aiLL 6/28/23 - The Innovations Transforming Our Future, By Kyle Shannon

Description

Chapters

Transcript

0:01 foreign
0:15 [Applause]
0:24 with the AI learning lab welcome welcome
0:26 welcome welcome
0:28 happy Wednesday night
0:32 yeah
0:34 we'll get rolling here
0:37 who's that Professor Sonya hello
0:40 what's shaking
0:43 just getting started eight o'clock on
0:45 the button here on
0:48 Wednesday in Denver
0:53 how is everybody
0:59 not a Missy
1:01 what's that yield
1:03 What's that name
1:05 went away
1:08 welcome everyone my name is Kyle Shannon
1:10 this is the AI learning lab welcome
1:11 welcome welcome I was wondering when you
1:13 would start I know I got a little bit
1:15 late start I was watching the news and
1:17 fell asleep so I had a little cat nap
1:20 after dinner good evening good evening
1:22 good evening cheers I got some uh some
1:24 iced coffee to us
1:26 Stave off the nap
1:29 don't want to nap while I'm live
1:31 but you know that could be entertaining
1:33 you know someone screen records it sends
1:36 it to me hello hello
1:40 all right oh wow 100 people here holy
1:43 crap my name is Kyle Shannon this is the
1:45 AI learning lab I'll talk about these
1:47 URLs behind my head in a second here
1:49 um a little bit about my background I am
1:52 an entrepreneur and thanks for tapping
1:54 the screen I think that'll let people
1:56 know we're here I'm starting to figure
1:57 out how this thing just what
2:00 what the kids call this tick ticker
2:03 talker thing starting to figure out how
2:05 this ticker Tucker works tap the little
2:07 screen that'll give us some little
2:08 hearts hearts go up goes out people come
2:11 in they say hi we hang out
2:13 uh this is the AI learning lab uh I'm an
2:17 entrepreneur I'm the current uh
2:20 co-founder and CEO of a company called
2:22 storyvine which is an automated video
2:24 storytelling platform I am completely
2:27 rethinking
2:29 um the future of the product and the
2:31 value proposition thanks to AI so thanks
2:34 Ai and also AI scarier than [ __ ]
2:39 um I do swear occasionally so if you
2:41 don't like that I'd go now
2:44 um what else I was a an early uh founder
2:48 in the early days of the World Wide Web
2:50 I was a co-founder of a company called
2:51 agency.com which is one of the first
2:54 digital agencies
2:55 and uh oh good looks like someone Lauren
2:59 Lauren lay looks like you're coming to
3:01 the uh to the salon I'll talk about the
3:03 salon in a second as well
3:05 um I was there for the early days of the
3:06 world wide web and why I mentioned that
3:08 is that
3:10 the early days of the web and the early
3:12 days of this generative AI stuff which
3:14 I'm sort of marking uh the launch of
3:17 chech EPT as kind of this the new launch
3:19 point of this generative AI stuff
3:22 um
3:22 very very similar a lot of parallels
3:24 from the early days of the web in the
3:26 early days of uh of this generative AI
3:28 stuff so that's what this channel is
3:30 about these URLs behind my head uh
3:34 chat.openai.com that's the official chat
3:36 GPT website bing.com
3:39 is Microsoft's search engine but if you
3:42 click on the chat button that's also
3:43 chat gbt it's gpt4 and you get to use it
3:47 for free it's got some nuances from the
3:50 official chat GPT I'm a fan of the top
3:52 one but if you don't want to spend the
3:54 money for gpt4 up here go there and it's
3:56 free and then you can also if you go to
3:59 poe.com that gives you access to six
4:01 different large language models that you
4:03 can kind of compare and contrast
4:06 um models from anthropic and from open
4:08 Ai and from Poe which is quora
4:12 um so that's that and then prompts.chat
4:14 that is it's really just a document
4:17 that teaches you a little bit about
4:19 prompting and then it's a whole long
4:21 list of different personas that you can
4:24 assign to Jet gbt like act like a
4:27 screenwriter act like a business coach
4:28 act like a life coach act like a I don't
4:31 know poet comedian whatever and it will
4:33 go off and do that good evening Emilio's
4:35 wife welcome welcome welcome if you've
4:38 got I'm echoing hmm I don't know what to
4:41 do about that
4:43 well
4:47 um
4:49 hey Tick Tock stop making me Echo
4:53 any suggestions for designing a clothing
4:55 line with Ai No Echo you're good okay
4:58 great it might be on your end there
4:59 Emilio's wife oh you might want to turn
5:02 off the echo feature get it it's comedy
5:06 we do we do stand-up comedy even though
5:08 I'm sitting oh
5:10 I get no respect
5:12 all right I'm a gen xer I'm bitter we
5:15 have fun and uh let's let's get into
5:18 this so if you've got questions about AI
5:20 pop them in the questions below I will
5:22 do my best to answer them all I will
5:24 fall behind that's a promise and let's
5:27 get started
5:29 what prevents Bad actors from making
5:31 Skynet
5:33 um they can't get cast in any films
5:35 [Laughter]
5:38 or joke
5:39 see I got a Hamilton shirt on that's
5:41 it's uh that's thespian comedy
5:44 um I think what you're talking about
5:46 however is um horrible people from using
5:50 AI to destroy all of humanity
5:54 um
5:55 this is this is perhaps a naive answer
5:59 hahaha
6:00 I think that we are early enough in
6:04 whatever this stuff's going to become
6:07 and and it's the reason this channel
6:09 exists right so if you haven't played
6:11 with chat EBT I strongly encourage you
6:13 go there and start playing while you
6:14 hear me ramble I think that any anything
6:17 that we do right now that bends the use
6:20 of AI toward the good is going to have a
6:22 disproportionately positive effect on
6:24 the long-term trajectory of generative
6:27 AI
6:28 um are there risks that bad actors are
6:31 going to do bad [ __ ] I think it's more
6:32 than risks I think Bad actors will do
6:34 bad [ __ ]
6:36 um but I also think that
6:39 other people are going to do remarkable
6:41 stuff like remarkably positive stuff and
6:44 so go start playing go start
6:47 um like like here's the thing
6:50 even though Bad actors might make the
6:52 next Skynet and might take down Humanity
6:56 um don't let that stop you right now
6:58 from getting curious about this stuff
7:00 and going and learning what it's capable
7:02 of and what it can do for you right now
7:04 for your community for your business for
7:07 your new company for your friends go
7:11 start playing with it
7:13 um learn about it talk about it
7:15 um share what you're learning you're the
7:17 CEO of Vine I love Vine no I'm the CEO
7:21 of storyvine which uh funny story we got
7:25 a little spat with the old Twitter
7:27 lawyers over that one uh because we were
7:31 there first and after I don't know about
7:34 a year of back and forth with their
7:35 lawyers they're like yeah you probably
7:37 on the trademark but we're Twitter so go
7:40 ahead and sue us if you want
7:42 so yeah it's life in a big city but
7:44 anyway we're still storybine
7:48 but it is a video platform which is why
7:51 we got in a trademark lawsuit with
7:53 Twitter
7:54 okay drinking cocktails like on the show
7:57 yeah well if there were booze in the in
8:00 in Guild coffee sure but I'm I'm uh I
8:04 took a little nappy right after dinner
8:06 so I'm still waking up
8:08 um okay
8:10 thanks for your reply today you're
8:11 welcome Troy I want to design a clothing
8:13 line with AI any any suggestions on
8:16 where to start so let's just go play one
8:19 of the things I like to do here is just
8:20 let's just flip the camera around
8:23 turn off the old wacky URL image which
8:27 by the way that was generated in stable
8:28 diffusion
8:30 and let's go do some live uh stuff let's
8:35 go do some live
8:38 come on
8:40 uh some live what do they call it
8:43 I don't know uh fashion designing what
8:46 are we what are we doing
8:49 suggestions for designing a clothing
8:50 line okay we're gonna design a clothing
8:52 line
8:53 all right so first thing we're gonna do
8:55 we're gonna say
8:56 um I
8:57 want you to act like a world
9:02 class
9:04 clothing designer
9:09 tell me oh wait let me
9:13 I'm gonna go use gpt4 connected to the
9:15 internet because maybe just maybe
9:19 it will connect to the internet tell me
9:21 what are the latest
9:26 fashion trends
9:29 and we'll get something here maybe it it
9:31 may possibly because I put latest in
9:33 there it may fire off a little plug-in
9:35 to go search the internet it may not
9:39 it may just hang because I'm doing this
9:41 live live from Denver
9:44 playing on chat gbt what'd you do
9:46 tonight I don't know I watched some old
9:48 guy try to load a website
9:50 how'd that work out for you yeah it's
9:52 pretty awesome
9:58 browsing the web see what did I tell you
10:00 I told you it was gonna do that
10:02 am I good or what
10:04 clicked on a link what did I click on
10:06 where did it go
10:08 the spring 2023 the Vogue all right the
10:11 Vogue editor shopping list all right so
10:13 vogue's gonna tell us what the latest
10:14 trends are
10:16 it's gonna be fantastic it's thinking it
10:18 went to some other site 2023 fashion
10:21 trends
10:22 so it's reading multiple sites that's
10:24 kind of cool so if you haven't if you
10:26 haven't seen plugins or if you haven't
10:27 seen chat gbt connected to the internet
10:29 this is kind of what it's doing in the
10:31 background when it goes and searches the
10:33 internet
10:34 that one said reading content failed
10:36 that might have been behind a paywall or
10:37 something like that
10:39 so it's trying it again it's probably
10:40 gonna fail again but we'll see
10:45 click three times now all right this is
10:48 getting boring come on chat gbt just
10:50 give me an answer
10:57 come on come on come on you can do it
11:00 you can do it
11:02 uh by the way welcome here my name is
11:03 Kyle Shannon uh what we're doing right
11:05 now is is watching uh Chachi PT in a
11:09 seemingly endless loop trying to read uh
11:11 a website and uh this channel
11:16 this live what you're watching right now
11:18 it may or may not get more exciting than
11:20 this chances are are uh
11:22 are good that it won't okay
11:24 sure I understand the following are some
11:26 major Trends in men's fashion I didn't
11:29 say men's fashion but that's okay relax
11:31 tailoring outdoor inspired pieces
11:33 sustainable materials bold prints
11:35 monochrome outfits unfortunately okay
11:38 that's good enough so I'm gonna say okay
11:41 great
11:42 write me a
11:44 visual prompt
11:47 of a
11:50 um
11:51 male model
11:54 on a fashion shoot in New York City and
11:59 describe
12:02 his uh whatever fashion
12:06 clearly I I purposely made that kind of
12:09 a shitty prompt because I'm going to
12:10 show you something fun here
12:12 oh no it's going back out to the
12:14 internet I'm gonna stop that stop that
12:16 we're gonna turn off the internet
12:17 because I just oh wait but I'll lose my
12:19 context ah dang it okay we'll do this
12:22 again I know this is horribly boring
12:23 people are leaving here in droves
12:25 because they're like why are we just
12:27 watching a screen of someone waiting for
12:29 a website to load seriously it's gonna
12:32 be so worth it
12:33 you don't even know how cool this is
12:34 gonna be man it's gonna be so stinking
12:38 awesome
12:39 lucky
12:41 name that movie
12:45 that was also a really bad uh impression
12:48 of yawn Dynamite
12:53 something is wrong yeah something is
12:55 wrong as you're taking forever tonight
12:57 because apparently people like playing
12:59 with chat chippy tea
13:01 by the way if you're here uh because you
13:04 watched a video that said you can make
13:05 money with Chachi pite
13:08 that's not what you're gonna find here
13:10 oh talk about some [ __ ] and I'll show
13:12 you some cool [ __ ] but
13:14 and we can talk about entrepreneurship
13:16 and starting businesses and making money
13:18 in general
13:20 but
13:21 AI is there's no get rich quick scheme
13:24 with AI any more than there's a get rich
13:27 quick scheme with okay I'm done with
13:29 this okay enough about that so that was
13:33 a bad example all right
13:34 we are now at po-po.com is one of the
13:37 sites that I tell people to go check out
13:39 and one of the things that Poe has is
13:42 it's got these little
13:45 um these Bots and and these are just
13:48 Bots that people made so this this one
13:50 right here MJ Captain is a bot that
13:53 helps write mid-journey prompts so I'm
13:55 gonna say
13:57 um male model
14:00 um
14:03 um photographed
14:06 in New York City wearing
14:10 relaxed
14:13 fit out
14:16 Dorsey
14:19 uh High and
14:22 fashion I don't know not great but what
14:25 this prompt will do is it will take that
14:27 and very very quickly give me something
14:28 better a photo of photo of a male model
14:31 walking down a busy Street in New York
14:33 City wearing a high-end relaxed fit
14:35 outdoorsy fashion so it kept that in the
14:38 style of Richard avadon nice shot using
14:41 a 35 millimeter camera with the title
14:43 Urban Explorer focal point on the
14:45 model's face black and white color
14:48 palette with high contrast capturing the
14:51 energy of the city around him okay I'm
14:53 going to change that a little bit but
14:54 I'm going to go copy that so now what
14:56 we're going to do is we're going to hop
14:57 over to Discord
14:59 and if you've never seen Discord before
15:01 I'll give you a little lay of the land
15:03 in the left-hand side of Discord these
15:05 are all different servers so those are
15:07 different communities the community I'm
15:09 in right now is the AI Salon which is a
15:11 community that I started about just a
15:13 bunch of people geeking geeking out on
15:15 AI stuff and within it these little
15:17 things here are channels so different
15:19 people having like here's General
15:21 discussion and they're talking about
15:24 plugins and chat GPT and here's video
15:27 tool talk and then here's mid Journey so
15:30 this is a mid Journey channel so if
15:32 you've never used mid Journey it is in
15:34 my opinion by far the best of all the
15:36 image things so if you're doing
15:38 if you want to do a fashion line and you
15:41 want to generate things I would I would
15:43 definitely recommend mid-journey because
15:45 it's just better and then how you make
15:48 an image is you type slash imagine
15:51 and then it gives you a little thing for
15:53 a prompt and I'm just gonna paste The
15:55 Prompt in there that we had before and
15:57 I'm gonna say I don't want this to be
15:59 I'm going to say instead of black and
16:01 white color palette I'm going to say
16:02 cross
16:04 processed
16:07 color palette with high contrast that's
16:09 good capturing the energy of the city a
16:12 three by two image no we want this to be
16:14 yeah that's good
16:16 we went three by two now we want two by
16:18 three we'll do two by three
16:21 it'll make it taller than it is so
16:24 that's aspect ratio AR no rigid upright
16:26 or boring uh version 5.2 so we're using
16:29 version 5.2 of mid journey and I'll hit
16:32 return and so now this is going to go
16:34 off and
16:36 um
16:37 start thinking a little bit hopefully
16:39 it'll work a little faster than chat gbt
16:41 did that that was excruciating earlier
16:43 uh welcome everybody looks like there's
16:45 some new people here if you're new here
16:46 uh someone asked me about how they might
16:48 get started doing uh high-end fashion a
16:52 clothing line with AI designing it so
16:55 I've popped over here to uh what's it
16:58 called uh mid-journey which is a
17:00 generative AI image tool that in my
17:03 opinion is one of the best now you'll
17:05 notice as these things as as this is
17:07 rendering it kind of started out as this
17:10 really four blurry blobs and then they
17:12 got slightly less blurry and slightly
17:14 less blurry what you're actually
17:15 watching there is you're watching the
17:18 the this diffusion model it's called a
17:20 diffusion model it's the way it actually
17:22 generates images it kind of averages
17:24 probably thousands of images together
17:28 um and and uh into basically kind of a a
17:32 highly noised image and then it denoises
17:35 that kind of random noise image over 20
17:39 or 30 or 50 or 70 steps
17:41 and so right now we're at 93 percent
17:45 and so there it is so
17:48 I can open this in a browser we can zoom
17:51 in on these we'll see like that one's
17:54 kind of interesting that one on the top
17:55 left there mid-journey's pretty good
17:57 with the fashion stuff that dude is
17:59 intense
18:01 there's
18:03 I don't like the color of that that
18:04 one's cool all right so so there's our
18:06 images so let me show you a couple other
18:09 things you can do with mid-journey which
18:11 are kind of fun
18:12 so when you create a set of images it
18:14 gives you four options and then you've
18:16 got these four buttons U1 through four
18:18 and V one through four V means variation
18:20 u means upscale and then and then that
18:22 redo button is a redo button so let's
18:25 take I guess uh let's do the the yellow
18:28 one that one's I don't know uh let's
18:30 let's do the third one so it goes one
18:31 two three four so I'm gonna do V3
18:35 and I'll do V1 just for the heck of it
18:38 so that's going to give us four
18:41 um alternates for each one and three
18:45 and it should do that pretty quickly
18:49 we'll see
18:52 all right
18:55 so here we go this is the yellow one
18:58 obviously you can see a yellow blob
18:59 there
19:00 and I mean this is kind of fascinating
19:02 stuff where you know it starts out
19:04 completely blurry and like now this he's
19:07 got like a face behind him here it'll be
19:09 interesting to see if that turns into a
19:11 face he's got a face behind him
19:14 not sure what that's about maybe they're
19:16 like billboards
19:20 that's wild
19:22 not quite sure why that happened but you
19:24 know that's part of the part of the joy
19:26 here is just experimenting and figuring
19:27 stuff out and seeing what works
19:32 all right so
19:34 that one is really creepy
19:36 let's go with the creepy one so I'm
19:38 gonna upscale number two here just
19:40 because it's creepy and if you're asking
19:41 why did it put the picture back there I
19:43 don't know because it did
19:46 and then you can do things like tell it
19:48 not to put other faces in the picture
19:49 and things like that but I'm going to
19:51 hit YouTube
19:52 fine band fine fine band
19:55 and now it's going to upscale that
19:59 and then there's going to be some one
20:01 other cool thing I'm going to show you
20:02 which is the unzoom feature so part of
20:05 mid-journey is this new zoom out feature
20:08 or not unzoom zoom out
20:11 Okay so
20:13 here's our here's our high-end thing
20:15 let's go look at it that's pretty nice
20:17 this dude's head is way too small he's
20:20 out of proportion so that's that's kind
20:22 of comically bad but that's okay we'll
20:24 go with it all right so now
20:26 let's say so so I did this Photograph
20:29 and let's say that I was hired to make a
20:32 vertical photograph but all of a sudden
20:33 I realized oh crap I need a landscape
20:36 photograph I need this to be wider than
20:38 it is tall and I can't really crop in on
20:40 him because it would just be too weird
20:42 so I can do zoom out two times zoom out
20:45 1.5 times or or do a custom Zoom so I'm
20:48 going to do a custom Zoom
20:50 and what I'm gonna do is I'm gonna do
20:53 aspect ratio of 16x9
20:58 which will give me a good cinema kind of
21:01 thing and I'm going to do Zoom 1.5 so
21:04 I'm going to zoom out 50 beyond what
21:06 we've got now but I'm going to do it in
21:08 16x9 and then I'll hit submit and so
21:11 this is now going to give me three
21:12 different or four different versions of
21:14 this that basically fill in the
21:17 background that doesn't exist right now
21:19 so we just created that image out of
21:21 nothing and now we're going to create
21:24 um a completely different aspect ratio
21:26 from this out of nothing you can also do
21:29 this in Photoshop in the Photoshop beta
21:31 right now with a thing called generative
21:33 fill and you'll you'll notice here
21:37 the part of the image that was in Focus
21:40 before is still in focus and we've still
21:42 got this weird blurry stuff in the
21:44 background so as these images develop
21:46 that'll come more and more into Focus
21:48 that looks like Macy's right there
21:50 that's kind of cool that makes that
21:51 thing look much more like a billboard
21:54 than a creepy dude looking over his
21:56 shoulder
21:57 it's given us some nice perspective with
21:59 these there's a Yellow Cab right so we
22:01 know we're in New York City
22:04 right now we're at 62 percent we're
22:07 going we're going Bueller Bueller
22:13 any bonus points for anyone here who's a
22:16 gen xer
22:17 laughs
22:18 because is that is that done that we're
22:21 at 93 percent
22:23 and then okay so let's look at these so
22:25 oh that one's kind of cool this is like
22:27 in the the meat packing District in
22:29 lower Manhattan with this with the brick
22:31 streets
22:33 um I think that's the one I like that
22:35 kind of looks like Soho a little bit and
22:37 Brick Street so let's do that so we're
22:38 gonna upscale number one
22:41 this should go pretty quick
22:44 and there you have it within
22:48 within a few minutes whatever however
22:50 long this has been 10 minutes of me
22:52 doing this
22:53 you know we went from I don't know
22:55 anything about fashion to there's our
22:57 fashion dude our New York City shoot so
23:00 you could see that if if you actually
23:01 had an eye for this and you knew the
23:02 terminology and you knew what you wanted
23:04 to do in in very short order you could
23:07 put together you know at a minimum mood
23:10 boards right uh potentially even
23:13 significantly beyond that so let me flip
23:16 my camera around welcome everybody my
23:18 name is Kyle Shannon this is the AI
23:20 learning lab uh I am learning this stuff
23:22 kind of in real time while talking about
23:25 it with y'all if you have questions
23:27 about AI pop them in the comments below
23:29 I do my best to answer everything
23:31 um I'm an entrepreneur by training I've
23:34 started 12 or 15 companies in my day and
23:37 that's why we're here so uh do that I'll
23:39 talk talk about stuff below and let's
23:42 get moving on what is the bot name
23:47 that
23:49 created that prompt am I toward the
23:50 bottom where am I
23:52 did I lose my place again the the bot
23:54 name was uh
23:56 MJ MJ Captain so if you go to poe.com
24:01 MJ captain or poe.com mid-journey both
24:06 of those are Bots that just regular old
24:08 people like you and I created on poe
24:11 poe.com poe
24:14 um you can make your own bot you could
24:15 make your own mid Journey bot if you
24:17 like figure out an interesting way to do
24:19 mid-journey prompts you could turn that
24:21 into a bot so someone gives you a piece
24:23 of crap prompt and you turn it into
24:24 something good that is in your power
24:26 I'll show you how to do that in a bit
24:29 let me go find out where I was in my
24:31 little questions here
24:34 and so because there were probably
24:36 there's a lot of questions here thank
24:38 you all for participating I really
24:39 appreciate it
24:41 all right here's where I was
24:45 Dolly too is not bad for visual creation
24:47 dolly two is okay if you go to bing.com
24:50 create that's dolly two so that's open
24:53 ai's image generation tool
24:56 um
24:56 of all the inner image generation tools
24:59 I think dolly is is kind of fourth from
25:03 a quality standpoint
25:05 um first is mid-journey in my opinion
25:07 second is stable diffusion stable
25:09 diffusion is way more
25:11 um there's a much healthier developer
25:13 community so there's lots more features
25:15 but you got to be a [ __ ] geek to use
25:17 it number three is uh Adobe Firefly
25:22 there are image Generation stuff is
25:24 really good and it's all commercially
25:26 safe
25:27 um and then they've got generative fill
25:28 within Photoshop that's really good and
25:30 then Dolly if you go to bing.com create
25:33 that's Dolly and that's pretty good you
25:35 can also call Dali from API calls
25:40 um I'll even come back okay beautiful I
25:42 can never evoke a reaction out of you am
25:45 I just not funny oh Eric wait what did
25:48 you say did you crack a funny joke that
25:50 I missed
25:51 I don't know I don't think you did I
25:54 don't think you did a funny Eric I don't
25:55 think it's true wait oh no that was to
25:57 RM Walter
26:00 or good actors for that matter oh I get
26:02 it it was an okay joke Eric it wasn't
26:06 great it was okay
26:09 now Echo New York you're good thank you
26:11 very much
26:14 um
26:14 yes you should sound like God yeah yeah
26:17 I'll turn on my echo
26:18 saw I am Thor no you're not dolly two is
26:23 not good I'll even come back okay here
26:25 we go
26:27 um I'm like 90 AI LOL you too
26:32 um what Masters should I do in AI uh I
26:36 have no [ __ ] clue what you're
26:37 interested in
26:38 um
26:41 if you're interested in the Science and
26:43 Technology and Engineering side of AI I
26:46 don't know go to Stanford go to MIT
26:48 there's there's probably lots of places
26:49 to go if you're interested in all this
26:52 generative AI stuff there's not a
26:53 masters in it right now this is all too
26:55 new this is all it is evolving so fast
26:58 no one knows anything do I love Denver
27:00 we almost relocated from Irvine
27:03 um Denver is beautiful the people are
27:06 super nice the weather's amazing
27:09 um
27:11 I miss the energy of the East Coast I
27:13 miss New York I've been out here for 14
27:15 years and
27:17 um
27:18 I miss Snappy people you know when you
27:21 like so so here's the thing when you're
27:23 in New York and you're like hey I got
27:25 this idea no matter who you talk to it
27:28 could be on the subway it could be a
27:29 stranger it could be your best friend
27:30 could be someone at work you're like hey
27:33 I got this idea
27:34 what happens in New York is you get
27:36 instant feedback oh that's a piece of
27:38 [ __ ] or oh that's really good hey you
27:41 know it would be really good
27:43 there's this there's just this drive
27:45 this energy of like let's [ __ ] go
27:47 let's [ __ ] get there
27:49 in Denver I'm like hey I got this idea
27:51 oh that's nice
27:55 it's just like the energy just goes hmm
27:57 out of the room it's really hard to have
28:00 an elevated conversation here and it
28:02 drives me [ __ ] batty
28:05 but I do like living here I just I miss
28:08 that energy I missed I didn't realize
28:11 how much
28:13 that kind of back and forth and that
28:15 energy to just [ __ ] do it the best
28:17 how much it drove me it drove me a lot
28:20 so I miss it hard I miss it hard that's
28:24 so hard well why don't you get up and
28:27 leave Kyle I don't know because I like
28:30 complaining
28:32 what could I complain about if I had
28:34 everything I wanted huh
28:38 um
28:40 T is hype Google of Google okay gumba
28:45 Limba
28:48 you clearly haven't used it so whole
28:51 purpose of this channel go there launch
28:54 that make an account
28:56 if you're too much of a [ __ ] to give
28:58 them your phone number because they do
28:59 ask for your phone number then go to
29:00 that one go to bing.com and click on the
29:02 chat button and start playing with these
29:05 things and if you don't know anything
29:05 about prompting or what to do with them
29:07 or if you can't get it to do anything
29:09 beyond what a Google search does then go
29:12 here to prompts.chat that will teach you
29:14 about prompting and it'll start to give
29:16 you it'll this document will help you
29:18 start to scratch the surface of what
29:21 that thing [ __ ] makes possible this
29:24 is
29:25 some of the most remarkable technology
29:27 ever put in the hands of mere mortals
29:31 this this represents like Decades of
29:34 machine learning development
29:39 put into
29:41 a package that is
29:43 if you use it right indistinguishable
29:45 from Magic
29:47 that's an Arthur C Clarke
29:49 quote I think it's Arthur C Clarke
29:53 any significantly advanced technology is
29:56 indistinguishable from Magic if you
29:57 think Chachi PT is just hype you haven't
29:59 used it go there I honestly honestly I
30:03 feel like everyone owes it to themselves
30:05 to go use this thing long enough to have
30:08 a Kevin McAllister moment which is it's
30:10 a U.S film for uh called home alone and
30:14 the little kid was left alone and he had
30:16 this moment
30:18 you will have that moment if you play
30:20 with this
30:21 if you sort of get out of your way and
30:23 stop judging it and stop diminishing it
30:25 all right years ago when AI started
30:28 running the financial system things got
30:30 really bad well
30:32 well yeah so so that an important thing
30:35 to note ai's been around for [ __ ]
30:37 decades
30:38 the difference now what the inflection
30:41 point that we have passed that that
30:42 started on November 30th 2022 six months
30:46 ago
30:47 is Chachi PT
30:50 made machine learning accessible
30:54 to the rest of us right
30:56 I don't disagree that a lot of the AI
30:59 implementations that have been done so
31:01 far haven't been great haven't been as
31:04 successful as predicted you know uh
31:07 Watson from IBM their Watson Health
31:09 Initiative kind of flamed out
31:12 um so so yeah there's some things it
31:14 does really good there's some things
31:16 that does really bad but this
31:19 and and
31:20 you know the similar tools like it and
31:23 how fast that got adopted by just
31:25 regular people and business people is
31:27 going to transform
31:29 the way we all work so so again
31:33 don't just sit back and wait for this to
31:36 get better jump in right now
31:39 ideally IA will reach a place where it
31:42 can defend itself against Bad actors
31:44 yeah hopefully so and I thought listen I
31:46 think that there's going to be lots of
31:48 efforts to do that I think I think at
31:50 this point there's enough awareness of
31:52 the good and bad potential of this that
31:54 people are now aware of it so I it's
31:57 it's looking promising
31:59 how would you use AI to scale up
32:01 Productions and help with onboarding I
32:03 don't know what kind of Productions like
32:05 like film Productions
32:07 um and help with onboarding
32:09 um
32:10 you can have it help you put together
32:11 project plans it's really good at taking
32:14 other documents and understanding them
32:16 if you're doing Film Production uh if
32:18 you use the
32:20 um here let me let me show you something
32:22 that's kind of cool this this may not be
32:24 at all what you do but I think this is
32:26 something worth worth sharing
32:28 um oh that's funny that dude
32:30 our little fashion dude
32:36 I'm a fashionista look at me and my
32:39 yellow jacket all right uh
32:42 that's pretty funny well it was funny to
32:44 me okay
32:46 so I'm gonna go to post so if you go to
32:48 poe.com
32:50 when you first get here
32:52 it will look like this where it's just
32:54 it's just like uh like a Chachi PT
32:56 window down below you can type in a
32:58 prompt and it'll it'll give you some
32:59 stuff
33:01 up at the top there's explore you can go
33:04 explore Bots and there's create a bot if
33:06 you want to create a bot you can do that
33:07 there but these first six choices here
33:11 sorry there's explore there's creative
33:12 bot these first six choices here
33:15 um are are different chat CPT models
33:17 basically Sage is Poe's model
33:21 gpt4 Claude plus and Claude instant 100K
33:24 those are all subscription only it's
33:26 like 20 bucks a month gives you access
33:28 to those three and then you get Claude
33:30 instant and chat GPT so the clawed ones
33:32 are from anthropic Sage is from Poe
33:35 which is quora and then Chachi BT and
33:37 gbt4 are from openai
33:40 this one right here Claude instant 100K
33:43 what is that 100K mean
33:46 with normal chat gbt you have what's
33:48 called a a 4 000 token context window
33:51 it's increasing and for some people it's
33:54 8 000 now but for the most part it's
33:56 four thousand what that means is three
33:58 thousand words you can you can have a
34:01 context window essentially a chat
34:03 history of about three thousand words
34:05 this hundred thousand is a hundred
34:07 thousand tokens that's about 75 000
34:09 words so
34:11 I I yesterday I was trying to find a
34:13 screenplay in PDF format and I couldn't
34:16 but I found one today so this is the
34:18 Matrix this is the screenplay for the
34:20 Matrix okay so screenplays are what 120
34:23 pages I I assume this one's longer than
34:25 that
34:27 124 126 okay so slightly longer than
34:32 normal
34:34 can I copy please tell me I can copy
34:36 here yeah okay so fade in
34:39 so I am going to
34:43 copy this entire screenplay so I'm going
34:45 to hold down my shift key
34:48 did that why did that not eh what's
34:52 going on
34:57 uh this is very weird
35:01 um
35:12 okay
35:14 this is the shooting script the numbered
35:16 shooting script okay there it is
35:18 okay so now I've copied
35:20 all of it I didn't get all the fade in
35:22 but I got it all
35:24 I'm gonna go to Claude instant 100 000.
35:27 and I'm just going to come into this
35:29 little chat box I'm going to say
35:31 based on this screenplay
35:36 and then I hit shift return to do some
35:38 line breaks and I'm going to paste in
35:41 all of The Matrix
35:42 so let's say you're you're uh in a
35:46 production thing uh or you're a
35:49 scriptwriter and someone says hey go
35:50 read the new version of the script let's
35:52 say
35:53 um
35:54 describe the main
35:57 characters
35:59 and what happens to them
36:04 question mark
36:06 and so now it's gone off and it's red
36:10 the whole of the screenplay
36:12 and it's we're waiting it's doing some
36:15 processing
36:17 still waiting
36:20 the main characters in The Matrix are
36:23 Neo he's a protagonist the story follows
36:25 this journey from being an ordinary
36:27 person living in the make Matrix to
36:29 realize he is the one Morpheus Trinity
36:32 agent Smith agents sentient computer
36:35 programs sent by machines in summary
36:37 these are the key events the key events
36:39 are Neo realizing he's the one
36:41 undergoing combat so now if I'm in
36:43 production again I don't know what
36:45 production you were talking about I
36:46 could say please
36:48 provide me a detailed
36:53 prop
36:54 list for Neo in the First Act
37:00 of the film
37:03 we'll see what it does see if it gives
37:05 us that
37:09 I didn't train it with a prop list was I
37:12 don't even know if it knows what a
37:13 propolis is but I assume it does because
37:15 these models have betrayed have been
37:17 trained on essentially all of the
37:19 internet
37:20 here's a detailed prop list for Neo in
37:22 the First Act of the Matrix home and
37:24 work computer equipment Neo works as a
37:27 programmer and a hacker so he has
37:29 numerous computers monitors Towers
37:31 keyboards clothes at the beginning he
37:32 dresses dresses like a typical late 90s
37:35 clothes for a hacker baggy jeans
37:37 t-shirts hoodies
37:39 so there you go
37:41 um
37:43 what are the first
37:47 uh 10 locations
37:51 and could any of them
37:54 be
37:56 re
37:58 -dressed
38:00 versions of the same location I don't
38:04 know if that's a thing
38:05 again but
38:07 it will probably
38:09 figure some [ __ ] out for us
38:16 um
38:18 here are the first 10 locations
38:22 subway station office virtual reality
38:26 space
38:27 as you can see several of the locations
38:30 like hotels Office Buildings and subway
38:31 stations could utilize redressed
38:33 versions to represent similar locations
38:36 within the confines of the Matrix so
38:40 that is from Poe using the Claude 100K
38:44 thing you could pop in any sort of large
38:47 document like that these models are
38:49 incredibly good at analyzing big
38:52 documents
38:53 and helping you do that kind of stuff so
38:58 hope that helps
39:05 um by the way if you're new here welcome
39:06 my name is Kyle Shannon this is the AI
39:08 learning lab
39:10 we'll sing Hamilton songs later we do
39:12 sing-alongs every night and uh these
39:15 URLs behind me that's not true by the
39:17 way I'm a smart ass that is true
39:21 uh if you're new to generative AI if
39:24 you've not played with chat GPT that
39:27 chat.openai.com that's the official
39:29 chatgpt website go there you owe it to
39:31 yourself to go there
39:34 um bing.com this is Microsoft search
39:36 engine but it's also chat GPT if you
39:38 click on the chat button that's gpt4 for
39:40 free po I just showed you it's got a
39:43 bunch of things there including that
39:44 hundred thousand token version
39:46 and then prompts.chat is
39:49 um a document that will teach you about
39:50 prompting and it'll give you a bunch of
39:52 ideas for personas that you can give
39:55 chat GPT to act like act like a
39:58 screenwriter act like an accountant act
40:00 like a counselor and it'll do that and
40:02 then futurepedia.io is
40:05 um
40:05 just a big directory of like thousands
40:08 of different AI tools most of which are
40:11 horrific pieces of [ __ ] that you
40:13 shouldn't waste your time on
40:14 but many of which are so mind-blowing
40:17 that you won't be able to get your head
40:20 around what in the hell is coming
40:23 good luck
40:26 how about Boomer points Boomer points
40:29 are okay you know you got your Boomers
40:31 you got your gen xers we get ignored in
40:33 most of the lists we're cool with that
40:36 we're bitter
40:38 we're better at you Boomers you [ __ ]
40:41 us up
40:44 all right but you made us strong
40:49 by the way if you haven't seen the bear
40:53 turn me off and go watch that [ __ ]
40:58 um episode six and seven of season two
41:03 holy [ __ ]
41:05 some of the best television I've seen
41:07 ever
41:09 if you haven't seen the bear at all
41:10 shame on you all right kind of toward
41:14 the bottom I'm new to Ai and an elder
41:16 Millennial welcome welcome you're on the
41:18 cusp huh
41:20 um do you know any free AIS that can
41:22 create and edit vids not really edit
41:25 okay so if you want to do editing there
41:28 are probably some AI tools that do some
41:30 you know chopping and snipping and
41:32 things like that so I would say go to
41:34 futurepedia.io
41:36 and um just look for the video tools
41:41 um like I said there's going to be a
41:42 bunch of them in there that are going to
41:44 want to charge you a lot of money and
41:46 they're big pieces of [ __ ] so just you
41:49 know go in knowing that if you want to
41:51 generate stuff there's a couple of uh
41:54 things out there right now probably the
41:56 one that's worth experimenting with is
41:58 Runway ml so go to run just Google
42:01 runwayml as in machine learning and then
42:04 they have an app but they also have a
42:06 Discord server and they've got this
42:08 thing called gen 1 and they've got one
42:09 called gen 2. so gen 1 you can upload a
42:12 short video clip and then give it a
42:14 prompt to basically use the action from
42:17 that video clip but give it a completely
42:19 different visual style that's pretty
42:21 cool and then Gen 2 is just you type
42:25 um a video scene and it will make a
42:27 video
42:29 um
42:29 that's relatively new technology if you
42:33 use those tools and you're and and your
42:36 thought is
42:37 um these are disturbing and creepy
42:40 so were all the image generation tools a
42:42 year ago so a year ago these like
42:45 mid-journey and stable diffusion the
42:47 images that they created were kind of
42:49 they weren't laughably bad but they were
42:51 disturbingly not good
42:53 that's where the video is today so just
42:55 sort of imagine it a year from now
42:58 um and it will it will get really good
43:00 so hopefully hopefully that helps all
43:02 right can you zoom out from an image
43:04 that's not AI generated yes you can let
43:06 me go show you that we'll do it in uh
43:08 Photoshop
43:12 [Music]
43:15 all right I am gonna go
43:18 so here's Photoshop
43:22 uh this is by the way to get to this
43:25 this was the thing I did this is an AI
43:27 generated pendant and it was
43:30 it was a still life with some plants
43:32 behind it
43:33 and I did some generative fill to
43:35 replace the plants with a with a person
43:38 um it kind of messed up the face the
43:40 face is weird but but uh it was
43:42 interesting
43:44 so let's see let me flip over to finder
43:47 and go find a photo photo graph what's
43:50 that that's a video okay here's okay
43:52 here's a bad photo of me so let's pop
43:55 this over to
43:56 to photoshop
44:03 did it bring it in oh I see it brought
44:05 it in as a
44:08 um
44:10 so I'm going to cut that I'm gonna go
44:13 new
44:15 all right let me do that
44:17 oh I see what it did
44:19 hang on a sec new
44:22 paste
44:26 damn it
44:27 oh
44:37 hang on
44:40 ah
44:42 I'm trying to move too fast here people
44:45 bear with me all right there we go
44:48 all right so so there's my own dumb fat
44:51 head
44:52 hey there my name's Kyle
44:55 um and let's say let's say we want to we
44:58 want to make a video frame
45:01 so so we'll move this I don't know kind
45:05 of there
45:07 and then we'll grab the
45:09 crop tool
45:13 [Music]
45:15 oh bottom purple that damn meow that's
45:36 um
45:38 all right so let's say we want my my
45:40 dang head
45:42 in some sort of rule of thirds thing
45:44 like that
45:46 so what I can do is I can just grab a
45:50 little photograph of myself here I'm I'm
45:52 selecting that let me make it a little
45:54 bigger so you can see it
45:55 so you can't really see the selection
45:57 tool but basically I've selected I've
45:59 selected the photograph of me and now
46:02 I'm going to invert that selection so
46:04 I'm now I'm selecting all the white
46:05 stuff in the background and you'll
46:07 notice there's this little contextual
46:09 menu thing here that's part of the new
46:11 beta and that what that button says now
46:14 is is generative fill
46:16 so I can either just hit return and it
46:19 will sort of figure it out or I could
46:21 type in modern
46:24 um
46:24 office with
46:27 uh impressive white board and windows
46:34 and then I'll hit generate and
46:38 oh I've got a quit Photoshop oh my God
46:44 why why
46:46 all right let me say this let me save
46:49 this image we'll do
46:51 KS test
46:55 where am I saving at desktop all right
46:57 it's gonna be so good
46:58 what'd you do tonight I don't know I
47:00 watched some guy try to launch Photoshop
47:03 that sounds awful it was kind of bad
47:08 okay so I'm relaunching Photoshop beta
47:15 which the code name was Strawberry
47:18 Letter apparently okay so we're back
47:20 here we're gonna make it big again I'm
47:23 gonna select this photo again
47:28 and we lost almost everyone in the chat
47:33 this is the die hards now we're gonna
47:36 invert the selection
47:38 and so now I'm going to do generative
47:40 fill I'm going to say modern
47:43 office with impressive
47:47 whiteboard and windows and we're going
47:52 to say generate and so now it will go
47:55 generate some stuff
47:59 time to update
48:01 oh I guess there's a new build of
48:04 of this version of Photoshop
48:09 so there you go so that is an office and
48:13 it'll give us three versions of that
48:16 and so it gave me a big giant shoulder
48:19 like the proportions are all messed up
48:21 there but you know it's making it out of
48:23 nothing right so it it kind of got the
48:26 glass windows right it got the the
48:28 Whiteboard here
48:30 um
48:31 let me
48:33 let me delete that layer
48:37 and then
48:39 let's go back here
48:45 select this we'll hit
48:48 inverse
48:50 and I'm just going to hit generative
48:52 fill but I'm not going to type anything
48:53 in we'll just see what it does on its
48:55 own
48:56 see if it does anything interesting
49:06 if it's more or less better oh give me
49:09 an office mate
49:14 he's got some sort of Lego toy or
49:16 something there
49:17 that's actually not bad some bikes
49:19 although it's really janky back there
49:21 that's actually pretty good that you
49:23 know
49:23 again it's sort of sort of messed up the
49:26 uh the shirt a little bit here but but
49:27 that's generative fill so yes you can do
49:29 that with any image you can also
49:31 completely replace the image and just
49:33 generate it wholesale
49:37 um
49:40 all right
49:42 let me
49:45 flip yarn camera back around
49:48 hello everyone
49:50 I'm back
49:54 if you have questions about generative
49:56 AI pop them down below I will do my best
49:59 Denver is boring I was just there
50:00 Brooklyn is great that's why it costs
50:02 way more
50:03 yes I agree David Horton
50:07 I concur
50:10 how do you answer the questions about AI
50:12 from audience in live sir
50:15 well Jesus Christ is my savior I read a
50:19 question like the one you wrote and I
50:21 talk words out of my mouth
50:23 I don't know
50:25 and if I can't answer them I won't
50:26 answer them
50:28 all right lol that's why I feel like I
50:30 could never leave Miami yeah exactly I I
50:33 know
50:34 I need to get back to the east coast
50:36 that low-key energy happens in Alaska
50:38 too people want to be polite instead of
50:40 helpful or honest agree fun fact if you
50:43 Google Google you break the internet I
50:46 like it fun fact if you delete your
50:48 recycle bin you break your computer
50:49 that's good Eric that was that see that
50:52 was good that was clever
50:54 oh you did the other fun fact Eric's in
50:56 here Eric needs to be a stand-up Eric
50:58 start a tick tock I think you could do
51:00 this
51:02 I've been working on SMS version but
51:05 needs some feedback okay cool I'm happy
51:08 to play Gen X here welcome Gen X Gen X
51:12 crowd
51:13 it's crazy good I don't know what you
51:16 mean I assume you meant mid-journey I
51:18 used AI today good
51:21 good
51:22 use it share it go to chat.openai um do
51:26 me a favor since we lost a bunch of
51:27 people when I was trying to run
51:29 Photoshop uh tap the screen give me a
51:32 bunch of likes and maybe they'll let
51:34 some people know we're in here having a
51:35 good time they'll want to come in and
51:37 play and watch an old guy try to use
51:40 tick tock in his computer
51:43 March how do you turn this thing on
51:46 Marge the computer's Off Of You the
51:49 screen went black
51:50 do we gotta take it to the Apple Store
51:52 mark
51:57 I love chats EBT I want to be able to
51:59 create a digital package of top prompts
52:01 uh you should be able to do that in
52:03 about 10 minutes have it write it for
52:05 you
52:06 um
52:07 how do you answer questions I answered
52:09 that I answer questions by answering
52:10 questions is it is Magic if you can't
52:13 distinguish the difference between
52:15 something and something Kevin McAllister
52:17 moment you realize Mom and Dad left for
52:20 France without you
52:23 but that's not that he realized they
52:25 left him that's that he put on the dad's
52:27 aftershave
52:29 but still good moment
52:32 um I'm surprised by what chat EBT is
52:34 capable of every single day you and me
52:37 both
52:38 I like I've spent the past two days
52:40 working on
52:41 um a big proactive project for a client
52:44 that I would not have done
52:47 like
52:48 what I did for the client is way above
52:51 and beyond
52:53 what's reasonable
52:54 and I put in like an hour or two but for
52:57 me to do what I did traditionally would
52:59 have taken me two or three days and I
53:01 wouldn't have done it
53:02 and and it's
53:04 it's it's astounding it's astounding
53:07 like you have a big long document and
53:09 you're like you know
53:11 we should kind of expand the focus and
53:13 add in this concept and that concept you
53:16 can take that whole document paste it
53:17 into chat GPT and go hey I wrote this
53:21 I want to add these two concepts
53:25 Focus Less on a and more on B
53:29 give me a draft
53:30 out it comes
53:33 and it's good
53:34 It's [ __ ] amazing it's just
53:37 it's just amazing
53:39 where do you find the best prompt I am I
53:42 am
53:43 if if you're a really organized person
53:46 which I'm not I I have ADHD like like
53:49 the laminated card version like they're
53:51 really the really good version the good
53:53 kind
53:54 um so my uh prefrontal cortex is just a
53:57 block of cheese so I have no executive
53:58 function so if I were good at that [ __ ]
54:01 I'd have like a whole spreadsheet or a
54:03 database or a notion set of notebooks
54:06 organized in 16 categories but I don't
54:09 um there's prompts.chat will give you
54:12 some there's there's a bunch of
54:14 different prompt libraries there's one
54:16 you can pay for them I wouldn't
54:17 recommend that at all
54:19 um I would really just recommend start
54:21 getting your head around here's the
54:23 biggest thing that made a difference for
54:24 me with prompting is
54:28 you don't have to have the perfect
54:29 prompt
54:31 what you actually have to have is the
54:32 right mindset and I know this is going
54:34 to sound like weird and creepy
54:36 when you use Google here's what Google
54:38 has trained us for the past 20 years
54:40 Google has said
54:42 write the most concise
54:45 version of what you want to find as
54:48 possible don't put in extra words don't
54:50 don't confuse me or I'll give you wrong
54:52 [ __ ] right so so here's the just give me
54:55 the [ __ ] you want to search for and then
54:56 I'll give you the 40 choices
54:58 chat jpt kind of works the opposite
55:00 where if you give it very little it will
55:03 oblige and return very little it like
55:06 it'll just return generic [ __ ] but if
55:08 you give it a lot of context like you
55:10 say hey I want you to act like an expert
55:12 historian for Egyptian hieroglyphics and
55:16 you know I'm a student of egyptology and
55:19 and then you tell it kind of who you are
55:23 and what you're trying to do
55:25 it will oblige you and it will get
55:27 better and better and better and it's
55:28 just the nature of how the tool Works
55:30 um so I I would strongly encourage you
55:32 to just really start thinking of chat
55:35 gbt more like a like a think of it like
55:39 this chat is the smartest intern you've
55:42 ever hired
55:44 um that needs an incredible amount of
55:47 guidance oh and by the way it lies a lot
55:49 it will lie to you but you give it you
55:53 give it lots of guidance and and sort of
55:55 think of yourself as the pilot it is the
55:56 co-pilot and and just start playing with
55:59 it like that and you'll kind of discover
56:01 your own kind of prompts um
56:04 but there are lots of tools out there
56:06 but but I just copying and pasting
56:08 prompts isn't going to teach you as much
56:10 about how the stuff actually works I've
56:12 Incorporated chat gbt into my daily
56:14 workflows so much so that I subscribe to
56:17 the service that's awesome Albert yeah
56:19 it just speeds things up chat is amazing
56:21 there's nothing natural about natural
56:24 language processing learn prompt
56:25 engineering yeah it's I think learning
56:28 learning to prompt right now is is a
56:30 valuable tool I think in two years it's
56:33 all this stuff's gonna drift into the
56:35 background it's going to be much more
56:37 automated and if you want to see an
56:38 example of that there's a plug-in for
56:40 chat apt called prompt perfect where you
56:43 type the word perfect before a prompt
56:44 and it will rewrite it for you kind of
56:46 like the mid Journey stuff I showed
56:48 earlier
56:50 um so so yeah chat GPT is amazing and
56:53 currently dictating my waking hours it
56:55 dictates mine it sometimes wakes me up
56:57 thinking about it too
57:01 I'm like oh I could do this with it
57:05 cyber security of the Future Good robots
57:07 and bad robots fighting on the internet
57:09 I think that's right I was talking to um
57:12 a
57:13 a group of uh people that work for like
57:17 a foundation you know they write grants
57:18 and read grants and give out money and
57:20 [ __ ] like that
57:22 and I was explaining chat GPT and this
57:24 woman goes well wait a minute if
57:26 if it does All That Couldn't someone
57:28 writing a grant proposal just write it
57:30 much more quickly and like yeah they
57:32 could and then she goes well then
57:33 couldn't the people reading those Grant
57:36 proposals read them much more quickly
57:38 I'm like yeah they could and I said you
57:41 can imagine a future where the robots do
57:44 the grant writing and the grant reading
57:45 and then I said and then what's left
57:47 what's left is you get to talk to the
57:49 human beings right so kind of there's a
57:52 very good chance that these Technologies
57:54 automate out all of the soul-crushing
57:57 time-consuming repetitive
58:01 just you know exhausting drudgery work
58:06 they automate all that [ __ ] and we get
58:08 to return to connecting as human beings
58:13 it might return us some empathy and some
58:15 human connection and things like that
58:17 that's where I'm hoping it goes so even
58:20 though cyber security of the future was
58:22 meant as a joke Eric which as a stand-up
58:25 you got some work to do on the writing
58:26 but I like the spirit
58:28 um I think that's actually not too far
58:30 from where things could go
58:33 um don't trust it because it learned
58:34 good info and bad info more bad than
58:37 good well
58:38 sure possibly but if you understand how
58:41 these things work there's a there's it's
58:43 worth Googling there's a YouTube video
58:45 of Andre carpathi from openai speaking
58:48 at the Microsoft developer conference
58:50 talking about prompting and how the
58:52 prompts work and how the tools work
58:54 and one of the things that he talks
58:57 about is he uses an example of a math
59:01 equation and if you go to chat GPT and
59:05 you're like here's a math equation go
59:06 solve that math equation for me
59:09 what it will do is it will go out and
59:11 look at everything that's been trained
59:12 on which is essentially all of the
59:14 internet right all the public internet
59:16 and what he talks about there is in all
59:19 of that data
59:21 there's
59:23 this professor that solved that equation
59:25 correctly and that information's in
59:28 there
59:28 and then there's this student that
59:31 didn't solve that equation correctly and
59:33 then 30 fellow students you know 15 of
59:37 which didn't solve it correctly so if
59:39 you just say go solve that math equation
59:41 it's going to look at all of what it's
59:43 learned related to that math equation
59:45 kind of amalgamate what it's learned and
59:48 give you some sort of answer it's just
59:50 spitting out individual words so it
59:52 looks like it's solving the equation but
59:54 that equation might be wrong because you
59:57 didn't narrow the search
59:59 because you just said go solve this math
1:00:01 equation but if you tell it I want you
1:00:02 to act like a Stanford math professor
1:00:04 that specializes in this branch of
1:00:06 mathematics and here's this equation and
1:00:10 I want you to not only solve this
1:00:12 problem but I want you to show me the
1:00:13 order of operations of how you did it
1:00:15 what you're doing is you're increasingly
1:00:17 narrowing the What's called the latent
1:00:20 space
1:00:22 of the information that it's looking at
1:00:23 and you have a much higher likelihood of
1:00:26 getting the right answer so don't just
1:00:31 don't diminish what it does and what it
1:00:34 makes possible without really
1:00:35 understanding how it works because
1:00:37 otherwise you'll you'll you're on the
1:00:39 risk of getting left behind because this
1:00:41 stuff's coming really really fast
1:00:46 yo make it rewrite the Matrix with
1:00:49 Batman and the Joker we can do that
1:00:52 um
1:00:55 all right let us go
1:00:58 let me flip the camera around so if
1:01:00 you're new here what I was showing
1:01:03 earlier was
1:01:05 um
1:01:06 if you go to poe.com and if you're a
1:01:08 subscriber
1:01:10 there's this thing called Claude instant
1:01:12 100K which is a large
1:01:15 um context window chat gbt clone
1:01:19 basically where I can put in up to 75
1:01:22 000 words so the 75 000 words I chose to
1:01:24 put in here are the the screenplay of
1:01:27 the Matrix right so so there's The
1:01:29 Matrix
1:01:31 and then I was asking it about plot and
1:01:33 characters and things like that so what
1:01:35 did you say rewrite
1:01:38 with Batman
1:01:41 wait where did it go
1:01:46 Tick Tock just blew me out of here damn
1:01:49 it
1:01:52 ah
1:01:56 oh here we go yo this is from Eric Eric
1:02:00 you're very you're very peppy tonight
1:02:02 make it rewrite the Matrix with Batman
1:02:04 and the Joker okay so I'm gonna say
1:02:06 re
1:02:08 write
1:02:11 the oak the
1:02:14 opening scene
1:02:17 um of a version of this screenplay
1:02:22 that also features
1:02:26 Batman
1:02:29 and the Joker
1:02:34 ain't better
1:02:42 waiting waiting we're waiting we're
1:02:45 waiting I'm sorry Dave I'm afraid I
1:02:47 can't do that
1:02:50 here's my attempt
1:02:54 computer screen night computer pulses on
1:02:57 dark computer screen a beating heart a
1:02:59 phone begins to ring the cursor as the
1:03:01 trace program activates yeah data
1:03:03 flashes on screen caltration initiated
1:03:05 we recognize the voices as Batman and
1:03:08 the Joker The Joker good evening Batman
1:03:10 I thought I'd give you a call Batman
1:03:12 where are you Joker
1:03:14 the trace Narrows down the phone Joker
1:03:17 you know we're around and on and on and
1:03:20 on
1:03:21 um let's see
1:03:22 um
1:03:23 what
1:03:25 uh what would be the log line
1:03:30 of this new
1:03:33 movie
1:03:35 so the log line is the little like the
1:03:37 Hollywood pitch
1:03:43 [Laughter]
1:03:49 here's a possible log line after
1:03:51 discovering he's actually living in
1:03:52 virtual reality created by artificial
1:03:54 intelligence a hacker must harness his
1:03:56 newly humanized super human abilities to
1:03:58 rescue his mentor from the clutches of
1:04:00 Agents while also confronting a deadly
1:04:02 chaos Unleashed in Gotham's most
1:04:03 notorious super villain
1:04:06 the key plot elements Neo discovering he
1:04:08 actually lives in the virtual reality
1:04:10 known as The Matrix gaining superhuman
1:04:12 abilities in The Matrix
1:04:14 his mission is to rescue Morpheus from
1:04:17 the agents the inclusion of Batman and
1:04:19 his Pursuit Of The Joker who's
1:04:20 unleashing the log line hits the major
1:04:22 plot points so I'm going to say
1:04:25 um
1:04:26 it
1:04:28 seems let's see it seems like
1:04:34 um you just
1:04:35 duct taped
1:04:38 the Batman characters
1:04:43 onto the existing plot
1:04:47 can you give me an original
1:04:53 idea where it is
1:04:56 more fully integrated
1:05:00 question mark so so one of the tricks
1:05:03 with these tools is if it gives you
1:05:04 shitty work tell it to give you shitty
1:05:06 work and tell it what you want better I
1:05:08 mean that's the kind of the role of
1:05:09 humans in these is you know you're the
1:05:12 director you're the conductor you're the
1:05:14 pilot
1:05:16 it's there to serve you neo Discovery
1:05:19 lives in a virtual reality known as The
1:05:21 Matrix however instead of facing mundane
1:05:23 agents Neo encounters some of Gotham
1:05:26 City's most infamous villains within the
1:05:28 Matrix who are working for the AI
1:05:31 so
1:05:33 let's see rescue his mentor Morpheus
1:05:36 who's been captured by the Joker
1:05:39 the Joker's testing dangerous Joy buzzer
1:05:41 devices on Morpheus that threaten to
1:05:44 drive him insane if activated
1:05:47 takes the role of Nightwing
1:05:51 so yeah so there's there's a there's a
1:05:53 version of the plot that's actually
1:05:54 decently integrated so yeah
1:05:57 you can even though you may have been
1:05:59 joking there Eric see we can go on the
1:06:02 internets with that chat GPT
1:06:04 and we can make some [ __ ] happen
1:06:07 we can rewrite some screenplays
1:06:10 all right
1:06:11 uh welcome everybody if you got
1:06:13 questions about AI popping below I'll do
1:06:15 my best to answer them could I connect
1:06:17 this via API to my bubble app yes you
1:06:21 can make all sorts of cool [ __ ] with
1:06:22 bubble and the open AI apis
1:06:26 they are in there what the heck and I am
1:06:28 having issues finding a job
1:06:31 wait what the heck and I am having well
1:06:34 I start playing with this stuff and
1:06:37 talking about it everyone that I know
1:06:38 that is actively playing with AI and
1:06:42 putting it out there that they're
1:06:43 playing with it is having
1:06:45 um
1:06:47 good luck getting jobs in fact a good
1:06:49 buddy of mine
1:06:50 this is crazy the guy that I work on a
1:06:53 newsletter with he just got a major new
1:06:56 job
1:06:57 um largely because of the stuff the
1:06:59 newsletter that we're doing on AI he's
1:07:02 like he like a massive increase for a
1:07:05 big public company is this answer a copy
1:07:08 and paste of someone else's answer
1:07:10 I don't know what you're talking about
1:07:12 possibly
1:07:13 is what answer I don't know what you're
1:07:15 talking about
1:07:16 um do you pay for the Claude instant no
1:07:18 well Claude instant 100K you have to pay
1:07:21 for Claude instant you don't and Cloud
1:07:23 instance really fast
1:07:25 Claude plus
1:07:26 gpt4 and Claude 100K you have to pay for
1:07:29 on Poe
1:07:32 um Sage Claude instant and chat gbt you
1:07:35 don't have to pay for it and the Bots
1:07:36 you don't have to pay for
1:07:39 what production work production for
1:07:41 employees I think is what he meant yeah
1:07:44 I don't if I don't quite know what his
1:07:46 question was
1:07:47 the Bing AI is better in your opinion
1:07:50 that's interesting I I
1:07:53 the the gpt4 for for me here is not as
1:07:57 good as gpt4 and open AI but if you find
1:08:00 it's better that that's the thing about
1:08:01 that's why I have these different models
1:08:03 here different people have different
1:08:05 ways of approaching this and if Bing
1:08:07 works for you like listen that's gpt4
1:08:09 connected to the internet and it's free
1:08:11 so if that's better for you [ __ ] more
1:08:13 power to you the whole point of the
1:08:15 channel is just if you haven't played
1:08:17 with these things go play with them go
1:08:19 play with them until you have your Kevin
1:08:21 McAllister moment
1:08:24 then you can decide it's all a bunch of
1:08:26 [ __ ] and you're not going to deal
1:08:27 with it but I don't think you will
1:08:29 because everyone that I know that has
1:08:31 gotten their head around this thing
1:08:33 says holy [ __ ] it can do that
1:08:37 and then they're Off to the Races even
1:08:39 Snapchat is a pretty good AI
1:08:41 everything's gonna have a pretty good AI
1:08:43 pretty soon I use mid-journey to copy
1:08:45 and paste same stuff you wrote down will
1:08:47 AI show me the picture and person wait
1:08:51 if I use
1:08:53 no so so
1:08:56 there's a couple of things
1:08:59 if you use the exact same prompt it will
1:09:02 give you four different images and then
1:09:04 if you use this exact same prompt again
1:09:06 it will give you the same four images
1:09:07 you can find this thing called the seed
1:09:10 which is a long number basically that is
1:09:15 the seed of that particular image so if
1:09:17 you use the same seed as I used it'll
1:09:20 get close but it'll still be kind of new
1:09:22 and unique the nature of how these
1:09:25 things work is they're not copying and
1:09:27 pasting elements from other people's
1:09:29 images what happens is when they
1:09:34 when they
1:09:35 learn all these images they they put
1:09:38 them into what's called a diffusion
1:09:39 model where they they turned every image
1:09:41 in the database into random noise they
1:09:43 use a noise generating algorithm and
1:09:47 then they combine a large language model
1:09:49 which understands your words and it sort
1:09:51 of points to all the images that have
1:09:54 been tagged with those words some it's
1:09:56 basically some version of that and then
1:09:58 it kind of averages all these random
1:10:00 noise images together into a single
1:10:02 random noise image and then over 30 to
1:10:05 70 steps it denoises that image into
1:10:08 some new original piece of work so every
1:10:11 time you render it unless you use the
1:10:14 same seed it's going to give you a new
1:10:16 seed a new randomized thing and a new
1:10:19 image
1:10:21 um and that's that's one of the
1:10:22 challenges of it
1:10:23 um is that it can be hard like if you
1:10:26 got something that you love it can be
1:10:27 hard to replicate that same thing in
1:10:30 different areas using things like
1:10:34 um zoom out and generative fill within
1:10:36 Photoshop you can improve that there's a
1:10:39 thing with stable diffusion called
1:10:41 control net that can control things a
1:10:43 bit better but that's one of the things
1:10:46 it's not great at is is consistency and
1:10:49 replicability
1:10:50 but in terms of generating new stuff
1:10:52 it's [ __ ] brilliant brilliant
1:10:56 brilliant Brown yeah
1:10:59 all right uh you don't need to ask
1:11:01 questions three or four times I'm very
1:11:03 far behind on the questions so uh
1:11:07 remind me where I find prompt perfect
1:11:10 please okay so let's go show let's go
1:11:14 show some prompts and I will show you
1:11:18 prompt perfect okay so we're gonna go to
1:11:19 chat GPT you can make money
1:11:22 okay so if you've not used chat GPT this
1:11:26 is your prompt history
1:11:28 um if you pay the 20 bucks you get this
1:11:30 little yellow icon and that's
1:11:33 um chat EBT plus which then gives you
1:11:36 access to gpt4 with like the normal
1:11:40 version the Internet connected version
1:11:42 and plugins and I'll show you these in a
1:11:44 second
1:11:45 if you pay your 20 bucks and you don't
1:11:47 see these options you actually have to
1:11:49 turn them on because I don't know why
1:11:51 they're doing that because they're
1:11:52 [ __ ]
1:11:53 I I don't know I think it's a bad
1:11:55 decision right now but it's probably
1:11:56 just a processor thing
1:11:59 um you go to your settings and then in
1:12:01 your settings there's a option called
1:12:03 beta features and then you turn on
1:12:06 browse with Bing and plugins and so
1:12:08 that's important you got to turn that
1:12:10 stuff on
1:12:11 and then once it's turned on we're going
1:12:13 to go to plugins
1:12:14 and now you have this little drop down
1:12:16 menu and the drop down menu is here's
1:12:18 all the plugins that I've got installed
1:12:21 which basically these are the ones I
1:12:23 pulled out of the the plug-in store that
1:12:25 I said these are these are the ones that
1:12:27 are interesting in any given chat
1:12:29 session I can enable up to three of them
1:12:32 so let's say I want to do something
1:12:33 about travel or food you know I would
1:12:36 turn on my travel or my food
1:12:39 plug-ins and then I would have kind of
1:12:42 like a little customized application
1:12:43 that can do chat GPT stuff but also can
1:12:47 use the plugins and do whatever they do
1:12:49 so the one that's called prompt perfect
1:12:51 is a plug-in so you just turn that on
1:12:55 and then the way it works is you you
1:12:57 type the word perfect as your first word
1:12:59 of your prompt
1:13:00 and then you type a crappy prompt and it
1:13:02 will rewrite it for you so let's say
1:13:04 um
1:13:05 uh let's say create a meal plan
1:13:12 for me for the week that isn't boring
1:13:17 uh but
1:13:19 is keto friendly all right
1:13:24 that's kind of a crappy prompt because
1:13:25 it's generic and whatever and so because
1:13:28 I typed the word perfect in front of it
1:13:30 it fires off prompt perfect
1:13:32 and now it took my crappy prompt and it
1:13:35 rewrote it please design a weekly meal
1:13:37 plan that adheres to the keto diet
1:13:39 ensuring it is both interesting and
1:13:41 diverse with a focus on providing a
1:13:43 detailed and comprehensive response cool
1:13:46 and then so it now gives me that
1:13:54 and actually let me let me show you
1:13:56 another thing I'm going to stop
1:13:57 generating here
1:13:59 I'm going to go copy that initial prompt
1:14:03 and I'm gonna go turn on two plugins
1:14:05 because this this kind of cool to to
1:14:07 watch this I'm going back to plugins so
1:14:09 I've got prompt perfect there
1:14:12 and I'm going to turn on
1:14:17 instacart because we were doing a meal
1:14:20 plan so now I'm gonna so now I've got
1:14:22 prompt perfect and instacart so I'm
1:14:23 going to take that same prompt I start
1:14:25 with the word perfect create a create a
1:14:27 meal plan
1:14:29 so it should fire off prompt perfect
1:14:33 it did
1:14:37 well there it rewrote
1:14:40 The Prompt it's not the same as it did
1:14:42 before this is
1:14:43 specifically designed to follow the
1:14:45 ketogenic diet for an entire week the
1:14:47 plan should prioritize variety and avoid
1:14:51 monotony
1:14:52 so there's day one day two day three day
1:14:54 four day five day six
1:14:57 day seven
1:14:59 and
1:15:00 notice what it did here at the end
1:15:03 would you like me to create a shopping
1:15:05 list for these meals using the instacart
1:15:07 plug-in so all I have to do is say yes
1:15:11 it knows the plug-ins in there it knows
1:15:14 it can create me a shopping list because
1:15:16 of how that plug-in was described
1:15:17 internally
1:15:18 and now it will probably
1:15:22 fire off the instacart plug-in it should
1:15:28 boom there it goes
1:15:31 and now it's going to give me a link to
1:15:33 that shopping list and I could have
1:15:35 those groceries delivered to my door
1:15:37 tonight
1:15:47 champ
1:15:50 come here
1:15:53 stop
1:15:54 [Applause]
1:15:56 did it do it all right there's your
1:15:58 instacart shopping list so if I go click
1:16:00 on that Bang
1:16:09 and there's all my stuff
1:16:12 right so from a single prompt it made me
1:16:16 a whole meal plan
1:16:17 there's all the ingredients
1:16:19 pop those in my cart have them delivered
1:16:22 Bing Bang Boom
1:16:25 Off to the Races isn't that nuts
1:16:28 nuts
1:16:29 freaking nuts
1:16:31 and and you know the future of these
1:16:33 plugins the plugins right now are really
1:16:35 kind of crappy and janky and the
1:16:37 interface is horrific
1:16:40 um but
1:16:42 here's how I like to think of all this
1:16:44 generative AI stuff right now
1:16:48 whatever you play with right now all
1:16:50 this stuff is as bad as it's ever going
1:16:52 to be today
1:16:53 so if you think about the amount of time
1:16:55 it would have taken you
1:16:57 if you wanted to go Google a meal plan
1:17:00 planner
1:17:01 and find a website and sign up for it
1:17:03 and most of them charge you money and
1:17:05 then
1:17:06 go make yourself a list and that
1:17:08 whatever however long that took what we
1:17:10 just did in three minutes
1:17:12 with those plugins this stuff's gonna
1:17:14 get way more transparent and way more
1:17:16 automated where you'll literally be able
1:17:19 to just say to your agent
1:17:21 um make me a meal plan that's different
1:17:23 than what we had last week it'll
1:17:25 remember what you had last week and have
1:17:27 the stuff delivered and have it be there
1:17:30 within a half hour of when I get home
1:17:32 and and it'll know that kind of stuff so
1:17:36 I'm here to tell you how stinking
1:17:38 adorable you are well thank you very
1:17:41 much
1:17:43 uh I appreciate that oh it's a little
1:17:47 warm and woman is it me
1:17:50 glare
1:17:52 ah okay first time you save yourself
1:17:56 countless hours of work you'll fall in
1:17:58 love with AI That's exactly it man it's
1:18:00 like
1:18:02 the the best way I can describe the
1:18:05 impact of Chachi BT and and mid-journey
1:18:08 like all of these things is that is that
1:18:11 they're just they're Humanity amplifiers
1:18:13 they're like Talent amplifiers like
1:18:15 whatever you're good at it will make you
1:18:18 way better at that and then all the
1:18:20 stuff that you are not good at that you
1:18:23 always wanted to be good at like you
1:18:25 have a desire to be good at something
1:18:26 but you're really shitty at it like me
1:18:27 withdrawing
1:18:29 now you can be good at it or maybe
1:18:31 you're not great at writing okay it
1:18:33 doesn't matter now you can be
1:18:36 um that's going to be remarkable now you
1:18:39 want to hear a downside to it
1:18:41 you know how when you get one of those
1:18:43 scam emails one of the ways you can tell
1:18:45 they're a scam email is because they're
1:18:47 English is really bad
1:18:50 that's going away
1:18:52 people with really bad English are going
1:18:54 to get really good at those letters
1:18:57 that's coming
1:18:58 uh can't you use AI to create it for you
1:19:01 I don't know what that was in reference
1:19:02 to but probably yes
1:19:04 yes use Reddit for prompt libraries yeah
1:19:06 that's a good idea yeah that's that's
1:19:08 another thing thank you for the little
1:19:09 hearty head
1:19:10 um
1:19:11 reddit's a really good source if you
1:19:13 want to kind of geek out on stuff or you
1:19:15 know dig into open source large language
1:19:17 models or prompt libraries or what are
1:19:19 the best prompt tools or who are the
1:19:21 best prompt Engineers Reddit all the
1:19:24 subreddits they're they're geeking out
1:19:26 on this [ __ ] so an important thing to
1:19:29 know about chat gbt it went from zero to
1:19:32 100 million users in six weeks it was
1:19:35 the fastest adoption of technology and
1:19:37 history so because of that
1:19:40 the entire developer Community is just
1:19:42 kind of laser focused on all this stuff
1:19:44 right now so the the subreddits within
1:19:47 within Reddit around all this stuff are
1:19:50 on fire so you want to learn about the
1:19:53 latest you know features of stable
1:19:55 diffusion that the community is putting
1:19:57 together and launching it's all in there
1:19:58 you want to figure out about the video
1:20:00 Generation stuff and who's doing
1:20:02 interesting stuff it's all there so
1:20:03 that's that's a really good call
1:20:06 run through where to find prompt perfect
1:20:09 again oh I just did that cool
1:20:12 um what app do you use for chat GPT I
1:20:16 don't I just use Safari or Chrome on my
1:20:19 mobile device and I just point it there
1:20:21 and make it a bookmark on my mobile
1:20:23 device although
1:20:24 I I heard from Rachel Woods uh he was a
1:20:28 tick tocker here
1:20:30 um today that the the official chat GPT
1:20:33 app is now connected to the internet uh
1:20:36 but but it's for plus users only so if
1:20:38 you pay the 20 bucks the iOS version of
1:20:43 the official openai app
1:20:45 for chat GPT is now connected to the
1:20:48 internet but I just use the website
1:20:50 because it's it's really good it's a
1:20:51 very simple website what's a prompt okay
1:20:54 so the prompt is just basically the
1:20:56 query that you type into chat GPT to
1:20:59 prompt it to do something so
1:21:02 um
1:21:03 explain
1:21:05 quantum mechanics to me
1:21:08 um
1:21:09 as if you were Shakespeare that's a
1:21:11 prompt right and you can get really
1:21:14 clever with them
1:21:15 um you can put really basic things in
1:21:18 if you put in things that are too basic
1:21:20 or too concise it doesn't have enough
1:21:22 context to kind of narrow
1:21:25 what you really want and it'll tend to
1:21:27 give you generic kind of bullshitty kind
1:21:29 of answers
1:21:31 um so the whole idea of prompt
1:21:32 engineering is really understanding you
1:21:35 know
1:21:37 how do I kind of logically put together
1:21:39 sort of language Lego blocks to tell
1:21:42 this thing what I actually want
1:21:45 and and the trick for me is don't think
1:21:47 of it like you've got to write the
1:21:49 perfect prompt at once think of it more
1:21:52 like a conversation you're going back
1:21:53 and forth with this tool saying I want
1:21:56 you to act like this kind of expert and
1:21:58 then I'm going to ask you questions
1:21:59 about this and then I want you to help
1:22:00 me with that can you do that yeah I can
1:22:02 do that Ah that's awesome you're the
1:22:04 best no you're the best oh Bubba I love
1:22:07 you Bubba I love you Bubba think of it
1:22:10 like your Bubba and it'll get better so
1:22:12 don't I think of it more like prompt
1:22:15 crafting than prompt engineering but
1:22:16 depending on how geeky you are it is
1:22:19 definitely thinking like a programmer
1:22:21 but using your native language
1:22:24 Googling on chat EBT has been the
1:22:26 biggest challenge for me having to
1:22:27 switch that mindset yeah exactly that
1:22:30 that that idea of it it's just not
1:22:33 Google it looks like a Google search
1:22:35 engine but that's not what it is and and
1:22:37 that that has taken me a while to get my
1:22:40 head around as well which prompt
1:22:42 formatter do you suggest for mid-journey
1:22:44 I don't know I there's there's a bunch
1:22:46 of them out there those those
1:22:48 um
1:22:49 those Bots that I used on Poe poe.com
1:22:53 mid-journey and po.com slash MJ captain
1:22:58 um are both good
1:23:00 um
1:23:02 the other thing that you could do if you
1:23:03 if you want to practice your prompt
1:23:05 crafting is go find there's websites out
1:23:08 there that will sort of give you the the
1:23:11 um
1:23:12 the syntax of a mid-journey prompt it'll
1:23:15 like here's all the options you can put
1:23:17 in right here's the modifiers and the
1:23:18 illustration Styles and the the
1:23:21 photography whatever bits and you know
1:23:25 all sorts of different elements you
1:23:27 could take that documentation paste it
1:23:30 into chat GPT say you know I want you to
1:23:33 read this documentation on mid-journey
1:23:35 and then I'm going to put in a prompt
1:23:38 and I want you to give me back a
1:23:40 well-crafted mid-journey prompt so you
1:23:42 can actually create your own bot to do
1:23:44 that
1:23:45 um which which would be good practice to
1:23:48 just practice your prompting thank you
1:23:50 I'm Gen X I try give it as much
1:23:53 descriptive info for specific requests
1:23:56 I'm getting better yeah Michelle Donald
1:23:58 that's that's
1:24:00 that's kind of the trick is just keep
1:24:02 experimenting with it just keep you know
1:24:06 when it gives you five shitty answers in
1:24:09 a row don't give up on it just keep
1:24:11 telling it why it's shitty like
1:24:13 that's the thing again like I equate
1:24:16 this very much to like working with an
1:24:17 intern you know you get an intern and
1:24:19 you're all excited and you're like
1:24:21 you're like okay I want you to come up
1:24:23 with a marketing plan and they go what's
1:24:25 marketing and you're like oh God oh God
1:24:28 okay okay here's what marketing is so
1:24:32 what's the plan huh you know and you
1:24:35 spend that first you know four hours or
1:24:36 four days or four weeks with the intern
1:24:38 trying to just get them to understand
1:24:39 the base Concepts so you can then have
1:24:42 them go off and do some research
1:24:44 um it's very much like that it's much
1:24:46 more compressed time-wise but but giving
1:24:48 it that context is really the key to to
1:24:51 making it better
1:24:52 you just gave me my next job good I
1:24:54 don't know what I did but oh is that the
1:24:56 stand-up comedy yeah start a tick tock
1:24:58 Channel Eric I think you'll be good AI
1:25:00 writes grants for you yes
1:25:02 um that means you could
1:25:04 you get to spend the money build your
1:25:07 future exactly you can make money with
1:25:09 chat GPT just have it write you a grant
1:25:12 submit that Grant get paid for that
1:25:13 Grant and go spend the money exactly
1:25:16 let's hope
1:25:18 um a cyber SEC joke was tongue-in-cheek
1:25:20 I figured it was but it's still not far
1:25:22 off from where I think it's headed I'll
1:25:24 get chat GPT to write my jokes one of
1:25:26 the in at imprompts.chat one of the
1:25:29 early things is act as a stand-up comic
1:25:31 and I've had it act as a stand-up comic
1:25:33 and it's pretty shitty at writing jokes
1:25:35 it understands
1:25:37 it understands the mechanics of a joke
1:25:41 but it doesn't quite understand jokes so
1:25:43 it will occasionally stumble on
1:25:46 something that's funny
1:25:48 I get Chachi PD to write my jokes yep uh
1:25:51 so a question go on
1:25:54 um fake news Okay biggest social
1:25:57 challenge to me with AI is that people
1:26:01 anthropomorphize AI That's gonna get
1:26:03 worse Eric so I I think that um
1:26:07 so right now
1:26:08 Chachi EBT just kind of looks like a
1:26:10 Google search form
1:26:12 um I think what it's going to start to
1:26:13 look like is
1:26:15 like your little buddy like an agent
1:26:17 like an agent that is either your little
1:26:19 friend it's
1:26:21 I people joke about you know well maybe
1:26:24 they should bring back clippy remember
1:26:25 clippy from Microsoft if you're if
1:26:27 you're a Gen X or or Millennial you know
1:26:29 you know clipping clippy was this failed
1:26:32 AI sort of recommendation engine living
1:26:35 within Windows it looks like you're
1:26:37 trying to write a business letter pring
1:26:40 want me to help you with that it was
1:26:41 just [ __ ] awful
1:26:43 um
1:26:44 I think there's going to be effectively
1:26:46 a more sophisticated version of that
1:26:48 that I'm sure Apple's gonna have an
1:26:50 amazing one I'm sure there's going to be
1:26:52 something within Windows that's like
1:26:54 that I think that where chat GPT goes or
1:26:57 we're pogos where some of these things
1:27:00 go is that right now like Poe has all
1:27:02 these Bots and all the Bots are just
1:27:04 kind of these type in the thing
1:27:06 I could imagine in the future those Bots
1:27:08 become like 3D avatars and they actually
1:27:12 talk to you and I can have one that
1:27:14 talks like Shakespeare and I can make my
1:27:16 Shakespeare bot look like Shakespeare
1:27:18 talk like Shakespeare and you can
1:27:20 actually interact with it like you're
1:27:21 having a conversation so I actually I I
1:27:23 think the answer
1:27:25 anthropomorphization is gonna get
1:27:28 um is gonna continue
1:27:31 um
1:27:32 and I think virtual companionship is you
1:27:35 know they're sort of the
1:27:38 you know the sort of uncomfortable you
1:27:41 know creepy version of that
1:27:45 um but I think there's also a really
1:27:46 healthy version of that where you could
1:27:48 have
1:27:49 virtual physical therapists that are
1:27:51 available 24 7 or you know virtual like
1:27:54 assistant
1:27:56 um uh you know therapist where you have
1:27:58 your real therapist but you have kind of
1:28:00 a virtual version of them that you can
1:28:02 talk through things and if there's
1:28:04 something serious that needs to get
1:28:05 escalated to the real person the the bot
1:28:08 will be good enough to do that
1:28:11 um Harvard today
1:28:13 announced that they're putting um next
1:28:17 fall their their computer science 50
1:28:20 they're like seminal computer
1:28:22 programming class at Harvard the
1:28:25 teaching assistant's gonna be an AI bot
1:28:27 so it's going to check your code it's
1:28:29 going to help you debug it's going to
1:28:31 help you document
1:28:32 it's available 24 7. so yeah so it's
1:28:36 like like this stuff is coming where
1:28:38 people are going to be interacting with
1:28:40 these things
1:28:41 very much like
1:28:43 you know companions
1:28:45 and they'll say please and thank you
1:28:47 like I say please and thank you when I'm
1:28:49 using chat GPT all the time
1:28:52 um I need to stop dominating your
1:28:54 comments with snarky remarks and pay
1:28:55 attention well then stop just then do it
1:29:02 then Tick Tock it might have some advice
1:29:05 all right watch I'm gonna be at work
1:29:07 tomorrow like oh it's like because I go
1:29:10 uh
1:29:14 does AI
1:29:16 have biases based on information like
1:29:19 religion or race yeah in fact so the
1:29:21 when I first started doing these lives I
1:29:23 was explaining a lot about biases and
1:29:26 how they work and how biases get in the
1:29:28 models and how to get them out of there
1:29:30 and I got banned twice in a row when I
1:29:34 started talking about bias saw I'm not
1:29:36 going to talk about it because they
1:29:37 kicked me off for seven days for I like
1:29:39 I don't think I said anything but it was
1:29:41 just I guess because I was talking about
1:29:42 bias it's like oh this is political
1:29:44 discourse he's saying bad things about
1:29:48 bad people or bad things about good
1:29:50 people or whatever
1:29:51 or someone reported I don't know what it
1:29:53 was but it doesn't matter but but
1:29:55 um
1:29:56 yeah I mean it's so the the way open AI
1:30:01 trains fine-tunes their models so they
1:30:04 train it on kind of all of the internet
1:30:05 so the internet's going to have whatever
1:30:07 biases it has so it's been trained on
1:30:09 all that [ __ ]
1:30:11 and then the way they train it is with
1:30:12 human reinforcement learning so it's
1:30:14 basically people going yeah that's a
1:30:15 good answer that's a bad answer that's a
1:30:17 good answer that's that answer well
1:30:18 people are biased right so that
1:30:19 introduces bias so
1:30:21 um there's there's a company called
1:30:22 anthropic so if you go to poe.com three
1:30:25 of the models there are called Claude
1:30:27 different variations of Claude that's
1:30:29 for a company called anthropic and they
1:30:32 have
1:30:32 this thing called the Constitutional
1:30:34 learning model so instead of humans
1:30:36 going thumbs up thumbs down
1:30:38 they've got a constitution like you know
1:30:41 um you know Do no harm increased
1:30:43 Prosperity
1:30:45 um whatever you know don't cause pain
1:30:46 whatever whatever the Constitution might
1:30:48 be
1:30:49 and then the the way the system works in
1:30:52 a dramatically simplified version is
1:30:54 when it generates an answer it Compares
1:30:56 that answer against the Constitution and
1:30:58 it will iterate on the answer until it
1:31:00 can pass the Constitution right so so
1:31:02 systems like that are going to start to
1:31:05 even out the bias and I see here's the
1:31:08 thing about bias is that
1:31:10 everyone's got biases so the the model
1:31:13 that doesn't look biased to me is going
1:31:15 to look completely biased to you so I
1:31:17 think what we'll see in the future is is
1:31:19 you'll have multiple models that you can
1:31:21 choose from and you can kind of opt into
1:31:22 the value set of your choice that's
1:31:25 that's my guess I don't know that it's
1:31:27 going there but there's not a single
1:31:30 perfect model because human beings are
1:31:32 [ __ ] up and biased and that's just the
1:31:33 way it's just the way it is
1:31:36 um
1:31:37 to access prompt perfect do I simply put
1:31:39 perfect before my prompt yeah you go to
1:31:41 you go to gpt4 with plugins you enable
1:31:46 prompt perfect
1:31:48 plug-in and then just type the word
1:31:50 perfect in front of your prompt and
1:31:53 that's all that's all all she takes
1:32:01 all right most of the data is probably
1:32:04 okay yeah
1:32:08 well I don't think that most of the data
1:32:10 is probably one race or another like
1:32:12 like if if you're if you're talking
1:32:14 about the training the people that do
1:32:16 the training for chat GPT because they
1:32:19 were in San Francisco the the that's
1:32:21 probably true of the people doing the
1:32:24 human reinforcement learning for open AI
1:32:27 but the
1:32:29 um
1:32:30 openai was trained on like all of the
1:32:32 public internet so not just America like
1:32:35 all of the public internet so to the
1:32:37 extent that all of the data on the
1:32:39 public internet is biased or skewed in
1:32:43 you know toward One race or another
1:32:44 that's going to be in there but but I
1:32:47 wouldn't it I wouldn't characterize it
1:32:49 quite so broadly as is what you put
1:32:51 there and I don't want to say it out
1:32:52 loud because I don't want to get the
1:32:53 channel shut down again
1:32:55 because I'm not saying anything I'm just
1:32:57 trying to explain what the [ __ ] is going
1:32:58 on I've seen a lot of mid-journey images
1:33:01 featuring people of color yet the it's
1:33:05 they're getting better and you but but
1:33:08 you have to really prompt for it
1:33:09 um the the people of color thing in the
1:33:13 in the stable diffusion and in mid
1:33:15 journey and probably to ex to an extent
1:33:18 with Adobe but I bet Adobe
1:33:20 has been adobe's been way more conscious
1:33:23 of what's in their training data so I
1:33:26 would imagine that adobe's worked really
1:33:27 hard to balance that so that would that
1:33:30 would be an interesting experiment is
1:33:31 compare mid-journey stable diffusion and
1:33:33 Adobe Firefly
1:33:36 for diversity
1:33:39 um I have plus and it's on I had
1:33:40 forgotten where prompt perfect was so
1:33:42 thank you for answering of course Silver
1:33:44 Fox that's why we're here
1:33:47 um have fun with that also go to uh
1:33:50 poe.com and go to the uh the mid Journey
1:33:52 prompt writers actually just go go
1:33:54 explore all the prompt all the the PO
1:33:56 Bots the the uh the chat bots in po
1:34:00 there's there's some really good ones in
1:34:02 there
1:34:04 great job you've been live for 90
1:34:06 minutes that's why you're shiny and your
1:34:09 voice doesn't work anymore I can't form
1:34:11 words thank you for the roses that is a
1:34:14 Rose Garden you've given me there I I
1:34:16 appreciate the hell out of that
1:34:18 wood
1:34:21 what is what is tricking me out with
1:34:23 roses I they're everywhere thank you
1:34:26 very much I appreciate that
1:34:28 sorry old question oh I got lost uh Tick
1:34:31 Tock just jumped me to the bottom so I'm
1:34:33 going to scroll up a little bit and grab
1:34:34 some questions here
1:34:37 did you see the documentary on him I
1:34:40 don't know what that was oh my God
1:34:43 it's great worth a watch I love that you
1:34:45 guys have conversations with yourselves
1:34:47 in the comments that's actually super
1:34:49 cool wow some more roses thank you so
1:34:50 much I really appreciate it
1:34:53 um
1:34:54 it's a human amplifier well said yeah
1:34:56 yeah it's
1:34:58 yeah it's yeah
1:35:01 I feel like for the last two days my
1:35:04 um
1:35:06 My my what do you call it uh my my uh
1:35:13 uh grasp of the English language has
1:35:16 been suffering like I I just I words not
1:35:19 so good what's that Steve Martin joke
1:35:21 some people have a way with words other
1:35:23 people oh not halfway I guess you'd say
1:35:28 that's what I felt like for the past two
1:35:30 days not halfway hey Kyle do you work
1:35:33 for open Ai No if not what's the reason
1:35:35 that you talk about chat GPT every day
1:35:38 because okay
1:35:41 so a couple of things I'm not a machine
1:35:43 learning guy I'm an entrepreneur
1:35:47 I was there
1:35:50 I was I was working in New York City
1:35:52 when the world wide web web came onto
1:35:54 the scene and I started one of the first
1:35:56 digital agencies I had a big Epiphany
1:35:58 about
1:35:59 the internet and the World Wide Web once
1:36:01 it came along
1:36:03 and what Chachi PT represents is is
1:36:07 essentially The Identical
1:36:10 um
1:36:11 inflection point as as what the World
1:36:14 Wide Web did for the internet so let me
1:36:16 explain a little bit
1:36:18 the internet had been around for decades
1:36:20 right it was you know invented whenever
1:36:22 it was invented 50s 60s and you know uh
1:36:26 and then and then in the 70s they
1:36:28 started connecting all this [ __ ]
1:36:30 and
1:36:32 um and then
1:36:33 it was largely you know researchers and
1:36:37 um
1:36:38 scientists and universities and research
1:36:42 paper you know repositories for decades
1:36:46 so if you weren't really a scientist or
1:36:48 a com you know an engineer of some sort
1:36:50 you really didn't have much use for the
1:36:52 internet you could use it but you had it
1:36:53 was all command line and then Tim
1:36:55 berners-lee invents the world wide web
1:36:57 and it allows us to create these web
1:37:00 pages with these blue underlined words
1:37:02 on them you can click on that and that
1:37:04 will jump you over to another page
1:37:06 and
1:37:08 that changed everything right so all of
1:37:10 a sudden the internet went from being
1:37:11 for scientists and researchers to anyone
1:37:14 that could learn a couple of HTML tags
1:37:16 could put up a page and Link that to
1:37:17 some other page and that changed
1:37:19 everything it transformed
1:37:21 you know it collapsed geography it it
1:37:24 you know collapsed time you could just
1:37:26 instantly go from one piece of
1:37:28 information to the other
1:37:30 um you know instant Communications all
1:37:32 that sort of stuff
1:37:34 radically transformed how we do business
1:37:36 how we communicate how we connect how we
1:37:39 date how we everything shop
1:37:44 so I talk about chat GPT all the time
1:37:46 because
1:37:48 machine learning and AI has been around
1:37:50 for decades and and you know certainly
1:37:53 in the past 10 to 20 years it's been
1:37:55 increasingly powerful and we hear it
1:37:57 about a lot and it's it's AI is is in
1:38:00 almost all the software we use like all
1:38:02 of the flip and Tick Tock filters that
1:38:04 make you pretty and give you a head of
1:38:06 hair and all that sort of [ __ ] that's
1:38:08 all AI That's AI how that's positioned
1:38:10 and tracked is all AI
1:38:12 chat GPT represents
1:38:17 putting that magic in the hands of real
1:38:20 people regular people
1:38:22 I have a degree in acting I've started a
1:38:25 bunch of companies I love technology but
1:38:26 I'm not a technologist
1:38:29 I like understanding how technology
1:38:31 works and then figuring out how that's
1:38:33 going to change the world that's what
1:38:35 I'm passionate about so for me chat gbt
1:38:38 is the seminal piece of technology that
1:38:41 broke the door open and and here let me
1:38:44 put another thing in context for you
1:38:47 the World Wide Web to get to 100 million
1:38:49 users took six years it was six years
1:38:52 from the Tim bernersley inventing until
1:38:54 we got to about 100 million users chat
1:38:56 GPT six weeks
1:39:00 fastest adoption of technology in
1:39:03 history the other thing about chat GPT
1:39:04 is this and why I talk about it all the
1:39:06 time
1:39:08 if they didn't improve it if it stopped
1:39:11 today like you know Elon Musk signed
1:39:13 that letter you know they should stop
1:39:15 innovating in AI for six months and let
1:39:18 me catch up
1:39:19 I think that's what that was about but
1:39:21 whatever
1:39:22 if they stop development of AI today I
1:39:25 feel like I could I could personally
1:39:27 spend the next decade
1:39:31 figuring out how to use chat GPT
1:39:33 it is it is so deep it is so profoundly
1:39:37 powerful
1:39:38 like I haven't scratched the surface of
1:39:40 it so
1:39:42 I just think it's that important and and
1:39:44 it's also
1:39:47 you don't need to have any computer
1:39:48 skills to use it you go to that web
1:39:52 address and there's like a Google box
1:39:54 there and you just type [ __ ] into it
1:39:57 tell me which plants like it's you know
1:40:01 whatever it is it's the end of June in
1:40:04 Denver Colorado what plants should I be
1:40:06 planting in my vegetable garden
1:40:08 boom out comes your answer
1:40:12 would I've taken the time to go Google
1:40:14 that no because
1:40:16 it would have taken me too long and
1:40:18 there'd have been too many ads and I
1:40:19 don't really give all that much of a
1:40:20 [ __ ] about it oh but now I got this
1:40:22 thing they'll just give me the answer
1:40:23 boom there it is
1:40:25 it's I I don't know because it's
1:40:27 profound
1:40:29 no hate just curious I love your lives
1:40:31 thank you yeah no no it's like it's a
1:40:34 good question and it's like I've been
1:40:36 thinking about that a lot actually like
1:40:38 at what when at what point am I going to
1:40:40 change these URLs like like there's
1:40:43 probably a better
1:40:45 prompting library to put up here than
1:40:47 this
1:40:48 but this is actually just a pretty good
1:40:50 primer it's got a decent primer in it
1:40:52 and and the Persona thing where it says
1:40:55 act like a screenwriter act like a
1:40:57 stand-up comic that's pretty good
1:40:59 this is just a list of list of tools and
1:41:02 I don't really follow all the tools
1:41:04 anymore so there's probably five or six
1:41:06 of those at some point I will probably
1:41:08 swap these out for something else I
1:41:10 don't know when like it still feels
1:41:12 relevant to me and I've got ADHD so I'll
1:41:16 get bored with this [ __ ]
1:41:18 I'm an artist and I'm disappointed how
1:41:20 cartoony most images I've tried I want
1:41:22 something painterly
1:41:23 okay so RM Walter I'm gonna go show you
1:41:27 something
1:41:29 because you can absolutely get there you
1:41:32 can absolutely get to painterly
1:41:36 um
1:41:39 [Laughter]
1:41:41 [Music]
1:41:42 go to
1:41:45 open art
1:41:51 Dot
1:41:52 is it AI yes
1:41:56 go to
1:41:57 openart.ai promptbook
1:42:01 openart dot a I slash promptbook
1:42:05 and it's going to take you to this
1:42:07 document that's a stable diffusion
1:42:09 prompt book
1:42:11 but I think this will help a lot because
1:42:16 these concepts are transfer transferable
1:42:19 to to other things like how you actually
1:42:21 you know put the the codes into the
1:42:23 tools might be slightly different but
1:42:24 the basic concepts are the same
1:42:26 so so there's a bunch of different
1:42:28 sections in it it's a 96 page document
1:42:30 so they talk in general about prompt
1:42:33 engineering they talk about your prompt
1:42:35 format like do you want a photo or a
1:42:37 painting what's the subject is it a
1:42:39 person an animal perhaps a landscape
1:42:41 what details do you want to add special
1:42:44 lighting environment color scheme point
1:42:46 of view is there a specific art style so
1:42:48 to your point if you want it to be
1:42:51 um you know claymation
1:42:54 sculpture you can put that if you want
1:42:56 it to be you know highly textured oil
1:42:59 paints you can put that
1:43:02 is there a specific photo type macro
1:43:04 telephoto so so it's starting to explain
1:43:07 to you that there are different elements
1:43:09 different structural elements you can
1:43:10 put into a prompt and then and then this
1:43:12 document just basically continues to
1:43:15 unpack all of that stuff so this section
1:43:18 on modifiers like here's photography
1:43:20 modifiers
1:43:21 shot type style subject lighting
1:43:26 context indoor outdoor at night in the
1:43:29 park studio right
1:43:31 um
1:43:32 it shows you little examples of if you
1:43:34 use the word Polaroid it understands
1:43:36 what that is if you put tilt shift or
1:43:38 product shot like these these systems
1:43:42 have an incredible depth of
1:43:44 understanding of art Concepts like the
1:43:48 the this document is like a good art
1:43:50 history
1:43:52 um it's like it's like it's like art art
1:43:54 school in a box
1:43:56 um lighting nostalgic Sunrise art medium
1:43:58 so this gets to what what we're talking
1:44:00 about with you right chalk graffiti
1:44:03 watercolor oil painting
1:44:05 um
1:44:06 artists so talking about specific
1:44:09 artists so if you've got specific
1:44:10 artists that have got a style that you
1:44:12 like you know you can invoke their names
1:44:15 if they're modern artists I would
1:44:18 encourage you to avoid using modern
1:44:21 artist names but certainly for the
1:44:23 classics
1:44:25 um
1:44:26 you know certain Styles you can invoke
1:44:28 them and it it is remarkably good at at
1:44:32 pulling that sort of stuff up so again
1:44:35 this is at
1:44:37 openart.ai promptbook
1:44:40 and that should help I hope that helps
1:44:48 all right I'm gonna answer a few more
1:44:50 questions I'm getting a little crispy
1:44:53 I needs my money someone set me up
1:44:56 you can make money with Chachi day
1:45:00 we all need money I need money
1:45:04 give me the money money
1:45:11 I never like clippy clippy 2.0 clippy
1:45:14 was annoying
1:45:15 [Laughter]
1:45:17 I remember clippy so annoying I know
1:45:21 I read today clippy was back I think
1:45:23 someone I think someone created a clippy
1:45:26 chat bot
1:45:28 um I don't think it's back within
1:45:29 Microsoft it would be a brilliant move I
1:45:31 mean quite honestly if Microsoft
1:45:33 relaunched clippy as not sucking and
1:45:37 quite honestly I think Apple's got a
1:45:39 relaunch Siri as not sucking
1:45:42 um
1:45:43 I think I like the Nissan The Nostalgia
1:45:46 thing and and the sort of
1:45:47 tongue-in-cheek marketing Buzz that
1:45:50 Microsoft would get by like leaning into
1:45:52 clippy and saying clippy was you know
1:45:55 was before it's time like that would be
1:45:59 pretty [ __ ] funny and then if clippy
1:46:01 were gpt4 you know super master clippy
1:46:04 that actually didn't suck cool
1:46:09 uh yes yes
1:46:12 does AI have a code tester built in yes
1:46:16 give it some code tell it to debug it
1:46:18 for you it will tell it to document it
1:46:20 for you it will tell it to write you
1:46:22 code it will
1:46:24 tell it to take code from this language
1:46:26 and refactor it into that language it
1:46:28 will now is it gonna be perfect every
1:46:31 time nope because here's the thing about
1:46:33 the way the large language models work
1:46:35 it doesn't actually understand what it's
1:46:38 doing
1:46:39 what it's doing is a probability
1:46:41 exercise you write a prompt it comes up
1:46:44 with a mathematical weighting that then
1:46:46 then says what's the next most probable
1:46:48 word after this long prompt and then
1:46:51 what's the next most probable word and
1:46:52 and so on and so when it writes all that
1:46:55 stuff it's doing it word by word as a
1:46:56 probability engine it's got no concept
1:46:59 of what the hell it just wrote so it may
1:47:01 give you code that looks right but is
1:47:03 actually wrong so it's good if you know
1:47:05 code
1:47:07 but if you know code you can
1:47:09 dramatically increase your productivity
1:47:10 with it it's it's remarkable
1:47:13 I hope it doesn't get so good that it
1:47:15 replaces my dog possible garbage in
1:47:17 garbage out correct
1:47:19 um well AI should take that Harvard bias
1:47:22 test it should
1:47:24 sorry old questions and it looks like
1:47:26 all right got a few more questions at
1:47:28 the bottom here and then I'm gonna
1:47:30 take off thank you again for all those
1:47:32 Roses woodsy appreciate that
1:47:39 is chat GPT constantly upgrading
1:47:42 changing yes on the Fly no so
1:47:46 when you use chatgpt if you just go to
1:47:48 chat GPT 3.5 or gpt4 not connected to
1:47:52 the internet
1:47:54 and you ask it for something current
1:47:56 it'll basically say I'm sorry I only
1:47:59 know things up to September 2021 and
1:48:02 then you're like well that's dumb
1:48:05 if you use Bing uh which is connected to
1:48:09 the internet or if you use gpt4
1:48:11 connected to the internet if you ask for
1:48:13 something recent it will actually fire
1:48:15 off a web search go search the web get
1:48:18 the answer and then incorporate that
1:48:19 into your answer so that's how it gets
1:48:21 around that limitation the limitation is
1:48:24 there because when they build the the
1:48:25 core large language models
1:48:28 it takes a tremendous amount of time to
1:48:30 identify and tag the the core data
1:48:33 sources that get what's called embedded
1:48:36 ingested into that base large language
1:48:39 model so that takes months if not a year
1:48:42 or more and millions of dollars in many
1:48:45 many gpus running 24 7 to do all that so
1:48:49 why this thing works so good is because
1:48:52 they did all that work for us
1:48:54 and then the fine-tuning the sort of
1:48:56 more recent stuff is sort of sitting on
1:48:58 top of that and then like the internet
1:49:00 searches and things like that so but yes
1:49:02 like they're they're training gpt5 right
1:49:05 now
1:49:06 um while we're using gpt4 but gpt4
1:49:10 hasn't even been fully rolled out like I
1:49:12 you may not know this gpt4 is multimodal
1:49:15 meaning
1:49:17 it can understand images
1:49:20 so you could upload a picture of a dog
1:49:22 and a fire hydrant and it would you'd be
1:49:24 able to ask chat GPT questions about
1:49:26 that photograph
1:49:28 describe this
1:49:29 um
1:49:30 when what year was that fire hydrant
1:49:33 made what if whatever it might be it
1:49:35 it's it it's going to be able to
1:49:36 understand image inputs that's not even
1:49:39 rolled out yet like that's not available
1:49:41 to us
1:49:42 there's a new thing coming called code
1:49:45 interpreter where you can upload 100
1:49:47 megabyte
1:49:48 um spreadsheet of data it will read all
1:49:52 the data and then act as a data analyst
1:49:54 for you like that's coming so yes so
1:49:57 they're constantly upgrading it but it's
1:49:58 not they're not dynamically updating
1:50:01 that base model just because of the
1:50:03 nature of how they're built is it's a
1:50:05 big expensive manual process
1:50:10 um how long how much storage does GPT
1:50:14 save
1:50:16 I don't
1:50:17 oh saving history how much storage oh
1:50:20 okay
1:50:21 okay
1:50:22 so there's
1:50:24 there's a thing called a context window
1:50:25 so any given session that you're doing
1:50:27 with chat GPT there's a essentially a
1:50:31 3000 word think of it like a sliding
1:50:33 3000 word window so if you start and say
1:50:36 write me a poem and then fix the poem
1:50:39 and then do this and then do that and
1:50:40 now turn it into a short story
1:50:43 at some point it will kind of forget
1:50:45 about the poem you told it to write
1:50:47 because it'll slip out of that context
1:50:49 window so that sliding window is about
1:50:51 three thousand words
1:50:53 um
1:50:54 there are tools out there like there's a
1:50:56 thing called Lang chain which is a
1:50:58 series of tools that are kind of glued
1:50:59 together that give it longer term memory
1:51:03 um and then there's the if you go to
1:51:05 poe.com one of the models there has a
1:51:07 hundred thousand uh like a 75 000 word
1:51:11 um context window so the the 75 000 word
1:51:14 context window at poe.com
1:51:17 um has a much longer memory so so that's
1:51:21 why sometimes it forgets what you're
1:51:22 talking about you're like well wait a
1:51:24 minute you just I told you all that
1:51:26 stuff and now you don't seem to
1:51:27 understand it that's why because it
1:51:28 slipped out of that context window so
1:51:31 um yeah that'll that'll help
1:51:34 hopefully explain what the [ __ ] going
1:51:36 on there
1:51:37 um again
1:51:38 a year from now
1:51:41 I I think these things will this
1:51:42 effectively have infinite memory
1:51:46 um chat GPT has already announced that
1:51:49 um
1:51:50 you'll be able to create a profile
1:51:52 within chat GPT where it will understand
1:51:55 who you are what you like what your
1:51:58 background is like what whatever stuff
1:52:00 is in there that you want to be kind of
1:52:02 your permanent memory you'll be able to
1:52:04 put in there it's not there right now
1:52:07 um so so just understand that we're
1:52:09 early and this stuff sucks for a while
1:52:13 um I made an Elvis bot cool
1:52:15 um in open chat to act as customers
1:52:18 customer support it responds with well
1:52:21 thank you very much
1:52:22 it's funny the last play I did I played
1:52:25 Elvis in Picasso at the lapana's yield
1:52:27 in Ridgefield Connecticut that was the
1:52:29 last play I did
1:52:32 that was in
1:52:35 2004 it's a while ago
1:52:38 how do you get money
1:52:40 you do stuff for people that are like I
1:52:42 want to give you money
1:52:46 I don't know what the question is how do
1:52:48 I get money I'm the CEO of a of a video
1:52:51 technology company and we do a lot of
1:52:53 work in the healthcare and Pharma
1:52:55 business and
1:52:56 that's how I make money right now
1:53:00 um if I keep talking about chat GPT this
1:53:03 much open AI should start paying me a
1:53:04 sponsorship [ __ ] fee I mean this is
1:53:07 like a nightly commercial for them
1:53:09 um
1:53:10 you need to create a commercial where
1:53:12 you can say you can make money
1:53:15 I should do that that would be funny
1:53:17 except I don't know how you make that
1:53:19 without like I feel like all those
1:53:21 people are already Lampoons of
1:53:23 themselves like they're just [ __ ]
1:53:25 ridiculous
1:53:31 is there a knowledge update coming in
1:53:34 gpt5 that's probably not gonna be for at
1:53:36 least
1:53:37 I'm guessing a year
1:53:39 like like I I don't think open AI is
1:53:42 interested in
1:53:44 um
1:53:45 speeding gpt5 to the market I think
1:53:48 they're probably Slow Rolling it because
1:53:49 they're still trying to they haven't
1:53:51 fully rolled out gpt4 yet
1:53:55 there's a lot there I mean the other
1:53:58 thing is if you think about it
1:54:00 when they created chat gbt
1:54:04 um the team put it together in like two
1:54:06 weeks it was like a very short project
1:54:07 like like the piece of software that
1:54:10 they launched on November 30th was
1:54:12 really just a proof of concept that that
1:54:15 kind of coincided with Microsoft saying
1:54:17 hey we're investing investing a billion
1:54:18 dollars in open AI
1:54:21 um
1:54:22 and then in five days they had a million
1:54:24 users and in six weeks they had a
1:54:26 hundred million users so I think that
1:54:27 open AI right now is is just dealing
1:54:30 with scaling
1:54:32 you know getting the safety guard rails
1:54:35 right like locking this thing down
1:54:36 because so many people are using it and
1:54:38 it like
1:54:40 this generative AI stuff went from like
1:54:43 people talking about it a little bit to
1:54:45 all of a sudden it's like the most
1:54:47 overhyped thing in the world and that
1:54:48 happened overnight like with the World
1:54:50 Wide Web
1:54:51 I started agency.com in 1994. it wasn't
1:54:55 until like mid 1997 until people really
1:55:00 started frothing at the mouth about the
1:55:02 internet and then it wasn't until 2000
1:55:04 that the the.com bubble burst so there
1:55:06 was a good six seven year period where
1:55:10 the World Wide Web awareness of it was
1:55:13 ramping up
1:55:14 this thing was like overnight like it
1:55:17 was literally overnight that that thing
1:55:19 just jumped onto the scene so
1:55:22 crazy crazy crazy
1:55:27 awesome show sir thank you very much I
1:55:29 appreciate that
1:55:31 all right a few more questions it's 9 56
1:55:35 if you make the commercial you just
1:55:37 demoed it'll be the next where's the
1:55:39 beach
1:55:46 oh my God by the way if if
1:55:51 if that if one of your questions is how
1:55:53 do you make money with that gbt here's
1:55:55 the here's the the non-smart ass answer
1:55:57 to that
1:55:59 is chechi PT is just a tool
1:56:02 so it's Photoshop so is Microsoft Word
1:56:04 so is a shovel
1:56:06 so is a lawnmower
1:56:08 so the answer to how do you make money
1:56:10 with chat EBT is actually identical to
1:56:12 how do you make money with any other
1:56:13 tool how do you make money with
1:56:15 Photoshop well you get good enough at
1:56:17 Photoshop that you can solve someone's
1:56:19 problem well enough that they're willing
1:56:22 to pay you for it
1:56:24 I can take your photographs and make
1:56:26 them suck less will you pay me for that
1:56:28 sure
1:56:29 here's 50 bucks great here's your
1:56:31 picture
1:56:32 if that's a good experience you got a
1:56:34 business if you suck at Photoshop you
1:56:37 got 50 bucks time to move on
1:56:40 same thing with this now here's the
1:56:42 difference with chat gbt
1:56:45 with Photoshop
1:56:47 you actually have to have some talents
1:56:49 and skills
1:56:51 with mid-journey and with chat GPT
1:56:54 the depth and power of what it can
1:56:58 create
1:56:59 is significantly beyond what any of us
1:57:03 can do individually right so I could
1:57:07 write in the next hour I could write a
1:57:10 plausible research paper on
1:57:13 a subject I know nothing about
1:57:15 on mathematics or on physics or on I
1:57:19 don't know
1:57:20 um organic gardening techniques or on
1:57:23 [ __ ] that I know marketing plans
1:57:26 um
1:57:27 um you know storytelling Theory like and
1:57:31 then I can go to Mid journey and I can
1:57:32 or or stable diffusion and I can create
1:57:35 sophisticated images even though I don't
1:57:37 have the talent to do that so so that's
1:57:39 the difference with it is that again
1:57:41 it's a Humanity amplifier so it can take
1:57:43 whatever you desire to do and it can do
1:57:46 it really really quick
1:57:49 um so that's that's how you make money
1:57:51 with chat EBT is you figure out a
1:57:52 problem to solve
1:57:54 you take something that you've got
1:57:56 knowledge about or something that you've
1:57:58 got passion about and then you figure
1:57:59 out which of these tools can help you do
1:58:01 that faster better or cheaper than the
1:58:04 next person down the line and then you
1:58:07 tell the world about it and if you do it
1:58:08 right they're like here's 50 bucks and
1:58:11 if you do it right they'll give you 50
1:58:12 more
1:58:13 and if you get 10 of those then you can
1:58:15 charge 500 bucks and if you get 10 of
1:58:17 those maybe it's 5 000 and then you have
1:58:20 a career or you have a job
1:58:22 so
1:58:23 I know nobody asked me how do you make
1:58:25 money with chat GPT but if we did start
1:58:27 with the commercial
1:58:29 s
1:58:30 [Music]
1:58:32 [Laughter]
1:58:36 Amelia's wife hears it in her sleep
1:58:38 that's funny
1:58:40 um all right a couple more a couple more
1:58:43 and then I'll go to go go go go go
1:58:48 Limitless ways to make money with chat
1:58:50 GPT have you tried AI powered Google
1:58:53 workspace I haven't gotten access to it
1:58:55 yet I'm a little cranky about it like I
1:58:58 think Bard sucks Bard compared to gpt4
1:59:01 is just it's like not anywhere in the
1:59:02 same neighborhood
1:59:04 um but
1:59:07 even the suckiest Bard is better than
1:59:11 what we had before in Google Docs right
1:59:14 so if I've got a tool that can can
1:59:17 basically be my writing assistant as I'm
1:59:19 writing things in Google Docs and when
1:59:22 it comes out as co-pilot and in Office
1:59:24 365 that's going to be remarkable I mean
1:59:27 I but no I haven't tried it when it when
1:59:30 it gets truly integrated into the tools
1:59:32 we use like like photoshop's generative
1:59:34 fill
1:59:35 is really just a natural extension of if
1:59:37 you've ever used Photoshop and you know
1:59:39 how to make a selection and invert it
1:59:41 and you paint [ __ ] outside of the thing
1:59:44 that you were doing before like anyone
1:59:46 who's used Photoshop has done that it's
1:59:48 like that except there's [ __ ] magic
1:59:51 pixie dust in it that just fills in your
1:59:54 photo for you and it looks world class
1:59:56 you don't have to photograph out the you
1:59:59 know the power line that went across the
2:00:02 tree and you had to like rubber stamp
2:00:05 all the leaves and no matter how hard
2:00:06 you tried it looked like [ __ ] and you'd
2:00:08 spend three hours trying to photoshop
2:00:10 out a [ __ ] wire well now you just
2:00:12 like Circle the wire and go remove wire
2:00:17 give me more background
2:00:20 Everything Changes Everything Changes
2:00:24 Everything
2:00:27 another great AI date night in the bag
2:00:29 thanks Kyle thank you Silver Fox AI date
2:00:32 night that's what we're gonna call these
2:00:34 hi I'm Kyle Shannon this is AI date
2:00:36 night
2:00:37 [Laughter]
2:00:39 all right everybody I am gonna go I I
2:00:42 hope you've had a good time tonight I
2:00:43 appreciate all the hearts and all the
2:00:45 roses and all the sweet sweet uh words
2:00:48 and didn't have any Power Trolls tonight
2:00:52 we did have Eric working on a stand-up
2:00:54 routine so Eric thank you for your
2:00:56 comedy
2:00:57 thank you frumple for the rose and uh
2:01:00 I'm gonna go take a nap or as as we like
2:01:03 to call it in the
2:01:04 in uh in Denver sleep
2:01:07 all right peace out everybody I'll see
2:01:10 you tomorrow