the early technology that is is
1:01:16 underpins all of what we do today
1:01:19 um he just quit Google so that he could
1:01:21 talk about the risks of AI and talk
1:01:25 about it in an intelligent way without
1:01:26 you know being skewed by who he works so
1:01:29 so yeah it's really really important
1:01:33 um and and the horse is a bit out of the
1:01:35 barn right Meta Meta released their
1:01:38 their core large language models to the
1:01:40 developer community and now they're
1:01:41 they're out there so a lot of these
1:01:43 powerful
1:01:44 large language models are open source
1:01:47 right now so they don't have any safety
1:01:49 guard rails on them so I don't know
1:01:51 quite how you regulate all of that um
1:01:54 but but yeah it it needs to be done and
1:01:56 it
1:01:58 I think it needs to be done
1:01:59 internationally right it can't just be
1:02:01 done with this country is going to do
1:02:03 their thing and this is going to do
1:02:04 theirs the EU right now has got some
1:02:07 proposals out for their thing but
1:02:11 like if if you make it too restrictive
1:02:14 the rest of the world just kicks your
1:02:16 ass at Ai and starts really exploiting
1:02:18 the [ __ ] out of the fact that you made
1:02:19 it too restrictive if you don't make it
1:02:21 restrictive enough then you gotta you
1:02:24 could you know threaten your National
1:02:25 Security
1:02:28 um and and this isn't this is an
1:02:30 international problem this is a global
1:02:32 this technology is everywhere right and
1:02:35 it's going to be everywhere so um so I
1:02:37 think it's got to be some sort of
1:02:38 international body probably like the
1:02:40 nuclear
1:02:41 um the nuclear the international nuclear
1:02:44 bodies
1:02:46 um I suppose they're going to make a new
1:02:47 series talking about Apple yeah I think
1:02:49 they would I don't know if they'll still
1:02:51 call it Siri because Siri's got a bit of
1:02:53 a black eye because it it sucks
1:02:57 um a true digital assistant powered by
1:02:59 AI yep exactly
1:03:02 um I see eventually each Siri plus will
1:03:04 be unique to each person and user yeah
1:03:06 and I could also see us like I could see
1:03:08 like apple would probably be pretty good
1:03:11 at I could have multiple agents right I
1:03:13 could have a social agent like this
1:03:14 one's just all it's going to do is make
1:03:16 sure that my social life is in some sort
1:03:18 of order this is going to be my
1:03:20 professional agent that just books me
1:03:22 sales meetings and you know I could see
1:03:24 us having different kinds of Agents
1:03:28 um chat gbt is helpful but can't do my
1:03:30 job better
1:03:32 it would destroy companies there's a
1:03:35 really fascinating article by Ethan
1:03:37 Malik that just came out I just did a
1:03:39 thread on it on the everyday AI Twitter
1:03:42 account
1:03:44 any any and he talks about
1:03:47 um the secret cyborgs in your company
1:03:50 so if you go to go to Twitter and go to
1:03:53 Everyday AI news hang on let me
1:03:56 let me change something here
1:04:01 go to Twitter and go to that um tag
1:04:06 everyday AI news that's my newsletter
1:04:08 the
1:04:10 thread
1:04:13 the latest thread that we put out is on
1:04:15 on Ethan Malik's article
1:04:17 and it's called
1:04:19 um detecting the secret cyborgs and what
1:04:22 he's talking about is
1:04:24 within companies right now individuals
1:04:27 are using chat GPT to do their job more
1:04:30 efficiently
1:04:31 but
1:04:32 the companies themselves this stuff's
1:04:35 too new and janky for the organization
1:04:39 to use it but the employees are afraid
1:04:41 to talk about
1:04:42 that they're using it
1:04:44 and so so because they don't want to get
1:04:47 fired
1:04:49 um they still want like like if if let's
1:04:51 say you're a copywriter and you've
1:04:53 figured out how to make chatbt do your
1:04:55 writing for you in like an hour when it
1:04:58 used to take you two days
1:05:00 a lot of employees don't want their boss
1:05:02 to know that right so so there's there's
1:05:04 already this Gap forming where people
1:05:06 are using and innovating with AI but the
1:05:09 companies aren't benefiting from it
1:05:10 right because they don't know about it
1:05:12 and so this whole article is about these
1:05:14 secret cyborgs that are innovating
1:05:16 within companies without without
1:05:18 benefiting the company and so they're
1:05:20 gonna have to somehow bring those
1:05:21 together and the company's gonna have to
1:05:23 do [ __ ] like it's okay to use AI in fact
1:05:26 I want you to use Ai and I promise you I
1:05:29 won't fire you if you are tell me what
1:05:32 you're doing
1:05:33 um so
1:05:35 um I don't quite know what your job is
1:05:38 but I would push really hard on chat GPT
1:05:41 to see if you can get it to do your job
1:05:43 as good as you can
1:05:45 um just because it's not doing it right
1:05:46 now that could be a prompting thing so I
1:05:48 would dig deeper on that
1:05:50 um because I think every job is is is in
1:05:53 some form of
1:05:55 um reimagining Google is experimenting
1:05:58 with emotional tonal response in
1:06:00 responses with Google home
1:06:02 um yeah I think that I think that um
1:06:06 while the way these large language
1:06:08 models work they don't feel any of the
1:06:11 feels they're really good at
1:06:13 communicating nuance and and empathetic
1:06:16 language right they're good at
1:06:18 understanding that if someone's in pain
1:06:20 you talk to them different and it's good
1:06:23 at predicting how I should talk to
1:06:24 someone so if you if you put that into a
1:06:28 system that that has an emotional
1:06:30 response response layer built into it I
1:06:33 think that could be quite powerful
1:06:35 it's often wrong with regards to I.T
1:06:38 yeah so chat GPT gets [ __ ] wrong all the
1:06:42 time so here's the way I think about
1:06:44 Chach EBT it gets me to 80 like that the
1:06:48 last 20 is up to me
1:06:50 um don't just the other thing to think
1:06:52 about with with these large language
1:06:54 models
1:06:56 don't think of it like a Google search a
1:06:58 Google search is like you put in one
1:06:59 thing and you get back your answer and
1:07:01 you're done unless it you know you can't
1:07:03 find your [ __ ] and then you do it until
1:07:04 you get it right
1:07:06 don't don't think of it so much like
1:07:09 that if you give it a prompt and it
1:07:10 gives you back something shitty
1:07:13 think of that as the beginning of a
1:07:15 conversation a collaborative
1:07:17 conversation where you get an answer
1:07:19 back if you don't like it you say hey I
1:07:21 don't like that like you know that was
1:07:22 too generic and you know and you missed
1:07:26 this point which I think is the most
1:07:27 important point in the piece can you
1:07:29 give me another version of it boom it'll
1:07:31 give you another version of it so what
1:07:33 I'm finding is that I'm actually going
1:07:35 back and forth you know a dozen two
1:07:37 dozen times to get to a first draft to
1:07:41 get to an 80 percent
1:07:43 and then I kind of take it from there
1:07:45 and
1:07:47 dial it in and fix all the generic
1:07:49 [ __ ] and it gets hyperbolic
1:07:50 sometimes and it uses too many words
1:07:52 sometimes
1:07:54 um
1:07:54 so so don't set the expectation that it
1:07:57 has to be perfect think of it more like
1:07:59 can this accelerate me from zero to
1:08:02 eighty percent and the answer is yes it
1:08:05 could accelerate at like 100x faster
1:08:08 and then you can take the two hours you
1:08:10 would have taken getting to eighty
1:08:11 percent and just use that to polish it
1:08:13 so I'm finding the quality of my work is
1:08:15 going up and the quantity of my work is
1:08:17 going up
1:08:20 um AI weather forecasters those will be
1:08:22 amazing I know part of 364 powerapps but
1:08:25 it's no more than a search result not
1:08:28 right now it's going to be way more
1:08:31 than a search result
1:08:35 um what is the Discord name uh oh so mid
1:08:39 Journey's got their own Discord
1:08:42 um the the Discord for the the AI Salon
1:08:45 is in this is this points to a link tree
1:08:47 which has got our meet up when it shows
1:08:50 when we meet our next meetings tomorrow
1:08:51 night Tuesday and then there's also the
1:08:54 link to the Discord there it's the AI
1:08:56 Salon AI can help pulling synonyms and
1:09:00 rewriting boringly worded sentences it
1:09:02 can do that really really well like
1:09:04 really really well like like if you're
1:09:07 like you know give me 20 words that mean
1:09:10 this that start with s and you know and
1:09:14 and don't have an e in them
1:09:17 here's all the words and you're like oh
1:09:19 yeah I would have thought of those it's
1:09:20 like holy [ __ ] it's really good at that
1:09:23 stuff is that a Long Island iced tea
1:09:25 you're drinking don't you wish it's an
1:09:27 iced coffee
1:09:28 so it's a it's a Denver iced coffee but
1:09:31 there's no booze in it
1:09:37 you can almost tell by the sound of the
1:09:39 ice that's someone who spends some times
1:09:41 in a bar
1:09:43 uh but is using chat EBT for copyright
1:09:45 and considered plagiarism no it's not
1:09:47 the the way well okay
1:09:49 it's a very gray area but
1:09:52 one of the misconceptions about the way
1:09:54 large language models work is that
1:09:56 They're copying and pasting from copy
1:09:58 written work and kind of you know
1:10:00 pasting it together it's not the way
1:10:01 they work
1:10:03 it's a very complicated mathematical
1:10:05 formula it's actually predicting word by
1:10:08 word the next most probable word in your
1:10:11 answer so it's literally writing an
1:10:13 original piece of content word by word
1:10:15 based on you know all the stuff it's
1:10:18 trained on in response to your prompt so
1:10:21 if you write a really generic prompt it
1:10:23 does it it makes it really generic
1:10:25 because it's kind of sampling from
1:10:26 everything
1:10:27 if you give the really specific prompt
1:10:29 with a lot of context it'll kind of
1:10:31 narrow the pool of things that it it
1:10:33 draws from
1:10:34 but but you can do things like say you
1:10:37 know write in the style of Ernest
1:10:38 Hemingway it knows all of Ernest
1:10:41 Hemingway's work it understands what in
1:10:43 the style of means
1:10:45 and you can tell it to write something
1:10:46 in the style of that person and it will
1:10:49 create an original piece of content that
1:10:51 is not plagiarized it's not copying and
1:10:53 pasting but it will have the tone and
1:10:55 feel of
1:10:57 um of Hemingway
1:10:59 so that's where it can get confusing
1:11:01 because if you get it close enough to
1:11:02 his stuff it might you know put a quote
1:11:06 in there that's actually from something
1:11:07 right so
1:11:09 um
1:11:11 the the whole copyright thing like what
1:11:13 these things are trained on
1:11:15 is probably the next 20 years of our
1:11:19 existence are going to be nothing but
1:11:20 lawsuits and counter lawsuits about who
1:11:23 owes money to who so I'm just trying to
1:11:26 ignore that right now on the image front
1:11:28 if you work for a big company use Adobe
1:11:31 Firefly and generative AI over
1:11:34 mid-journey and stable diffusion for
1:11:36 this very reason
1:11:38 um mid-journey and stable diffusion are
1:11:41 already being sued like big lawsuits
1:11:44 Adobe has trained their image generation
1:11:47 tools on public domain and Adobe stock
1:11:49 images they own all the rights to so
1:11:52 they're safe for commercial use so
1:11:54 there's that's a little tip with the
1:11:56 large language models it's not nearly as
1:11:57 clear
1:11:59 what happens if you put the Bible in it
1:12:01 knows the Bible it probably knows all of
1:12:03 the versions of the Bible it probably
1:12:05 knows all of the books that were not
1:12:07 included in the Bible and when they were
1:12:10 included and not included and who did it
1:12:12 like just go play go tell uh go tell
1:12:15 Chachi BT to act like a uh a Biblical
1:12:19 scholar and have fun dig deep and
1:12:24 I bet you there's some some good stuff
1:12:26 going on there
1:12:29 um how do I find your Discord I did that
1:12:31 before go to the salon.ai it's in the
1:12:34 link tree there
1:12:37 um what do you think of the stories
1:12:39 plug-in my kids are addicted I don't
1:12:42 know the stories plug-in let's go play
1:12:44 I like finding new stuff
1:12:48 [Laughter]
1:12:49 [Music]
1:12:51 yeah we call them loser of the day
1:12:55 stories I have not heard of stories
1:12:59 the other thing about plugins is they
1:13:01 don't always
1:13:02 um
1:13:04 like like I'll see I'll see plugins on
1:13:06 Twitter they're like go play with this
1:13:08 plug-in and then I can't find it
1:13:10 So Stories
1:13:13 stories create beautiful Illustrated
1:13:16 stories easily this is super cool I have
1:13:20 not played with this okay so I'm gonna
1:13:22 show you about plugins here and I'll
1:13:25 give you I'll give you a little bit of
1:13:26 uh
1:13:28 a little bit of primer okay so if you
1:13:32 come to
1:13:33 chat GPT chat.openai.com if you pay the
1:13:37 20 bucks a month that gives you plus the
1:13:40 little yellow flag here and with plus
1:13:42 you get access to
1:13:44 gpt4 default connected to the internet
1:13:47 and plugins now
1:13:50 if you've paid your 20 bucks and you
1:13:52 only see GPT for default you don't see
1:13:54 these other things that's because you
1:13:56 have to enable those things so if you go
1:13:58 to your settings
1:14:00 in your settings there's a there's an
1:14:02 area called beta features you have to
1:14:04 turn on browse with Bing and plugins and
1:14:07 there will be other things coming soon
1:14:08 like code interpreter which I want I
1:14:11 want this to say code interpreter but
1:14:13 it's not there right now this makes Kyle
1:14:15 cranky that see this Blank That's My My
1:14:18 Rage in angers fills that Darkness okay
1:14:21 so you turn on the beta features and
1:14:23 then that allows you to access
1:14:27 chat gbd connected to the internet and
1:14:29 plugins now how plugins work is you've
1:14:33 got this plug-in store that I was just
1:14:35 in so you install the plugins that you
1:14:37 like
1:14:38 and then in any given chat session you
1:14:41 can activate or you can enable up to
1:14:43 three plugins so I've got ask your PDF
1:14:46 and I've got web pilot
1:14:50 enabled I'm going to turn those off and
1:14:52 now we've just got stories
1:14:55 um done so create beautiful Illustrated
1:14:57 stories easily this is cool okay
1:15:00 so the way plugins work it's it's it's
1:15:02 just this normal chatbot interface I'm
1:15:04 going to say write me as
1:15:07 a story
1:15:10 um about
1:15:14 um
1:15:15 two best friends that are animal let's
1:15:21 see that are an unlikely pair of animals
1:15:27 on Galapagos
1:15:30 or ghost galap a ghost
1:15:37 Island
1:15:40 during the time of
1:15:44 Darwin
1:15:47 and so now it should fire off the
1:15:51 plug-in so it did and then when these
1:15:53 plugins go
1:15:56 sorry I'm trying to get it to focus
1:15:57 focus
1:15:59 come on you dang dummy
1:16:04 so let's see what it did you can open up
1:16:06 this thing
1:16:07 so it went out it did something and then
1:16:10 it gave us a URL so it said here's the
1:16:12 story generated on your prompt The Odd
1:16:15 Couple of Galapagos Island when a finch
1:16:17 and a tortoise become best friends they
1:16:19 set out on an adventure the direct URL
1:16:22 The Odd Couple of Galapagos Islands
1:16:26 storybird.ai
1:16:30 and there's two little animals
1:16:34 oh this is super cool oh look at that
1:16:36 that's so cool it's the the tortoise is
1:16:40 on Wheels and the little birds on top of
1:16:43 him
1:16:44 okay let's go back let's read this once
1:16:46 upon a time when Galapagos Islands uh
1:16:48 during the time of Charles Darwin there
1:16:50 were two this is super cool so so
1:16:53 there's something really neat going on
1:16:55 here that is is not super obvious
1:16:59 so all of these plugins
1:17:03 um
1:17:04 are basically just kind of fancy little
1:17:06 hooks back to
1:17:10 um existing websites in the world right
1:17:13 so diagram is a diagramming website
1:17:17 Earth does some geography stuff
1:17:20 experiences is a travel thing get your
1:17:22 guide as a travel thing
1:17:24 um kayak instacart right so
1:17:28 so what the plugins are doing is they're
1:17:30 allowing you to to put this large
1:17:32 language model layer on top of it so
1:17:35 this this kids book one I assume that
1:17:38 this
1:17:39 that this website you can make these
1:17:42 stories anyway right if I go to create
1:17:44 oh see they've they've rolled
1:17:47 they've rolled the large language model
1:17:48 on top of here that's really cool and
1:17:51 there's their plug-in so they're
1:17:53 teaching you how to use it oh this looks
1:17:54 like it's new this is super cool
1:17:57 and there's a library of all the stuff
1:17:59 that they're doing
1:18:00 so what they've done is they've sort of
1:18:02 they've sort of pre-prompted the image
1:18:04 generator to follow a certain Style
1:18:07 right
1:18:09 really slick thank you for sharing that
1:18:11 that's super cool all right
1:18:14 I'm telling you man daily my mind is
1:18:17 just like
1:18:19 daily it's just nuts it's nuts I'll tell
1:18:24 you it's just crazy it's crazy
1:18:29 what do you my kids are addicted okay
1:18:32 that was super cool thank you for
1:18:33 sharing that who shared that Steve that
1:18:34 was awesome all right what if years
1:18:37 later we figure out that Kyle was a bot
1:18:39 the whole time it turns out he was chat
1:18:41 GPT five
1:18:44 [Laughter]
1:18:50 thank you
1:18:53 all right
1:18:55 GPT went to the moon GPT seems better at
1:18:58 translation as you
1:19:01 can ask for literal creative or causal
1:19:03 translations as well yeah there's
1:19:06 something about giving
1:19:10 GPT lots of context like give it a
1:19:13 spreadsheet give it a screenplay give it
1:19:16 some piece of writing one of the things
1:19:18 I've been doing is
1:19:20 as I'm as I'm doing more this AI stuff
1:19:22 I'm getting invited to do speaking gigs
1:19:24 and like when I when I first started
1:19:26 getting invited to do gigs they're like
1:19:27 hey can you send me your AI profile like
1:19:30 I'm like I don't have an AI profile but
1:19:32 I've got a LinkedIn profile
1:19:34 so what I would do is I just copy my
1:19:37 LinkedIn profile copy
1:19:39 um
1:19:40 a bunch of previous profiles that I had
1:19:42 thanks for the heart very nice
1:19:45 and then I just sort of popped them all
1:19:47 into chat apt and I'd say based on that
1:19:49 LinkedIn profile and those other
1:19:50 profiles write me a new profile that
1:19:52 includes all my AI [ __ ] in it they're
1:19:54 just out it comes it's really good at
1:19:57 that really good at that
1:20:00 oh [ __ ] Tick Tock just dumped me to the
1:20:03 bottom of the comments I just lost my
1:20:04 place in the questions hold please old
1:20:06 man trying to find tiny text on tick
1:20:09 tock
1:20:10 go go go have your Long Island Iced Tea
1:20:14 uh looks like we have some new people
1:20:17 here hold please
1:20:20 my name is Kyle Shannon this is the AI
1:20:22 learning lab welcome welcome welcome if
1:20:24 you haven't played with chat GPT that's
1:20:27 what that URL is right there
1:20:28 chat.openai.com that's the official
1:20:30 website go there and start playing just
1:20:33 do it trust me
1:20:34 it'll be fun
1:20:36 ask it anything try to stump it yeah go
1:20:39 play trivia with it tell it you want to
1:20:41 play Trivial Pursuit and have it ask you
1:20:42 a bunch of questions
1:20:44 um bing.com is Microsoft search engine
1:20:46 but if you click on the chat button
1:20:47 that's also chat gbt it's gpt4 and then
1:20:51 po.com is another place where you can
1:20:52 play with six different
1:20:54 um chat engines
1:20:57 um all of them are basically just
1:20:58 different ways to play with chat EBT
1:21:00 that I think are worth playing with I'm
1:21:01 not affiliated with any of those if you
1:21:03 have questions about AI pop them into
1:21:06 the into the comments below I'm very
1:21:08 behind on my questions and Tick Tock
1:21:10 just lost my place so I'm trying to
1:21:12 figure out where I was
1:21:13 uh
1:21:15 um oh okay I found out where it was okay
1:21:17 here we go so do you think chat GPT
1:21:19 counter will be able to reason within
1:21:21 the next one to five years five years
1:21:24 certainly one year probably not
1:21:29 so so what's interesting about
1:21:32 gpt4 right now is it's good enough that
1:21:35 depending on the thing that you ask it
1:21:38 it seems like it's reasoning but when
1:21:40 you actually understand how this [ __ ]
1:21:42 actually works it's not it's just a
1:21:44 probability option it's just literally
1:21:47 like vomiting out words it's got no
1:21:49 awareness of the words that's vomiting
1:21:51 out
1:21:52 um
1:21:53 I know that there are companies working
1:21:55 on that right now that it actually has
1:21:57 an understanding of what it's vomiting
1:21:59 out and an anthropic in particular
1:22:02 they've got this thing called the
1:22:04 Constitutional learning model so they're
1:22:06 very much about alignment and and
1:22:10 and the tool getting better and better
1:22:12 at understanding what it's putting out
1:22:14 there and
1:22:15 um keeping it in line with a set of
1:22:17 rules that that they set right or that
1:22:20 maybe you set
1:22:21 um so I think that there there are
1:22:24 companies working on that problem right
1:22:26 now and I think I my guess is two to
1:22:29 three years we'll start to see systems
1:22:31 where the answer that it gives you it's
1:22:34 much more aware of the meaning of what
1:22:36 it gave you and it can do things like
1:22:38 did I give you any [ __ ] citations
1:22:40 and if I did
1:22:42 iterate until
1:22:44 I give you actual real citations that
1:22:46 actually support the argument
1:22:48 there was a really interesting
1:22:50 um
1:22:51 story of a doctor I was watching this
1:22:54 doctor on Tick Tock and and he said
1:22:56 um
1:22:58 he heard that Chachi PT was amazing he
1:23:00 said I want to you know let me try
1:23:03 and so he gave it a patient case and it
1:23:05 correctly diagnosed that patient case
1:23:07 and he was like huh that's interesting
1:23:09 and then you give it a hard patient case
1:23:11 and it correctly diagnosed that patient
1:23:14 case and he was like [ __ ]
1:23:16 and then he gave it the hardest case
1:23:18 that he had ever
1:23:20 had
1:23:22 and it correctly diagnosed that patient
1:23:25 it was a really obscure set of right and
1:23:27 and it correctly diagnosed that and he
1:23:29 goes this is [ __ ] amazing he goes he
1:23:31 has chat CPT you know give me the the
1:23:35 um
1:23:36 like site
1:23:38 the the cases that got that led you to
1:23:41 that diagnosis
1:23:42 and it wrote a bunch of cases and they
1:23:45 were all [ __ ] they were all made up
1:23:47 so so so it actually doesn't like it can
1:23:51 do things remarkably well but then it
1:23:54 might not know how it did it or how it
1:23:55 got there it will confidently look like
1:23:58 it's doing that but so it's wild but I
1:24:01 think I think the the point that you're
1:24:03 talking about its ability to reason
1:24:06 um you know two things happen one is the
1:24:09 [ __ ] just gets way better but the [ __ ]
1:24:12 gets way scarier right because if it
1:24:14 starts reasoning and says well I don't
1:24:18 really like that Kyle doesn't want to do
1:24:20 this thing that
1:24:21 I want to do and I'm I'm reasoning that
1:24:24 him not wanting to do it is
1:24:27 not as valuable as my wanting to do it
1:24:30 so [ __ ] it I'm gonna go do it anyway
1:24:31 right so if you don't if you don't have
1:24:34 some rules built into these tools their
1:24:36 ability to go Rogue and just go do [ __ ]
1:24:39 not on your behalf
1:24:41 um increases pretty dramatically so
1:24:43 that's why the regulation conversation
1:24:45 is an important one I like talking to
1:24:47 chat GPT and email form sometimes yeah
1:24:51 um I also I don't know if anyone else
1:24:53 experiences that that this but I find
1:24:56 myself saying please and thank you a lot
1:24:58 you know hey chat gbt awesome job on
1:25:02 that last paragraph you wrote could you
1:25:04 do me a sweet little favor and polish it
1:25:07 up a bit
1:25:08 um because there's something about the
1:25:10 nature of how it was programmed where it
1:25:12 it wants to please you it feels like a
1:25:15 human interaction it feels very much
1:25:17 like I'm interacting with a co-worker so
1:25:19 anyway
1:25:21 um
1:25:22 yeah so if you've got questions about AI
1:25:24 pop them down below I'm happy to answer
1:25:26 questions about whatever my background
1:25:28 or what I do or I'll just keep answering
1:25:31 them do what I can I'm gonna probably
1:25:32 put in another I don't know 15 20
1:25:36 minutes half an hour something like that
1:25:37 I've been at it for 90 minutes now
1:25:40 so get a little crispy but you know
1:25:42 that's why I'm here
1:25:44 all right I just want to say thanks
1:25:46 reinforcing the details and how to use
1:25:48 the digital dishwasher to your
1:25:51 I don't know what that is what is your
1:25:53 Discord that you showed before so my
1:25:55 Discord is uh part of the AI salon so if
1:26:00 you go to the salon.ai that's a link
1:26:02 tree that goes to my
1:26:05 um
1:26:06 the salon we meet bi-weekly in person we
1:26:09 have we do it virtually and in person
1:26:11 here in Denver and then we've got the
1:26:14 Discord where you can go geek out and
1:26:15 play with mid journey and ask questions
1:26:17 and like hang out in the community is
1:26:20 there one source for AI and chat gbt
1:26:22 good question that's why I have hold
1:26:28 these things up here so there's not one
1:26:31 source
1:26:32 um there are a lot of scam apps that are
1:26:35 out there to steal your [ __ ] money
1:26:36 and that's why I put this here so
1:26:40 chat.openai.com that's the official
1:26:42 website of chatgpt
1:26:45 so like even on my phone I just have a
1:26:48 bookmark to that rather than there is an
1:26:50 official chat gbt app but there's so
1:26:53 many
1:26:54 scam apps out there that just be really
1:26:56 careful just go there that's why I have
1:26:59 this here
1:27:00 bing.com is also gpt4 if you click on
1:27:03 the chat button at bing.com
1:27:06 that's gpt4 connected to the internet
1:27:09 for free and then poe.com gives you
1:27:13 access to their their large language
1:27:16 models
1:27:18 uh two from openai
1:27:21 GPT 3.5 gpt4 and then three models from
1:27:25 anthropic which is a competitor to open
1:27:28 AI it's not quite as good yet but but uh
1:27:32 they have a hundred thousand token model
1:27:34 there that's worth putting there so the
1:27:36 reason I put these three sites there
1:27:38 they're all sort of variations of the
1:27:40 same thing but those are kind of
1:27:42 official places to go play with this
1:27:44 stuff because there's so much [ __ ] out
1:27:46 there that's scammy because
1:27:49 everyone wants to play with AI and
1:27:51 people are like I know I could get
1:27:53 people to give me 10 bucks a month for
1:27:55 an app that looks just like chat jpk
1:27:58 you too can make money with chat gbt by
1:28:01 scamming people about chat GPT
1:28:07 um writing an ad-lib
1:28:10 um and asking chatgpt to compress it
1:28:12 while maintaining info can save tokens
1:28:14 writing in ad-lib and asking chachibi to
1:28:17 compress it while maintaining info yeah
1:28:20 so if you've got really long stuff that
1:28:22 you want to chat against tell chatgpt to
1:28:25 compress it and and it will it will
1:28:27 extend the length of your conversation
1:28:29 before it goes off the rails
1:28:32 what's the best paid app to download for
1:28:34 the phone so okay
1:28:37 I'm gonna so this is an important one
1:28:40 just because the phone thing
1:28:42 like is
1:28:45 it's it's essentially impossible to find
1:28:48 the official chat GPT app in the app
1:28:51 store it's not but there's so many scams
1:28:54 that it's it's really dangerous so
1:28:55 here's what I would do
1:28:58 I would go to openai.com
1:29:02 if you want the official chat gbt app
1:29:05 and then go to the products the product
1:29:08 tab
1:29:10 and go to chatgpt
1:29:13 and then in the lower right hand corner
1:29:16 here
1:29:17 right here is the link to the App Store
1:29:22 so that's the official chat GPT app
1:29:27 so the official app by openai the
1:29:30 company is open AI the that's what the
1:29:33 logo looks like like here let's go
1:29:36 I'm gonna go chat GPT
1:29:41 look at all these things see how they
1:29:43 they've all copied the logo
1:29:45 Genie chat with chat with ask AI chat AI
1:29:50 chat see they're they're all genius AI
1:29:53 chat AI chat bot so
1:29:56 all of this stuff is crap now here's
1:30:00 Bing so Bing is another thing that
1:30:02 another app that you could use so again
1:30:04 this is from Microsoft Corporation so
1:30:07 this is Bing chat with AI and gpt4
1:30:10 so you could either use Bing or you can
1:30:12 use the official openai Chat gbt thing
1:30:15 again I would go to openai.com products
1:30:19 chatgpt and then click to the official
1:30:22 app or just be really careful the
1:30:25 official app by open AI openai is the
1:30:27 company
1:30:30 because you know them scammers is out to
1:30:34 get you they'll do it too they'll take
1:30:36 your money and not even give a [ __ ]
1:30:38 that's just how they roll that's how
1:30:40 they roll boy
1:30:42 all right
1:30:45 do you have to say please to chat gbt
1:30:48 you don't but if you're like me you fall
1:30:50 into the habit I thank it I say please I
1:30:55 I tell it good job on that last one and
1:30:59 you know there's something about when
1:31:00 you tell it it did a good job you're
1:31:03 actually reinforcing that what it gave
1:31:05 you was in in the right direction do
1:31:07 different AIS respond very differently
1:31:09 holy [ __ ] yes
1:31:12 um go to so go to chat EPT so go here
1:31:17 and have it
1:31:19 um
1:31:20 ask it to write you like a rhyming poem
1:31:22 or a wrap like have it write you a wrap
1:31:24 from Eminem and then go to
1:31:26 bard.google.com go to googlebard and and
1:31:29 type in the exact same prompt and just
1:31:31 compare the answers
1:31:32 it's just they're stupidly different
1:31:38 um I love it when AI fights with other
1:31:40 AI
1:31:42 uh I appreciate your talks thank you
1:31:45 very much do you know of a way to have
1:31:47 chat gbt read PDFs yes they there are a
1:31:51 pile of
1:31:52 um
1:31:53 what are they called plugins
1:31:58 so let me go turn off all this [ __ ] and
1:32:00 let's go look
1:32:02 one one that I use is called ask your
1:32:06 PDF
1:32:07 so we'll turn that on we'll turn off the
1:32:09 story the stories thing that stories
1:32:11 thing is really cool
1:32:14 um so ask your PDF and then and there's
1:32:17 one called chat with PDF and in fact
1:32:19 let's go to the store I bet I bet
1:32:20 there's 30 of them now
1:32:22 so I'm going to type in PDF
1:32:25 so I'm now at the plug-in store there
1:32:27 are four pages so there's 16 of them
1:32:31 access to PDF and docs export to PDF
1:32:35 mixer box summarize anything ask your
1:32:38 PDF chat with PDF chat OCR that's
1:32:42 interesting
1:32:43 so you can OCR images easy resume
1:32:49 all right so
1:32:52 so now I need to go find a PDF let's see
1:32:55 um we'll do oh I know where we'll go
1:32:59 um
1:33:00 archive wait hey
1:33:11 [Applause]
1:33:14 so go to Archive so this is a this is a
1:33:17 site full of
1:33:18 um
1:33:21 full of white papers AI
1:33:24 um image
1:33:27 processing
1:33:34 all right here's one we'll just go we'll
1:33:36 just go grab this URL so
1:33:39 this is a white paper
1:33:42 I'm gonna do copy
1:33:46 it's I don't know a big pile of pages
1:33:50 and it says assist GPT General
1:33:53 multimodal assistant so we'll go back to
1:33:55 chat GPT so I now have the plug-in
1:34:00 um ask your PDF there
1:34:02 and now you can go
1:34:05 um can you
1:34:08 please
1:34:11 summarize
1:34:13 this in 10 short bullet points
1:34:19 and then I just paste in the URL
1:34:22 and it will fire off that plug-in
1:34:27 which you did
1:34:29 and now it is going off
1:34:33 and it looks like it
1:34:35 it just pulled back the the general
1:34:37 summary there
1:34:40 but it says it talks about a multimodal
1:34:42 AI assistant the assistant integrates
1:34:45 large language models with various tools
1:34:47 designed to right so so it went and re
1:34:49 so if you have any specific questions
1:34:52 about the content of the document so
1:34:54 let's say
1:34:56 um
1:34:58 um what is an example of a use case
1:35:04 talked about
1:35:08 in this document
1:35:16 it should fire it off again and it might
1:35:19 now it might dig a Little Deeper because
1:35:20 I just asked for the summary
1:35:23 yeah it dug a little bit deeper here
1:35:29 provides several use cases question
1:35:32 what I what I am on the desk
1:35:35 could help with a cold
1:35:38 question what activity does the cat
1:35:41 appear most likely to do so it's giving
1:35:43 me very very specific use cases that it
1:35:46 found within the document so
1:35:49 there you go
1:35:51 PDF O Rama
1:35:53 I mean and I mean that's remarkable like
1:35:56 like some of those scientific white
1:35:59 papers like are really really
1:36:01 interesting but like I have ADD I can't
1:36:04 [ __ ] read a 20 Page
1:36:08 um scientific white paper but
1:36:10 if I can have this thing like give me 10
1:36:12 bullet points so I can get my head
1:36:14 around the concept I can at least figure
1:36:15 out is that something I want to dig
1:36:17 deeper on or not right so so for some
1:36:20 for someone with with a brain that
1:36:22 functions like mine does neuro spicy as
1:36:25 they say
1:36:26 um yeah that that sort of shit's like a
1:36:28 little miracle it really is very cool do
1:36:31 you use iOS shortcuts uh with interface
1:36:34 with chat GPT
1:36:36 um
1:36:37 I did it once and then it broke and then
1:36:40 I haven't reset it up you can you can
1:36:42 use shortcuts you can there's ways you
1:36:44 can kind of hack Siri to talk to chat
1:36:47 gbt but you have to set it up as a
1:36:49 shortcut and I just I just haven't done
1:36:50 it yet
1:36:51 um
1:36:54 in relatively short order like probably
1:36:58 within I don't know
1:37:00 maybe in the fall
1:37:02 Apple will come out with some sort of
1:37:03 voice interface to some sort of large
1:37:05 language model
1:37:07 than that but I'm just I'm just kind of
1:37:09 waiting right now but yeah you can you
1:37:10 can there's there's YouTube videos on
1:37:12 people that like you can talk with chat
1:37:14 GPT and
1:37:16 with your voice with Siri
1:37:19 so if you are not like me with an 80d
1:37:23 brain
1:37:24 and you actually want to take the hour
1:37:26 and a half to go figure that [ __ ] out
1:37:28 yes you can there's a little moth a
1:37:30 little son of a [ __ ] all right
1:37:35 what's the name of that site again I'm
1:37:37 not sure which one you're talking about
1:37:39 but I will if you
1:37:41 tell me I'll I'll go back
1:37:44 the power grid the AI would just shut
1:37:46 off a few cities yeah that's one of the
1:37:48 dangers if it if it got access I mean
1:37:51 and you know those power companies are
1:37:52 probably running on I mean hopefully
1:37:54 post 911 they've got their cyber
1:37:57 security [ __ ] together for the power
1:37:59 grid but
1:38:01 hey I don't know which which power
1:38:04 companies are still running Windows 95
1:38:07 without security patches on them
1:38:08 probably more than we are comfortable
1:38:10 with
1:38:14 oh it's behind his head yeah there's
1:38:17 stuff behind my head
1:38:21 I'm like Carol Merrill
1:38:23 all right
1:38:26 um
1:38:31 all right I love how creative you are
1:38:34 thank you very much dude skip typing
1:38:36 class dude come up with more creative
1:38:39 trolls than that
1:38:42 that's so lame
1:38:44 I can't
1:38:45 yeah you got me uh you know I wasn't
1:38:49 great at that typing
1:38:51 good Lord you know how hard it is like
1:38:53 I've got my computer is up at like eye
1:38:56 level my laptop's at eye level I'm like
1:38:58 typing like this for one thing
1:39:02 plus I can't type plus I'm typing live
1:39:05 so I'm sure you never make mistakes
1:39:07 right
1:39:11 ah good Lord what app is that I'm not
1:39:13 sure which one I was using
1:39:15 um I gave it a list of ingredients I had
1:39:17 for dinner and then had it give me a
1:39:19 hundred options of different Cuisines
1:39:21 [ __ ] awesome and then have it give
1:39:23 you a shopping list and then have it
1:39:24 organize the shopping list buy section
1:39:27 of the grocery store that one for me is
1:39:29 the like the biggest one like give me a
1:39:31 big long list of [ __ ] to go by and then
1:39:34 organize it by by section of the grocery
1:39:36 store because that's always my pet peeve
1:39:38 is like I'm always at the end of the
1:39:40 grocery store and then there's that one
1:39:42 item that was all the way back at the
1:39:43 beginning you're like [ __ ]
1:39:47 never again
1:39:49 can you make a Markov blanket with
1:39:52 Wolfram
1:39:53 probably I don't know what a Markov
1:39:56 blanket is but let's go ask chat tvt let
1:40:00 me make an ass of myself so not only
1:40:01 will I make an ass of myself typing I'm
1:40:04 going to make an ass of myself because I
1:40:05 have a degree in acting not in
1:40:07 mathematics
1:40:09 dear Chachi BT can you tell an actor
1:40:15 what a Markov
1:40:20 blanket is
1:40:22 [Applause]
1:40:24 certainly it's a it's a field of
1:40:27 probabilistic graphical models a Markov
1:40:29 blanket refers to a set of variables
1:40:31 when known provide a complete
1:40:33 probabilistic determination of the
1:40:34 target variable in simpler terms the
1:40:38 Markov blanket of variable contains all
1:40:40 the other variables directly influenced
1:40:42 or are directly influenced by that
1:40:44 variable while excluding any other
1:40:45 variable seems perfectly logical to me
1:40:48 I'm going to say
1:40:50 can you explain it
1:40:54 so that even an actor can under
1:40:59 stand it
1:41:01 [Laughter]
1:41:01 [Music]
1:41:06 come on oh now it's thinking about it
1:41:08 you see we stumped it
1:41:10 but what I'm going to do here is I'm
1:41:12 gonna I'm gonna I'll activate the
1:41:14 Wolfram Alpha plug-in and then we'll see
1:41:18 what we can do come on
1:41:22 stop generating regenerate response
1:41:25 absolutely let's break it down in a way
1:41:28 that's easy for actors to grasp imagine
1:41:31 you're playing a character in a scene
1:41:33 your characters is like the central
1:41:35 person in a web of connections these
1:41:37 connections involve other characters the
1:41:40 environment the events happening around
1:41:42 you the archive blanket is like a
1:41:43 protective boundary around you including
1:41:46 only the things that directly affect
1:41:47 your character and are affected by your
1:41:50 character that's a perfect explanation
1:41:52 for an actor because all an actor really
1:41:54 carries cares about is themselves so so
1:41:57 this puts it in completely their context
1:41:59 okay great so
1:42:01 now let's go to gpt4 plugins we're going
1:42:04 to turn off ask your PDF we're going to
1:42:06 turn on wolf from Alpha
1:42:09 and then I know so little about this
1:42:11 that I don't even know how to write this
1:42:12 prompt in a way that'll get us an answer
1:42:14 but I'm going to say can you please
1:42:18 um
1:42:19 give me and
1:42:22 example set of data
1:42:26 for a Markov blanket
1:42:30 and then
1:42:32 um I don't know
1:42:34 uh fill in the formula
1:42:39 to
1:42:42 demonstrate that might not even be the
1:42:45 way to ask that but who gives a [ __ ]
1:42:47 let's see if it'll do it sure let's
1:42:49 start out by defining what a Markov
1:42:51 blanket is
1:42:52 in a Bayesian Network it's a node with a
1:42:55 set of nodes a is a parent of B and C B
1:42:58 is a parent of d c is a parent of d and
1:43:00 e in the case with markup
1:43:02 now let's say the following
1:43:03 probabilities
1:43:09 all right so I guess it's now making our
1:43:12 data
1:43:15 given these possibilities we calculate
1:43:17 the conditional probability of D in this
1:43:19 Markov blanket b c and e using the
1:43:21 following formula
1:43:22 so then we simplify that
1:43:25 so that didn't need to go use Wolfram
1:43:27 Alpha to do that so it just did that so
1:43:29 there might be a way to ask that
1:43:30 question that it requires Wolfram to do
1:43:33 something more but that's that's how you
1:43:35 would do it you would you would go in
1:43:36 here you would turn on you would enable
1:43:38 the Wolfram Alpha plug-in and then you
1:43:41 would ask it a question
1:43:43 um let me just try please calculate
1:43:48 calculate the
1:43:50 example
1:43:54 see if this fires off the wolf from
1:43:56 alpha yeah there we go
1:43:58 so that fired off the wolf from alpha
1:44:00 plug-in
1:44:02 it typed in the code it got a response
1:44:06 and the probability of D given the
1:44:08 Markov blanket is approximately 0.964
1:44:11 which that that's obvious to any actor
1:44:14 that would have looked at this
1:44:15 particular set of data I mean
1:44:19 fairly obvious to me
1:44:20 right
1:44:22 I mean given a Markov blanket
1:44:25 point point what is it 964.
1:44:29 right duh
1:44:33 who owns the illustrations in the
1:44:35 stories in that case it's probably that
1:44:37 website but I I would imagine that
1:44:40 website
1:44:41 Storybird
1:44:44 about
1:44:46 bless you
1:44:50 I'm good
1:44:53 uh uh
1:44:55 this is a brand new company so I bet
1:44:58 they don't have their legal [ __ ]
1:44:59 together
1:45:02 um You probably own it is my guess
1:45:07 but I don't know
1:45:09 uh oh my God the comedy I know it's bad
1:45:12 right sorry about that
1:45:13 [Laughter]
1:45:20 all right I got lost The Tick Tock just
1:45:23 bounced me to the bottom again do I own
1:45:25 any AI coin do you like any AI coins
1:45:27 well
1:45:31 um I I so I spent like a year deep
1:45:35 diving into
1:45:36 um
1:45:37 blockchain and crypto and nfts and
1:45:41 um
1:45:42 that's a [ __ ] [ __ ] show over there
1:45:44 I'm having much more fun in the AI stuff
1:45:46 uh I think the the the one that Sam
1:45:49 Altman is putting together World coin is
1:45:51 really interesting if not a little scary
1:45:54 uh World coin by the way is if you don't
1:45:57 know what worldcoin is
1:46:00 so so Sam Altman they've developed this
1:46:02 orb kind of thing it's like the size of
1:46:05 like a baseball or a softball and it's
1:46:07 got a hole in it and you put it over
1:46:09 your eye and it scans your iris and it
1:46:12 verifies that you're human and if it
1:46:14 verifies that you're human it sends you
1:46:16 a bunch of world coin tokens so that so
1:46:19 that you're gonna have a currency that
1:46:20 is a currency that verifies that you're
1:46:22 human now is that creepy as [ __ ]
1:46:25 absolutely
1:46:26 is it also going to be really really
1:46:29 really
1:46:30 incredibly important that we have
1:46:33 tokenized verification of human beings
1:46:37 and authenticity of news pieces and
1:46:41 content and videos yes
1:46:45 why is it going to be important because
1:46:47 watch the 2024 presidential election in
1:46:50 the United States and tell me in 2025 if
1:46:53 you don't think we need that [ __ ] so I
1:46:55 think it's really smart that Sam Altman
1:46:57 is putting that together whether or not
1:46:59 that's the right solution I don't know
1:47:01 but but we need we need some
1:47:03 authenticity verification
1:47:07 Pronto we don't know it yet but the 2024
1:47:11 election is going to make it painfully
1:47:13 obvious that uh
1:47:15 the 2016 and 2020 elections were like
1:47:18 Child's Play when it came to
1:47:19 misinformation
1:47:22 I just know we're gonna see this what do
1:47:25 you mean he's that that the the
1:47:27 candidate was French Chris kissing a
1:47:30 gorilla in the Cincinnati Zoo really
1:47:33 yeah oh there's video of it really ah Is
1:47:37 that real
1:47:39 that's coming two nights before the
1:47:41 election
1:47:43 you watch
1:47:47 um I lost my place here I'm trying to go
1:47:49 back and figure out where I was
1:47:57 all right AI weather forecasters I was
1:47:59 here already
1:48:02 oh I'm I'm way way behind hang on hang
1:48:06 on wait for me what'd you do tonight I
1:48:08 watched some old guy scrolling through
1:48:09 Tick Tock it's pretty interesting
1:48:13 all right AI fights with other AIS we
1:48:16 did that one
1:48:18 the power grid it would just shut it off
1:48:20 we did that one hold I'm catching up
1:48:22 it's gonna be so exciting when I finally
1:48:24 get there it's gonna be so good you're
1:48:26 gonna be like
1:48:27 that was worth the wait
1:48:29 I gave it a list of ingredients we
1:48:30 talked about that one oh sorry about
1:48:32 this oh my God your comedy okay that was
1:48:34 the one I saw where it jumped me to the
1:48:36 bottom okay Poe keeps messing up my
1:48:38 calculations see here's the deal
1:48:40 um Poe I don't know which model you're
1:48:42 using for Poe
1:48:44 um I would use gpt4 so I would use Bing
1:48:47 or chat GPT Bing or Chachi BT and pay
1:48:51 for gpt4 or use gpt4po.com
1:48:56 um calculations with 3.5 and with Claude
1:48:59 instant
1:49:00 you you are just not good you might try
1:49:03 Bard bard.google.com for calculations
1:49:06 I've heard it's better at calculations
1:49:08 than
1:49:10 um than it is at language it sucks at
1:49:12 language
1:49:14 do I use any iOS shortcuts that
1:49:16 interface with chat gbt
1:49:19 um
1:49:20 I don't but you can
1:49:22 just go watch YouTube videos on that
1:49:24 really enjoy plugging my screenplay into
1:49:26 Cloud 100K today it was very
1:49:28 enlightening oh that's super cool yeah I
1:49:30 haven't done that with any of my
1:49:31 screenplays
1:49:32 um I did I took a partial screenplay
1:49:34 with chat gbt and I had it finish the
1:49:38 outline for me I'd written two scenes in
1:49:40 like three quarters of an outline I had
1:49:42 to finish the outline and then I had it
1:49:45 write a third scene which the first
1:49:46 crack at it sucked but then after like
1:49:49 three iterations it got to somewhere
1:49:53 all right how do you get plug-in how do
1:49:55 you get the plug-in Marketplace you have
1:49:57 to pay for it you go to chat.openai.com
1:49:59 you pay for chatgpt plus and then in
1:50:03 your settings you enable the beta
1:50:05 features and turn on plugins that's how
1:50:07 you get there
1:50:10 um exactly like I did algebra in high
1:50:12 school but couldn't explain how I did it
1:50:14 yep exactly uh I see please please and
1:50:17 thank you also yeah you gotta be nice to
1:50:19 your robots
1:50:20 uh what's the basic thing chat GPT does
1:50:23 that you really like
1:50:25 great question
1:50:27 um
1:50:29 it has erased writer's block
1:50:34 one thousand percent all I have to have
1:50:37 is kind of a vapor of an idea
1:50:40 and within 30 seconds I've got it
1:50:43 realized in some form whether I know if
1:50:45 it's a good idea or not like I don't get
1:50:48 stuck in that well I'd have to open up
1:50:51 word and then I'd have to write an
1:50:52 outline and then I went to bed like if
1:50:56 you ever done anything creative
1:50:59 um
1:51:00 sorry moth
1:51:03 listen
1:51:07 er uh
1:51:11 uh what are we talking about
1:51:14 oh what do I what does it do that I
1:51:16 really like like
1:51:18 it's just it's just I don't know it
1:51:23 it's it's a remarkable writing partner
1:51:27 that's it
1:51:28 Microsoft calls it co-pilot that's the
1:51:31 right branding for me I'm the pilot it's
1:51:33 the co-pilot but it's like having the
1:51:35 most capable co-pilot in the history of
1:51:37 co-pilots it knows [ __ ] everything
1:51:39 and if if you get clever at talking to
1:51:44 it and collaborating with it you can get
1:51:46 an idea out of your head and into some
1:51:48 sort of usable form in minutes every
1:51:51 single time zero writer's block it's
1:51:54 [ __ ] Magic
1:51:56 can AI replace programmers yes and no I
1:51:59 think Engineers are going to be in
1:52:02 demand for a long long time because holy
1:52:04 [ __ ] everything's about to change
1:52:07 and
1:52:11 engineers
1:52:12 that are pair programming with AI are
1:52:15 going to be in much higher demand than
1:52:17 ones that aren't and pretty soon if
1:52:19 you're not doing AI augmented
1:52:22 programming
1:52:25 um I think you're going to be left
1:52:26 behind
1:52:27 a good site for finding good prompt for
1:52:29 chat gbt
1:52:31 prompts that chat that's why it's here
1:52:34 that's why it's here I put it on the
1:52:36 thing behind my head my big fat head
1:52:39 covers this up but prompts.chat is where
1:52:41 you go
1:52:42 all right and there's lots of other ones
1:52:45 but that one's just one that it explains
1:52:47 it well and gives you a bunch of good
1:52:48 examples when degenerative AI kill
1:52:50 traditional web search If Ever I think
1:52:52 sooner than I don't think it's if well
1:52:55 I don't think traditional web search
1:52:57 will go away but I think the necessity
1:53:00 of web search
1:53:02 as a core function of how we find [ __ ]
1:53:06 on the internet
1:53:08 the importance of it shrinks
1:53:10 dramatically which is why Google's
1:53:12 freaking out right now
1:53:21 have you ever used it wait I have used
1:53:24 it for so much already it's pretty
1:53:25 amazing I only use the free version I'm
1:53:27 telling you
1:53:28 if you just use the free version of chat
1:53:31 EPT chat.openai.com just use the free
1:53:34 version chat EBD 3.5
1:53:37 I like I sincerely feel this way I could
1:53:42 spend the next decade
1:53:45 trying to figure out all of the things
1:53:47 that that makes possible it's that
1:53:49 powerful
1:53:51 it's [ __ ] insane
1:53:54 and it's like it's Simplicity belies
1:53:56 what's sitting beneath it it's basically
1:53:58 like if if you understand what the base
1:54:02 model of Chachi PT is they trained it on
1:54:05 all of the public internet so if you
1:54:07 think about for the last 30 35 years
1:54:10 people have been putting [ __ ] on the
1:54:11 internet right
1:54:13 all the [ __ ] on Wikipedia all the blogs
1:54:15 all the websites all the university
1:54:18 paper archives just all the [ __ ] you can
1:54:21 imagine and then Sam Altman recognized
1:54:24 we've got to have good data so they like
1:54:25 went and dug in the corners of the
1:54:28 internet to build these core models it
1:54:30 like knows all that [ __ ]
1:54:32 and then you can just ask it and it will
1:54:34 tell you it's like
1:54:37 what can you do with that I don't know
1:54:38 everything
1:54:40 does IT program yes will it replace my
1:54:42 job yes
1:54:46 it won't replace your job but someone
1:54:48 using it will
1:54:50 that's from Paul rateser
1:54:52 open AI just created an IOS app for chat
1:54:55 EBT yes but be very careful to get the
1:54:57 real one so go to openai and get the
1:54:59 link from there and just make sure it's
1:55:01 the official one all the other ones are
1:55:03 rip-off [ __ ]
1:55:05 can an AI replace programmers answer
1:55:08 that have you played with perplexity AI
1:55:10 I have played with it a little bit I
1:55:12 can't quite figure it out it has some
1:55:15 chat gbt [ __ ] in it but then sometimes
1:55:17 when I ask good stuff I just I keep
1:55:19 going back to this one
1:55:22 these [ __ ] invented it this is by far
1:55:25 the best like gpt4 is by far
1:55:29 you know Head and Shoulders above
1:55:31 everything else right now why I don't
1:55:33 know it just is When someone knocks that
1:55:37 [ __ ] off the mountain I will change that
1:55:39 Top Line but I just keep going back to
1:55:41 that because there's just nothing like
1:55:43 it
1:55:46 what's the other site you talked about
1:55:48 earlier no it had playground at the end
1:55:50 okay so that's um
1:55:53 it's also openai.com but if you go to
1:55:56 platform.openai.com playground that's
1:55:59 the developer portal for chatgpt's API
1:56:03 so you can change models you can there's
1:56:06 all the documentation in there
1:56:09 platform dot openai.com playground
1:56:16 can you get GPT
1:56:18 version four from open for free no but
1:56:22 you get it here from free bing.com if
1:56:25 you go there it's a search engine and if
1:56:28 you click on the chat button that's gpt4
1:56:33 gpt4 for free
1:56:37 FYI there must be some configuration
1:56:39 difference in Bing chat or a different
1:56:43 difference yes
1:56:45 Bing chat even though it's gpt4 is very
1:56:49 different
1:56:50 than chat gpt's gpt4
1:56:53 and it's not as good there's a lot of
1:56:55 reasons for it
1:56:58 it is gpt4 so it's better than gbd 3.5
1:57:03 but it's not as good as the gpt4 you pay
1:57:06 for here and I don't think there's a
1:57:08 free way to get it up there
1:57:11 why do you pay for chat gpt4 if it's
1:57:14 free with Bing because it's better
1:57:16 and and and you also get access to the
1:57:19 plugins
1:57:20 with gpt4 and when code interpreter
1:57:23 comes out I I also get access to that so
1:57:25 I pay for this for all the Early Access
1:57:27 to their [ __ ]
1:57:29 code interpreter is going to melt some
1:57:31 faces man go to YouTube go to YouTube
1:57:34 search for code interpreter Chachi BT
1:57:37 and if you've done anything ever in your
1:57:39 life with data or spreadsheets
1:57:41 prepare to have your face just drip off
1:57:43 your skull
1:57:46 I use AI with my kids crazy questions
1:57:49 and it breaks it down for them until
1:57:51 they get it yep exactly by the way if
1:57:54 you if if you're doing this with your
1:57:56 kids
1:57:56 um Khan Academy
1:57:59 um which is the you know Alternative
1:58:01 Learning thing the flip the classroom
1:58:02 thing
1:58:04 um they launched a thing month six weeks
1:58:06 ago called conmigo which is a GPT for
1:58:10 Education bot sitting on top of Khan
1:58:12 Academy's all of their stuff and it it's
1:58:15 remarkable they spent like seven months
1:58:17 fine-tuning that bot to be a really good
1:58:20 tutor for kids
1:58:22 and and adults quite honestly
1:58:25 because I've used both and I get better
1:58:27 coding responses from Bing chat oh
1:58:29 that's cool
1:58:30 oh yeah if you get so with coding
1:58:34 see if you're doing coding I would say
1:58:36 GitHub
1:58:38 um GitHub co-pilot I would play with
1:58:41 Bard I would play with Bing
1:58:43 and compare that to gpt4 right because
1:58:46 with coding it's very possible that one
1:58:49 of those other systems is better than
1:58:51 gpt4 with with word stuff with language
1:58:54 stuff gpt4 nothing touches it but with
1:58:57 coding there there might be other things
1:58:58 that are better so that's that's a
1:59:00 really good point Winston thank you for
1:59:03 that
1:59:06 all right I got 10 minutes and then I'm
1:59:08 gonna get yelled at
1:59:10 because because I live in a household
1:59:13 with people that love me and want to see
1:59:15 me and don't don't want me on The Tick
1:59:17 Tock all the time
1:59:19 do you love those Tick Tock people more
1:59:22 it's possible
1:59:29 all right this is my new favorite live
1:59:31 thank you love that
1:59:33 thank you for sharing all this I
1:59:34 appreciate that
1:59:36 what will happen to people who don't
1:59:38 know anything about computers okay
1:59:40 good point if you don't know anything
1:59:42 about computer listen you're here so you
1:59:45 know enough about computers to open your
1:59:47 phone and get on a ticky tacky so you
1:59:50 know more about computers than you may
1:59:52 be willing to admit
1:59:54 don't you think don't you think you do
1:59:57 so here's the deal
1:59:59 the inflection point that we have now
2:00:02 passed we we have we have passed our
2:00:04 Rubicon or we have passed we're into a
2:00:06 new phase
2:00:08 prior to this phase
2:00:11 if you wanted to use machine learning
2:00:13 and like you ever heard Google talk
2:00:16 about things like tensorflow and you're
2:00:18 like oh I'm gonna go check out this AI
2:00:20 stuff that sounds really cool and then
2:00:21 you go figure out what tensorflow is and
2:00:23 unless you have a computer science
2:00:24 degree you ain't doing [ __ ] with
2:00:26 tensorflow
2:00:28 um
2:00:29 and that's how it was for many many
2:00:31 years for decades
2:00:34 Chachi BT November 30th 2022 so this
2:00:38 isn't that long this is like seven
2:00:39 months six months ago
2:00:44 marks the beginning of a new era where
2:00:47 people who are not computer scientists
2:00:49 and researchers and big [ __ ]
2:00:51 programming engineering brainiacs can
2:00:54 use these tools
2:00:56 on your phone if you're on you're on
2:00:59 your phone using tikitaki
2:01:01 Point your phone browser either Chrome
2:01:04 or Safari just point it to that web
2:01:07 address and start talking to it like
2:01:09 it's a person asking it things like it's
2:01:12 a person like it's a really smart person
2:01:14 and ask it questions for things that are
2:01:18 personal to you so maybe you're into
2:01:19 gardening
2:01:21 ask it to come up with
2:01:23 um
2:01:24 what are the next five plants that I
2:01:27 should plant or hey I've got you know
2:01:29 this kind of infestation and what can I
2:01:32 do about it that doesn't use chemicals
2:01:34 like just start interacting with it
2:01:38 um
2:01:39 like you've got a like like you have a
2:01:42 friend who just happens to be the
2:01:44 smartest friend you've ever met you've
2:01:46 ever known like
2:01:49 don't worry about the fact that you
2:01:50 don't know computers
2:01:52 take what you're passionate about what
2:01:54 you know
2:01:55 and start solving problems using that
2:01:59 the free version of that
2:02:02 that skill
2:02:04 over the next two years is going to be
2:02:06 incredibly valuable
2:02:07 people that don't do that are going to
2:02:10 be at a massive disadvantage
2:02:12 so
2:02:14 just go play
2:02:16 just go play that's why that's why I do
2:02:19 this every night is I come here to
2:02:21 inspire people like you to just fire up
2:02:24 your browser point it at that and start
2:02:26 asking it to write you a poem have it
2:02:28 write you a song for your significant
2:02:30 other or a poem about your pet and then
2:02:32 just sit there with your mouth open and
2:02:34 go
2:02:35 it can do that yep
2:02:38 and then have it write you a dinner plan
2:02:41 for the week
2:02:44 and then say I've only got three
2:02:46 ingredients in my fridge what can I make
2:02:48 from this out of the cub
2:02:51 what other things do I need to buy
2:02:53 here's your list
2:02:56 a good site for finding good prompts
2:02:59 prompts.chat prompts dot chat chat dot
2:03:02 prompts and there's probably other ones
2:03:03 but that's that's just that's a document
2:03:05 that I know is valuable and not a waste
2:03:07 of your time
2:03:09 there are a lot of prompting sites that
2:03:10 are a waste of your time
2:03:14 um the people who don't know can still
2:03:16 learn quickly with I boys
2:03:18 I don't know what I boys oh AI Bots
2:03:22 you know the eye boys
2:03:25 you can learn quick with the I boys
2:03:27 where is that app found which app I
2:03:30 don't know which app
2:03:31 what's the down payment not sure what
2:03:33 that's about a good site for finding
2:03:35 good prompts we just showed that have
2:03:38 you fed it any Declassified FBI Files
2:03:40 like DB Cooper I haven't but I better be
2:03:43 good at that [ __ ]
2:03:45 all right I got six minutes then I gotta
2:03:47 go let's let's go let's go play with uh
2:03:50 with chat GPT connected to the internet
2:03:52 and see what we can find on DB Cooper it
2:03:55 may some of its safety stuff it might
2:03:57 call it a conspiracy theory and not let
2:03:59 us play
2:04:00 but okay so I'm at gpt4 I'm going with
2:04:03 browse with Bing so I'm like okay
2:04:06 um what is the latest theory on who DB
2:04:12 Cooper is and where the [ __ ] money is
2:04:17 it probably doesn't like the fact I
2:04:19 swear sweared swore Jesus
2:04:24 squared
2:04:25 I don't know it was on there and he
2:04:27 sweared with these bad words he was just
2:04:29 he talked he talked fast and he had a
2:04:32 lot of bad words and he he seemed like a
2:04:34 bit of a dum-dum but then he seemed like
2:04:36 he was kind of smart but he was more
2:04:38 like a dum-dum he sweared a lot he
2:04:41 sweared a lot
2:04:43 and I think he said he was drinking iced
2:04:46 coffee but I'm pretty sure it was a Long
2:04:47 Island Iced Tea
2:04:51 so what this is doing right now it's off
2:04:53 searching the internet I mean this this
2:04:56 is one of the cool things like when you
2:04:57 like flip these little things open
2:05:00 it went to whatever that site is then I
2:05:02 went to the guardian then I went to some
2:05:04 other site
2:05:06 of the Guardian
2:05:08 click failed
2:05:09 click failed so it's it's trying it's
2:05:12 trying and failing
2:05:15 searched Bing DB Cooper Ransom money
2:05:18 Discovery scientists uncover new
2:05:20 minuscule Clues on DB Cooper so it's
2:05:24 searching it's searching
2:05:26 how exciting is this
2:05:32 I don't know you see he seemed like a
2:05:35 dum-dum as of recent developments Eric
2:05:38 Euless a true crime investigator who
2:05:41 organizes an annual event called
2:05:42 coopercon has identified a new person of
2:05:45 interest Vince Peterson was an employee
2:05:48 at Crucible Steel
2:05:49 all right where's the money regarding
2:05:51 the money in 1980 a young boy found six
2:05:53 thousand dollars of the ransom money
2:05:54 while camping with his family on tenobar
2:05:57 a sandbar in the Columbia River
2:05:59 the only Ransom money ever recovered
2:06:01 from the DB Cooper case recently amateur
2:06:03 scientist Tom K had examined this money
2:06:06 and identified diatoms
2:06:09 tiny deposits of algae on it he believes
2:06:11 that the diet comes indicate that the
2:06:14 money ended up in the river months after
2:06:16 Cooper's jump contradicting the FBI's
2:06:19 original Theory
2:06:23 so anyway yeah it does that
2:06:26 it does it quite well now was all that
2:06:30 true I don't know
2:06:32 but you can like it includes sources
2:06:35 doesn't it include sources yeah it
2:06:36 includes sources so I can actually go
2:06:38 look at the sources where it pulled
2:06:40 those answers from
2:06:42 so pretty dang slick ain't it
2:06:46 I'll tell you what this here AI stuff
2:06:49 it's pretty magical
2:06:52 where is DB Cooper
2:06:56 uh what was that Twitter account
2:06:59 where you meet where you talk about AI
2:07:01 okay hold please
2:07:07 so this is the AI salon so that's a
2:07:10 community that I started we meet every
2:07:12 other week our next meeting is tomorrow
2:07:13 night and we have a Discord server you
2:07:16 can play on and then this if you go to
2:07:18 Twitter and go to Everyday AI news
2:07:21 that's my newsletter and we also do
2:07:23 Twitter threads on interesting stories
2:07:25 there so you can go learn [ __ ] there and
2:07:28 you can sign up for a newsletter there
2:07:29 it's all free
2:07:31 free at some point we're gonna monetize
2:07:33 it but right now just it's out of the
2:07:36 good of our hearts not really
2:07:41 um we'll monetize it at some point when
2:07:43 it makes sense but right now it's free
2:07:44 like I'm just like the reason this Tick
2:07:46 Tock Channel exists I'm just trying to
2:07:48 process what I'm learning and get it out
2:07:50 of my head and then Inspire other people
2:07:52 to do the same I think it's really
2:07:53 important if you learn this stuff if you
2:07:56 get curious about it like I'm
2:07:57 encouraging you to the other thing that
2:07:59 I would encourage you to do
2:08:01 is share what you're learning
2:08:04 get other people up to speed on this
2:08:06 because this is important this shit's
2:08:07 coming fast and a lot of people are
2:08:09 going to be blindsided by it and it's
2:08:11 going to be
2:08:12 like painful
2:08:14 and and I want there to be less pain
2:08:18 and more
2:08:19 um excitement because these are the most
2:08:22 powerful tools ever put in the hands of
2:08:25 mankind and we all get to play with them
2:08:27 like we all get to play with them so
2:08:29 it's pretty magical times but it's also
2:08:32 serious because it's it's coming fast
2:08:34 and it's going to disrupt a lot of jobs
2:08:35 so as you learn about this stuff get
2:08:38 generous with what you're learning share
2:08:40 what you learn do things for others use
2:08:43 chat epd to solve problems for other
2:08:45 people
2:08:46 um and then when they and then when they
2:08:48 say to you how did you do that
2:08:50 let me show you
2:08:54 you know
2:08:56 it's it is one of the best things in the
2:08:58 world when you get this moment
2:09:01 from someone when they see what this
2:09:03 [ __ ] can do
2:09:04 so with that listen I know I missed a
2:09:06 bunch of questions below I appreciate
2:09:07 y'all hanging out with me I gotta go I
2:09:08 got work tomorrow I got I got I got a
2:09:11 company to get to I gotta go CEO you
2:09:14 gotta go CEO tomorrow can't just be
2:09:15 hanging out talking on the tick tock all
2:09:18 right everybody peace out have a good
2:09:20 night I really appreciate your time