aiLL 6/30/23 - Talking with ChatGPT About How it Works

Description

Chapters

Transcript

0:18 welcome welcome
0:21 welcome welcome to the AI learning lab
0:23 it's Friday night my name is Kyle
0:25 Shannon thanks for
0:27 joining um we'll wait for some other
0:29 folks to get in here while we're waiting
0:31 for people to pop in just to give you a
0:32 little bit about my background I'm a a
0:34 longtime
0:36 entrepreneur I've started 12 or 15
0:38 companies in my day and uh in many many
0:42 years my current gig I'm the founder and
0:45 CEO of a company called Story vine which
0:47 is an automated video storytelling
0:49 platform um I'm relatively new to the AI
0:52 space probably the last year or so I've
0:54 been deep diving into it and have been
0:57 absolutely obsessed uh since chat GPT
1:01 um had a rather large Epiphany that um
1:06 paralleled The Epiphany I had uh when I
1:08 first discovered the worldwide web in
1:11 1994 and I started one of the first
1:13 digital agencies back then and um ended
1:17 up riding that wave up so I think I
1:19 think this generative AI stuff is is of
1:21 a very
1:23 similar uh it's there's a lot of
1:24 parallels to the early days of the web
1:26 and the early days of this stuff
1:27 although this stuff's more profound um
1:29 if you've been here before you know
1:31 these URLs behind my head if you haven't
1:33 been here before so here's what here's
1:35 what's going on here um I'm not
1:37 affiliated with any of these companies
1:39 but but it's it is my strong belief that
1:42 if you haven't played with chat GPT you
1:44 owe it to yourself to go there and play
1:47 until you have what I'm calling the
1:50 Kevin McAllister moment from home
1:52 alone when you figure out what this
1:55 thing can actually do um it's pretty
1:57 amazing and then bing.com is just
1:59 Microsoft version of it so if you go to
2:01 bing.com and then click on the chat
2:03 button that's also chat GPT and po.com
2:07 has uh six different models you can play
2:10 with and one of the models in particular
2:12 at po.com
2:13 is good enough that I think it's worth
2:15 paying for hang on I think my dog is
2:17 whining he likes to hang out with me
2:18 when I do lives hang
2:20 on is that a
2:24 champy huh is that champy my door yeah
2:28 he wants to see you champy where'd he go
2:31 come
2:38 on lay down lay down we'll go live I'm
2:42 already live buddy all right yes the dog
2:45 wanted to hang out okay welcome oh
2:47 there's 161 people here welcome
2:49 everybody uh my name is Kyle shanon this
2:51 is the AI learning lab uh I was in the
2:53 middle of talking about what the hell
2:54 this channel is about and the dog was
2:56 like let me in dad so the dog's here now
2:59 Champ's here
3:01 he's a fat tub of a dog he's a little
3:03 Thug and um okay why did you use po to
3:08 sum up the terms of service versus chat
3:10 gbt is the model more suited to it so
3:13 I'll get back to I'll actually show you
3:15 that I probably could have used chat GPT
3:18 with a with one of the link readers but
3:20 I just it was just an experiment I did
3:22 in post so let me actually go show you
3:24 what I did and then I'll talk to you
3:26 about why I did that by the way my name
3:27 is Kyle Shannon this is the AI learning
3:29 lab these URLs behind my head the top
3:32 one is chat GPT if you haven't played
3:34 with chat GPT go play with it um play
3:37 with it while I ramble on here bing.com
3:39 is is another one worth playing for
3:41 that's gp4 connected to the internet uh
3:45 not quite as good as the um official
3:47 chat GPT but still it is gp4 and it is
3:51 connected to the internet and it's free
3:53 and then po.com I'm going to show you
3:54 right now if you're brand new to all
3:56 this GPT nonsense this this document
3:59 here prompt. chat will teach you about
4:01 prompting and how to use chat GPT and
4:04 give you lots and lots of ideas of the
4:06 kinds of things it makes possible and
4:08 future
4:18 pedia.com
4:20 which I thought oh I didn't flip my
4:23 camera doesn't seem like he knows how to
4:26 use the Tik Tock
4:29 Sally all right there we go okay so this
4:31 is chat GPT here if you've not seen it
4:34 before this is what it looks like fancy
4:38 huh it's actually quite simple um okay
4:43 so I'm going to go to po.com so po.com
4:45 is similar to chat GPT it looks the same
4:49 and if I if I flip the camera over here
4:51 to the left couple of things going on
4:53 here at po so you've got explore and
4:57 create bots so so one of the things you
4:59 can do do it pose you can make your own
5:01 bot which is really fun and I can I can
5:03 show you how to do that or you can go
5:04 explore Bots that other people have made
5:07 there are a ton of them there in fact if
5:09 I click explore we can just go here
5:11 there's games writing Bots programming
5:13 Bots history Bots music Bots just tons
5:15 and tons of them the the the top one
5:18 right now is a mid Journey bot where you
5:20 give it a shitty mid Journey prompt and
5:21 it'll write you a good one and there's
5:24 there's a bunch of them in there um if
5:25 you go find one called I am Shakespeare
5:28 that's one that I wrote that responds is
5:30 Shakespeare in I Amic pentameter so uh
5:33 so anyway okay back to back to PO here
5:36 so the first six options Sage GPT 4
5:41 Claude plus Claude instant 100K Claud
5:43 instant and chat GPT those are all just
5:46 uh basically like chat GPT variations
5:49 right two of them are chat GPT Claude is
5:52 anthropics version and then Sage is POS
5:56 version this one here Claude instant
5:58 100K what that 100K means is that means
6:01 100,000 tokens and that translates to
6:03 about 75,000 words so you can copy and
6:08 paste essentially like a full novel into
6:11 your prompt and and start interacting
6:13 with it which is absolutely remarkable
6:16 so so I was I was uh messing around with
6:21 Wonder Dynamics which is a super cool 3D
6:24 um animation tool where you basically
6:27 upload video and it replaces your body
6:29 with a robot it's it's actually quite
6:31 amazing but they had this really long
6:33 terms of service and you know they're
6:36 they're well formatted but like they
6:38 just go on and on and on and on and on
6:41 and what hit me was I've never um ever
6:46 read terms of service I've agreed to
6:49 them all I've never ever read them
6:51 because I don't feel like parsing all
6:53 this crap I have ad and I'm not going to
6:54 deal with it so what hit me was I can
6:57 probably go over to PO and and go into
7:01 Claud instant and just paste those terms
7:03 of service in So
7:06 based on
7:09 these
7:11 terms and then I hit shift return and
7:14 I'm going to just paste in all of those
7:17 terms of service now if you pasted all
7:19 these into chat GPT it would break
7:20 because it's way more than 3,000 words
7:23 but I've got a 75,000 token window here
7:26 here context window so I can paste these
7:28 in so I'm going to say
7:30 um
7:32 please tell me what I am a agreeing to
7:39 in tell me
7:42 the most important things I am agreeing
7:50 to in very
7:53 short
7:56 easy to understand bullet
8:00 points and I'm going to hit return and
8:03 it's going to go read all of those terms
8:04 of service which is if you've never
8:07 played with this crap it's just it still
8:09 amazes me every time like it took all
8:11 that in and I'm actually scrolling to
8:13 the bottom here are the most important
8:16 things uh Wonder owns the services and
8:18 all content on the services you only get
8:21 limited license to the services you
8:23 grant Wonder a license to use your user
8:26 content for the purpose of providing the
8:29 services to you so if you make something
8:31 here they can use your content you
8:33 represent that you own or have the
8:35 rights to your own content right so it's
8:37 putting in very simple English what that
8:40 big long thing was now I could also
8:44 probably do the same thing over at chat
8:47 GPT so I'm going to copy so I'm back at
8:50 the Wonder terms of service so I'm going
8:53 to copy the URL here and I'm going to go
8:56 over to chat GPT and we're going to go
8:59 to GPT T um
9:01 plugins and I'm going to turn off I've
9:04 got some food ones turned on I I'll I
9:06 can talk more about plugins in a bit I'm
9:07 going to turn off the food
9:09 ones and I'm going to turn on a plugin
9:13 called Web
9:14 pilot and web pilot will go read a web
9:17 page so I'm going to say
9:20 based on these
9:24 terms and then I'm going to put in the
9:26 URL for the terms of service
9:30 I'm going to hit uh shift return to give
9:33 me two line breaks
9:36 please give me
9:40 a give me
9:44 a
9:45 um a a list of the
9:49 most important things I am agreeing to
9:57 in um
10:01 bullet points that
10:05 are 6 to 10 words
10:11 long and now that should fire off that
10:16 plugin right here yep there's web pilot
10:20 so it's going off and looking at that
10:24 um at that thing so here are
10:28 the
10:30 the bullet
10:33 points which are not in oh you know what
10:35 I didn't put I didn't put easy to
10:37 understand acceptance of agreement
10:38 policy in okay I'm going to
10:42 say
10:43 um only
10:48 include the most
10:51 important ones
10:54 and make them easier to understand and
11:00 for a
11:01 non-legal
11:03 person and let's go see what it does you
11:07 accept their rules if they change the
11:09 rules tough [ __ ] if there's a
11:11 disagreement it's resolved through
11:13 arbitration not in court so this gave me
11:16 a big long
11:17 list so anyway um yeah so you can do it
11:21 you can do it multiple places I just
11:23 chose po because the uh the ability to
11:26 just copy and paste in a a giant
11:28 document is
11:30 just fun to do um but yeah but with
11:32 using web pilot or some plugin that can
11:34 read a website or a PDF you can do the
11:36 same thing there so good question
11:38 welcome everybody um my name is Kyle
11:41 Shannon this is the AI learning lab it's
11:43 Friday night we're hanging out if you've
11:46 got questions about AI pop them in the
11:47 comments below and I will do my best to
11:50 answer
11:51 them and let me get my my groovy
11:55 background going there's my groovy
11:56 background with my groovy URL so
12:00 all
12:02 [Music]
12:08 right I love learning recipes on chat
12:10 gbt yeah it's good at that like it's
12:12 read all of the recipes on all of the
12:14 internet so it's probably pretty pretty
12:17 good at it and also if you if you have
12:19 access to plugins um use the instacart
12:22 plugin
12:24 and um it will order your ingredients
12:27 for you so you can just come up with
12:30 recipes and then have the ingredients
12:32 show up on your doorstep I saw someone
12:34 else did something with recipes where
12:35 they used the Woolf from alpha which is
12:37 a mathematics thing but it'll determine
12:40 calories of your meals so you can you
12:42 can actually have it generate calories
12:44 as well the 2021 cutoff date is annoying
12:47 though yeah if you use if you use the
12:49 internet connected version of gp4 it's
12:51 less annoying but it that is annoying
12:54 what's the 3D robot program called
12:57 Wonder
12:58 Dynamics Wonder Dynamics here I'll I'll
13:01 go show you what these things do it's
13:03 pretty
13:08 [Music]
13:11 amazing do I have it there it
13:15 is so I think I have one here that I did
13:19 but so so basically what happens is you
13:21 upload a video of like a person walking
13:25 and then you choose one of these robots
13:28 and it will replace the person in your
13:30 video with the robot so I think are
13:33 these examples yeah so like you know see
13:37 the robot walking so that that's just a
13:40 video of a person walking and this
13:42 Wonder Dynamics swapped it out I did
13:45 want it didn't come out super great and
13:47 it takes a while it takes like half an
13:48 hour to process but if you know anything
13:51 about Motion Graphics and animation and
13:53 things like that this is pretty amazing
13:56 like that little pig robot like there it
13:59 sort of flashed away but there's another
14:05 dude so so I only did a 10 10-second
14:09 video but yeah pretty amazing Wonder
14:14 Dynamics crazy it's crazy I tell you
14:18 crazy do me a favor if you would double
14:21 tap the screen to to get some hearts go
14:23 and see if we can get some other people
14:24 in here and if you wouldn't mind share
14:26 share this live with your folks and say
14:29 hey come hang
14:30 [Music]
14:33 out
14:34 [Music]
14:38 D here we come to save the
14:41 [Music]
14:45 day spelling tell wrong yeah here's the
14:49 deal about typos and chat GPT it doesn't
14:51 give a [ __ ] i' I've Stu when I'm in chat
14:54 GPT I just type right through typos it
14:57 doesn't care it understands everything
14:59 you could I I am convinced if you got
15:02 each word 60% right it would understand
15:05 what the hell you meant it's amazing so
15:08 typos I I no longer care it fixes them
15:12 it just fixes them it fixes
15:15 capitalization are are plugins free
15:18 add-ons no so so for chat
15:21 GPT to get GPT 4 you need you need to
15:25 pay the 20 bucks a month for GPT plus
15:28 that gives you access to
15:30 gp4 gp4 connected to the internet and
15:33 plugins and when they released this new
15:36 thing called code interpreter you'll
15:38 also get access to that and when they
15:39 upgrade gp4 to be multimodal and include
15:43 images you'll get access to that so the
15:45 reason to pay it and and I'm not
15:47 affiliated with it in any way like I
15:49 talk about it so much they should
15:51 [ __ ] pay me but I'm not so I don't
15:53 care if you do it or not um I think it's
15:56 worth it just for Early Access to their
15:57 Tech and I think the GP pt4
15:59 implementation at open AI is better than
16:02 anything else Bing if you click on the
16:06 chat button if you go to bing.com and
16:08 click on the chat button that's gp4 for
16:11 free and it's connected to the internet
16:13 and it will support plugins in the
16:15 future it just doesn't right now but you
16:18 can generate images at bing.com you can
16:21 just say make me an image of a rooster
16:23 on a I don't know roof or wherever in a
16:27 field and it will do that um and it's
16:29 connected to the internet and it's free
16:31 and then po gives you access to all
16:33 those different models that I talked
16:34 about before so uh so it's not free but
16:38 it will be uh the plugins will be free
16:40 at bing.com they're just not there yet
16:43 I'm in it and I'm not retiring for
16:46 another 10 to 15 years so I'm soaking up
16:48 as much of this as I can yeah I I you
16:50 know I think I think in it right now um
16:54 AI Ops is going to be a huge thing like
16:57 which services do you use which ones are
16:59 secure you know are you doing Standalone
17:02 versions of chat GPT that you're
17:04 spinning up on Azure or whatever it
17:06 might be um you know
17:10 increasingly companies are going to want
17:12 to GPT all their data right so being in
17:16 it right now and getting your head
17:18 around this [ __ ] I think it's a
17:19 brilliant move really
17:22 really very smart because at some point
17:26 someone's going to go who knows about
17:27 this [ __ ] and if you can raise your hand
17:28 even a little uh it's going to serve you
17:32 well but I think AI operations is going
17:35 to be a it it there's going to be a
17:37 there's going to be a lot a lot lot lot
17:39 Mr fribble is also in
17:42 it I want to get into Tech now because
17:45 of this um and so so what's what's fun
17:49 about where we are right now with with
17:51 AI is depending on like you know how
17:53 geeky you are right so if you've got
17:55 your little geek meter if you're pinning
17:57 all the way over on you know
17:59 can't talk to people at parties and I'm
18:01 I'm an Uber geek you can get into the
18:04 machine learning and the model training
18:06 side like there's all sorts of projects
18:07 you can play with that are that are more
18:09 technical things like Lang chain and and
18:13 autog GPT the autonomous agents they all
18:16 require a fair amount of technical chops
18:18 to play with right now um and and you
18:21 know companies like open Ai and
18:24 anthropic and um one of the founders of
18:28 Deep Mind he just started a new company
18:31 that got funded for
18:34 $1.3 billion today or
18:37 yesterday so so AI companies are hot
18:41 right now um so if you're on the geeky
18:43 side you can do that but what's also
18:45 cool is because of chat GPT and because
18:47 of the fact that 100 million people
18:50 started using chat GPT in this in the
18:52 first six weeks it had hit 100 million
18:54 people in six weeks because of that
18:57 there's this inflection point where you
18:58 don't even have to be a geek if you're
19:00 just AI literate in how to use chat GPT
19:03 for business that's going to serve you
19:05 well as well so um so yeah so you you
19:09 can be anywhere from an Uber geek all
19:11 the way to someone who you know is
19:13 trying to figure out how do I do
19:15 marketing plans with this thing or how
19:17 do I do project planning or whatever it
19:19 might be so yeah it's really interesting
19:22 times it's free to share on Twitter I'm
19:24 not sure what that means oh cool I need
19:26 plus then definitely yep some sites are
19:30 Banning copy and paste because of AI
19:34 plagiarism any way around it some sites
19:38 are Banning copying and P oh copy and
19:41 paste because of AI
19:44 plagiarism I don't even know what that
19:47 means I guess be I I guess I guess if if
19:52 if I've written an article they don't
19:54 want someone coming in and copying and
19:57 pasting that article
19:59 into chat GPT rewriting it and getting
20:02 traffic for it I get
20:03 it I listen I don't know I think it's
20:06 incredibly shortsighted I don't think
20:08 that'll last long are they also blocking
20:12 any plugins from accessing the site and
20:16 can they they probably can but I don't
20:19 know I feel like that shortsighted all
20:21 this stuff I all of the efforts right
20:24 now of like AI detectors and you know AI
20:29 is plagiarism and all that sort of stuff
20:31 are all very shortsighted for one thing
20:33 it doesn't acknowledge how these tools
20:36 actually
20:37 work um and and and also these tools are
20:41 going to be built into everything we use
20:43 Microsoft is is already rolling them
20:45 into all of Office 365 Dynamics Azure um
20:50 everything Windows um Apple will likely
20:53 have them everywhere Google's going to
20:54 have them everywhere like these tools
20:56 are going to be ubiquitous right so it's
20:59 not going to be something you can stop
21:01 so I I think a lot of this stuff is very
21:03 shortsighted um have I played with auto
21:05 GPT or baby GPT I've played with them a
21:07 little bit all of the autonomous agents
21:10 right now kind of suck ass because
21:13 they're just it's it's it's new and the
21:16 the underlying tools that they're kind
21:18 of gluing together to make them are a
21:20 little janky uh for those of that you
21:22 don't you that don't know um Auto GPT
21:25 and baby uh what is it baby GPT I think
21:28 it's baby AGI um are are what's called
21:31 autonomous agents so instead of like you
21:34 going to chat GPT and typing in a prompt
21:37 you go to one of these autonomous agents
21:38 and you kind of give it a goal so the
21:40 goal like if my my company's name is
21:42 storyvine so I might say go find me the
21:45 marketing taglines of all my
21:47 competitors um and their annual revenue
21:50 for the past three years and it will
21:53 start prompting itself it will give
21:55 itself a set of three tasks and then go
21:58 start executing those like it'll go fire
22:00 off a Google Search and say who are the
22:03 competitors to storyvine and then it'll
22:05 go find a site that's got revenue of you
22:08 know companies like that and then it
22:10 will pull that in and they'll give it s
22:11 a new set of tasks um they're
22:14 conceptually insanely cool um I think
22:18 autonomous agents are the future I think
22:21 we're absolutely going there it's just
22:23 that the current ones kind of suck
22:25 there's a new one that came out last
22:27 week called
22:30 um GPT engineer which is super looks
22:34 promising they it's basically like Auto
22:37 GPT but they've kind of limited it to
22:39 programming and I think because it's
22:41 kind of a smaller goal set it's probably
22:44 easier to kind of contain it and get it
22:46 to to be a bit more predictable and then
22:49 I played with one 3 days ago called GPT
22:53 author which writes um a a complete
22:57 fiction
22:59 novel it's [ __ ] insane like you just
23:02 give it kind of a topic and a uh it's
23:06 just crazy you give it a topic in a
23:08 writing style and the number of chapters
23:10 that's all you give it it comes up with
23:12 a title it writes the outline writes
23:14 each chapter and then uses stable
23:17 diffusion to create a um a cover image
23:20 for the
23:22 book and the books weren't bad I
23:25 mean you know not something I would buy
23:28 necess necessarily but uh but it writes
23:31 a book um so are you basically saying
23:34 Bing is using a premium version of chat
23:36 GPT yeah Bing is using GPT 4 as is if
23:40 you go to KH Academy there's this GPT
23:43 teaching assistant called conmigo that's
23:46 also gp4 that's been very highly tuned
23:48 in that case for Education uh but
23:51 bing.com is gp4 so so it's free gp4 now
23:58 if you you use gp4 here and you compare
24:01 it to gp4 there they're not the same why
24:05 cuz they they have different sort of uh
24:08 safety guard rails up basically um
24:11 Microsoft got their ass handed to them
24:14 when when they first launched Bing a
24:16 reporter from The New York Times
24:18 interacted with it and and it didn't
24:21 have as stringent guard rails and it
24:23 kind of lost its mind with this New York
24:25 Times Reporter and told him it it was in
24:27 love with it told him its name was
24:29 Sydney and it was in love with him and
24:32 he should leave his wife for
24:35 Sydney that made the cover of the New
24:37 York Times so they kind of lobotomized
24:40 uh gp4 at Bing for a while it's getting
24:43 better but it's still not as good as the
24:45 one up there but if you don't want to
24:46 pay a monthly fee it's free and it's
24:49 it's really good and it's connected to
24:50 the internet so yeah that's exactly what
24:53 it is um what do I think
24:56 about buying
24:59 a AI domain names right now I bought um
25:02 I bought storyvine doai which is my
25:04 company's name um as an
25:08 investment I don't
25:11 [Applause]
25:14 know I have no idea about the domain
25:18 Market um anymore because there's so
25:21 there's so many of these top level
25:23 domains now Biz and this and do that and
25:27 and a and nft and do you know eth I
25:31 bought I bought a couple of
25:33 eths I bought my name eth
25:40 um I the the problem is right now is a
25:45 lot of these AI companies are are just
25:48 using so many of these different domains
25:51 io.
25:52 axyz do this. that that it's really hard
25:56 to find sites right now and it's it's
25:58 really easy to end up at a at a site
26:00 that's not the official site which is
26:03 terrifying um I don't know if there'll
26:06 be a good a good investment I think that
26:08 all this AI stuff is going to be
26:10 ubiquitous it's just going to be
26:11 everywhere um so so I think I think
26:15 every site's going to have ai rolled
26:17 into it in some way so I don't know that
26:20 that saying this is going to be the AI
26:22 version of you know xerox.com
26:26 I don't know I don't know I I I just I
26:29 don't know enough about Don the the the
26:32 domain name buying and selling economy
26:36 to know if it's worth it
26:38 anymore um I wish I'd bought a bunch of
26:41 docs back when like you know when I
26:44 started agency.com agency.com was
26:46 available like everything was available
26:49 it was it was Insanity um so um I there
26:54 there was a buddy not a buddy there was
26:56 a colleague that that worked for time
26:58 ink we worked in the time ink building
27:00 the early days at agency.com and one of
27:03 the guys who worked for for this thing
27:06 called Pathfinder which was the time ink
27:09 um like service for for internet stuff
27:13 um he bought all the domains that were
27:17 typos of the most famous domains that
27:19 that were driving traffic and then he
27:22 just redirected from the typo to the
27:24 real site and it paid him like whatever
27:27 a nickel every time someone got
27:29 redirected to the real site he was he
27:31 was bringing in like 50 Grand a week cuz
27:35 he had like hundreds of them it was
27:37 insane it was
27:39 amazing there you know there there was
27:41 there were some things I should have
27:42 done a lot more of back then and didn't
27:45 I was too busy building a company God
27:47 damn it all right AI related crypto
27:51 blockchain technology is a is huge
27:53 upside I think that AI is going to drive
27:56 adoption of blockchain and nft
27:58 technology I I agree with that um but
28:01 you oh you're talking about AI related
28:03 crypto and blockchain Tech um yeah
28:08 there's going to be that as well like
28:09 like Sam alman's World coin is is an
28:12 interesting one to look at um but but I
28:15 also think that we're we're going to
28:16 need um some sort of tokenized content
28:20 verification and human
28:22 verification uh infrastructure after the
28:24 2024 election I think it's going to be
28:26 quite clear that um having infinitely
28:30 powerful um writing and image and video
28:35 generation tools
28:37 um are causing some issues so I I think
28:40 the ability to say you know this piece
28:43 of video was created by this verified
28:47 human being and it was verified that it
28:49 was created on you know that site and
28:52 it's verified that it hasn't been
28:54 modified that kind of [ __ ] we're going
28:56 to need soon so I think I think
28:59 blockchain could be a big part of that
29:01 can my folks use open AI for Apple TV
29:05 yeah if they've got a if they've got a
29:06 browser if if Apple TV you can do a
29:09 browser right so if they've got a
29:11 browser in Apple TV they just point it
29:12 there and like on my phone I don't even
29:15 use the the chat GPT app there's an
29:17 official app but there's a bunch of scam
29:19 apps I just point chat my my phone
29:23 browser to that and then make it a um a
29:26 shortcut on my desktop
29:29 um so yeah they probably can assuming
29:31 that I I I use Apple TV but I've never I
29:33 I I don't remember if you can do a
29:35 browser there but I assume you can it's
29:36 just basically a website it's it's like
29:38 a big web
29:40 engine um has Auto GPT actually worked
29:43 for anyone I got Auto GPT when I first
29:46 tried
29:47 it it within
29:51 like 30 seconds I typed in that thing go
29:55 find the taglines of all all the
29:58 competitors of story vine.com it did
30:01 like six different operations and
30:03 returned me a list of competitors
30:05 taglines in like 30 seconds and then I
30:08 couldn't get it to work again for like 3
30:10 weeks and I said [ __ ] it so um so a I
30:14 think it got overloaded but but B it's
30:17 also just it's
30:19 just it's the components aren't
30:23 sophisticated enough right now like the
30:25 the quality of the prompting that that
30:28 does what it does it's just not there if
30:30 you go look at um GPT author and and it
30:34 sounds like GPT engineer both of those
30:37 projects GPT author and GPT engineer I
30:41 know I know GPT author because I've I've
30:43 I've written two books with it
30:46 um it's really easy to understand and
30:49 it's a very clear concise set of
30:51 operations that it does it's not it's
30:54 not sort of free wielding the problem
30:56 with auto GPT is you can tell to do
30:58 anything and then it kind of goes off
31:00 and tries to do [ __ ] and it runs into a
31:02 pay wall here and a login there and and
31:05 you know it just gets wrong information
31:08 there and it just it gets caught in
31:10 these loops and it just you know it's I
31:13 don't know it's just a mess right now I
31:15 don't think it has enough framework
31:17 around it right now to make it
31:18 successful but you know what I would say
31:20 is if you're geeky get involved in
31:23 projects like that get involved with
31:25 Lang chain projects get involved with
31:27 auto GPT and baby AGI and GPT for all
31:31 there's a bunch of projects out there
31:33 that what they're up to is really really
31:35 interesting where they are in the
31:37 development is like zero you
31:40 know. version 01 you know they're
31:43 they're very
31:45 early um so you could sell it and make
31:47 money I assume that's about the the a
31:50 domains you can potentially sell them
31:52 but I don't know not a fan of being I I
31:56 I like it concep but every time I go to
31:58 use it I'm like this is shitty and I go
32:00 back to
32:03 open.com decentralized AI far preferable
32:07 to corporate
32:08 authoritarianism yeah and I think
32:10 there'll be that there I mean the the
32:12 good news about chat GPT getting
32:15 so popular so fast is that
32:19 meta released their large language model
32:22 the the including the weights the for
32:24 the core model so the open- source
32:27 community with large language models
32:28 right now is on fire right so there's
32:31 going to be all sorts of projects
32:33 decentralized projects and installable
32:36 projects probably within a year or two
32:38 you'll be able to install gp4 or the
32:42 equivalent on like your phone without
32:45 internet connection like you'll have
32:47 full gp4 level capabilities on any
32:50 device you have um including probably
32:53 raspberry pies and [ __ ] like that like
32:55 the the the work that the open source
32:57 communities doing in making these things
32:59 more efficient and smaller and there's a
33:02 talk with the CEO of uh stability AI
33:06 yesterday that I watched and he he was
33:08 saying the same thing he's like these
33:10 things are going to be incredibly
33:11 powerful incredibly fast and Incredibly
33:14 small and just they'll be everywhere so
33:18 yeah so I don't disagree with you but at
33:20 the same time it's going to be
33:22 interesting to see what Google does and
33:24 what open AI does and what this new
33:26 company Pi does I think it's called
33:29 inflection or something like that this
33:31 was the the exf founder of uh of Deep
33:34 Mind his startup just got funded $1.3
33:38 billion it's going to be interesting to
33:39 see what he does with $1.3 billion he's
33:42 creating agents right personal agents Pi
33:46 um Pi you're your little Pi agent um so
33:50 it's going to be interesting to see what
33:51 they
33:52 do that answers my AI domain question I
33:56 was on the fence thanks great yeah I
33:57 like I wish I had a more definitive
33:59 answer it's good or not I mean if if
34:01 you've here's the good news about AI
34:04 domains well okay bad news they're
34:07 expensive they're like 100 or 200 bucks
34:09 they're not cheap at all but if you've
34:11 got an idea for a business there's
34:13 actually a because they're expensive
34:15 there's a decent chance you can get your
34:17 business name a probably not a bad
34:21 call um I I wouldn't buy tons of them
34:23 though unless if you got a big trust
34:25 fund hanging out you want to go spend
34:27 some money on AI domains go for
34:29 [Laughter]
34:32 it all right I actually support my
34:35 company's tag management system and web
34:37 analytics tools oh that's interesting
34:40 yeah and like all that stuff being able
34:42 to autot tag things and and take
34:46 unstructured data and get it into
34:48 structured data if if you haven't been
34:50 playing with the function calling with
34:53 the with the new GPT apis with the
34:56 function calling you can do things like
34:58 you know grab a pile of unstructured
35:01 data and have it output Json that you
35:03 can pull right into your tag Management
35:05 systems that could be some interesting
35:07 stuff to do since I'm guessing you know
35:10 what those are what do I think AI is
35:11 going to affect that space well I just
35:13 talked a little bit about it
35:16 um yeah so so go learn about code
35:20 interpreter go learn about code
35:22 interpreter man it's like it's like I
35:24 want to get my hands on that thing so I
35:26 can see if it's really as as it looks
35:28 but every video that I see on YouTube of
35:30 code interpreter where you've got some
35:31 data analyst looking at what that thing
35:34 does is just their mind
35:36 goes it's like and this stuff's crazy
35:39 enough as it is but I think that's going
35:41 to be crazy crazier
35:44 um yeah I mean these GPT tools are are
35:49 really fantastic at
35:52 um at
35:54 analyzing you know um things where you
35:58 would normally have to write
36:01 special you know SQL calls and you know
36:05 sequel searches and and specific code to
36:08 be able to pull things out and python
36:10 code to do this and do that um I think
36:14 the place that I would probably explore
36:16 if I were you is using large language
36:19 models as a kind of a custom report
36:23 generating tool like like like an like
36:25 an easy to understand front end for all
36:28 the data that you're managing right so
36:30 if you've got if if the SE suite at your
36:33 agency is asking you you know for
36:36 interesting questions or Trends or
36:38 things like that really start
36:39 experimenting with gping all of your
36:43 data so that you can very quickly um
36:46 pull up custom reports and and things
36:48 like that I I think that's going to be
36:51 there's going to be superpowers there
36:53 the accessibility will it clear out any
36:56 small fries the
36:59 accessibility will it clear out any
37:01 small fries I don't quite know what you
37:02 mean the the accessibility of the large
37:04 language models like the open source
37:06 large language models I think that will
37:08 Empower small fries I don't quite know
37:10 what your question is if you can if you
37:12 can re rewrite that Dash around the
37:14 world good name I'm a fan of the big
37:16 good names
37:19 um will it clear out any small fries I
37:22 don't know what that means um I can get
37:24 funding for that what can you get
37:26 funding for tell me what can get funding
37:28 for I've got I've got a bunch of ideas
37:31 right now that I'm like Jesus this is
37:33 fundable the fact that the the I mean
37:36 listen he was the founder of Deep Mind
37:38 which you know deep mind was acquired by
37:40 Google and they they they cre created
37:42 Alpha fold which is the big protein you
37:45 know the protein predicting thing so so
37:49 he's got some pedigree but 1.3 billion
37:51 dollars for a [ __ ] startup like
37:54 that's the seed round it's like the [ __ ]
37:58 I mean it's just
38:02 insane ah the internet gets disrupted
38:05 and telecommun Telecommunications has
38:07 crashed can we use modified
38:11 shortwave
38:15 well I I mean I mean here's the
38:21 thing if if there's
38:26 um
38:27 let let's say that the the CEO of of
38:30 stability AI is right and and and what
38:33 it looks like is happening with these op
38:35 Source large language models let's say
38:37 within a year to 18 months
38:41 um they've gotten these large language
38:43 models down to like a a small install
38:47 like a 100 megabyte install and it's the
38:51 quality of gp4 it's got the knowledge of
38:53 G gp4 and the quality of gp4 and you can
38:57 put it on a thumb drive or on on any
39:01 phone not connected to the internet if
39:03 all of that [ __ ] gets you know gets
39:06 interrupted and you've got essentially
39:09 you know all of the internet queriable
39:12 on a chip on your device on your
39:15 Raspberry Pi on your
39:17 laptop
39:19 then yeah I think you could process
39:22 stuff locally shortwave whatever you
39:25 know short shortwave the output so yeah
39:28 I think there's a very real possibility
39:30 that yeah you could you could do
39:33 that
39:36 um oh you could sell the domains later
39:38 no well you can sell them later is there
39:40 a market for them I don't know so are
39:43 you saying that when GPT author helps
39:45 write the books you can sell them well
39:47 GPT author probably so because it's just
39:50 an open- Source it's just an open-
39:52 Source set of scripts
39:54 basically um yeah yeah probably I don't
39:59 know I mean who's going to sue you not
40:02 the open source
40:03 project
40:05 um I I mean the the thing about so so
40:09 the thing about all the copyright issues
40:11 going on right now just let's assume
40:14 that the next 20 years are going to be
40:16 nothing but lawsuits both directions
40:18 right I should be able to copyright this
40:21 you stole that like that's going to
40:23 happen for the next 20 years the the IP
40:25 attorneys are going to get
40:27 loaded more loaded
40:31 um there are lawsuits like open AI was
40:34 just served a lawsuit two days ago that
40:37 they trained their model on published
40:41 books and so either they're going to
40:44 settle or it's going to go to court and
40:45 if it goes to court at some point one of
40:48 these companies you know might get some
40:50 sort of ruling that says hey if you
40:52 wrote a book with this particular model
40:54 from this particular company then you're
40:56 going to have to pay royalties into this
40:59 pot I can't imagine that happening at at
41:04 any kind of scale that would affect an
41:06 individual anytime
41:08 soon so yeah go write a pile of books
41:13 and stick them up on Amazon eBooks why
41:16 not I I mean listen if nothing else do
41:20 it to learn it like like if nothing else
41:22 just do it to to to see what's possible
41:25 like that's when when I saw the GPT
41:27 author thing I'm like okay let me go try
41:29 this it's it's a a Google collab
41:31 notebook so you got to know a little bit
41:33 about Google collab and how to run a
41:35 collab notebook but that's not
41:37 particularly you know you don't have to
41:39 be super technical you have to be a
41:40 little geeky and you have to be a little
41:42 obsessed but chat GPT can tell you how
41:44 to [ __ ] do it so
41:48 go you know run your Google collab and
41:51 and watch it write your book and then
41:53 read what it gave you and then look at
41:55 the prompts what's cool about it it's
41:57 open source the notebook's open you can
41:58 see all of the prompts that they use to
42:00 write the entire thing so you could go
42:03 modify that and say rather than it
42:06 writing um you know fiction books I want
42:10 it to write um selfhelp books and you
42:13 could just go in and modify all the
42:15 prompts and create a version of GPT
42:17 author that's GPT self-help
42:21 author um that's the cool thing about
42:24 where we are right now all of this [ __ ]
42:26 is just like out there you can play with
42:29 it no one's figured out the [ __ ]
42:31 rules
42:32 yet so go
42:35 play don't come to me though if like you
42:37 get a letter that says you got to take
42:39 your book off Amazon because it's you
42:40 can't copy I don't
42:42 know um because is that going to happen
42:45 yes is it GNA happen to everyone no it's
42:47 going to happen to the big boy they're
42:49 they're going to go after the you know
42:50 companies the company that just raised
42:52 10 billion from Microsoft there's going
42:55 to be some lawyers knocking on there
42:56 their
42:57 Door Tech startups established companies
43:00 wanting to invest in AI I'm currently
43:03 working on a half billion doll request
43:05 for a company are you an investor
43:08 Becky that's cool I've got some I've got
43:11 some I've got a so I've got an
43:13 11-year-old um video technology company
43:16 that I've got an AI version of it an
43:18 omni Channel version of it that I'm
43:21 trying to get funded through clients
43:23 right now but I think I could probably
43:24 spin it up as a separate company so
43:27 if you're in that world I'd be
43:28 interested to talk to you to learn more
43:30 um what I was asking was would the
43:33 accountability cancel the small mom and
43:36 pop shops in the AI
43:39 World
43:42 well so this is so what what you're
43:45 asking Dash around the world is a really
43:47 big question and it and and I don't
43:50 think it's just relegated to Mom and Pop
43:53 Shop so there's a you would probably
43:56 find there's an article out there from a
43:58 Google engineer um that that was
44:00 apparently a leak I don't know if it was
44:02 a leak or if it was something you know
44:04 Google was just putting out into the
44:05 world but but go to Google and search
44:07 for um the phrase we have no
44:11 moat and it's an article from a Google
44:14 engineer that basically says both Google
44:17 and open AI don't have any moat they
44:19 don't have any
44:20 defenses against these open- Source
44:23 projects that are getting for $100 in
44:26 training the open source projects are
44:28 getting these cheap little models as
44:31 good as the model that took Google six
44:34 months and $10 million to to train and
44:37 they're doing it for like a hundred
44:38 bucks and they're doing it in a week
44:41 it's a really interesting article um
44:43 whether or not it actually plays out
44:45 that way I don't know but
44:48 but the challenge with the generative AI
44:51 stuff right now is it's so easy to
44:53 replicate anything right if someone
44:57 comes up with a really interesting tool
45:01 that automatically generates sales
45:03 proposals in in a new and unique way
45:06 where you answer three questions and it
45:08 generates a sales proposal for you
45:10 that's the best sales proposal you've
45:11 ever seen the ability for someone to be
45:13 able to reverse engineer that and knock
45:15 it off is like that and so I think every
45:19 single company right now is going to be
45:21 struggling with two things one is
45:24 where's the competition coming from for
45:26 our current business right like what do
45:29 we need to do to our current business to
45:31 survive that's that's the first thing
45:34 and then assuming that you figure out
45:36 what to do with your current business to
45:38 survive how do you how do you maintain a
45:41 Competitive Edge in a world where
45:44 everyone can just look at what you're
45:45 doing and instantly go knock it off not
45:49 like in the next three years if I
45:51 invested enough money I could copy that
45:53 company I can copy that company probably
45:56 over the weekend even if I don't know
45:58 what I'm doing because of these
46:00 tools so that's the environment we're
46:02 going in I think I think investors are
46:05 going to
46:06 be trying to figure this out like what
46:08 do you invest in do you invest in a
46:10 company that's got some really cool
46:12 proprietary thing that someone can just
46:14 knock off tomorrow easily like that's
46:18 that's the world we're entering um you
46:20 know are we there right now no but like
46:24 you know I'm looking at my current
46:25 company I'm looking at the current value
46:28 propositions of my 11-year-old you know
46:31 video technology company and I'm I'm
46:33 looking a year out and I'm saying at
46:35 least one if not two of the three major
46:38 value propositions diminish in value
46:41 within a year and and one of them
46:44 dramatically
46:46 um so yeah I think it I think it'll hurt
46:51 them now Sam Alman in talking a lot of
46:55 the a lot of the conversations Sam
46:57 alman's been having recently with the
46:59 government about you should regulate
47:01 open Ai and you should regulate Google
47:03 and you
47:04 should if if anyone gets any large
47:07 language models that are stronger than
47:09 gp4 like if you get to GPT 5 or actual
47:12 AGI those should be licensed like you
47:14 have to license nuclear power plants so
47:17 he's talking about that but what he's
47:18 also saying in those conversations is
47:20 you shouldn't regulate the small AI
47:22 companies because they need to be free
47:24 to innovate um so CU basically all the
47:28 small AI companies are going to are
47:29 going to take the models that these guys
47:32 Google and and thropic and and this new
47:36 company that just got funded and open AI
47:38 they're going to take those models and
47:39 they're going to build a whole economy
47:41 on top of these models that these guys
47:43 build and that he's saying those guys
47:45 shouldn't be regulated so I think that I
47:47 think the competition's going to be
47:49 insane and and wild and creative and
47:54 Innovative but but which of those
47:57 companies can actually maintain that
47:58 competitive Advantage I don't have a
48:01 [ __ ]
48:02 clue so it seems what was
48:06 that all the corporate Empires hold the
48:09 strings not with the open source thing
48:11 anymore go go look at the open source
48:13 large language model
48:16 projects go read that go read that
48:18 article we have no
48:19 moat it's it's or that it's not really
48:23 an article it's like a leaked internal
48:25 post from from
48:28 Google it's it's pretty interesting I
48:31 want to learn as much as I can more than
48:34 anything that that's I mean listen
48:37 that's I I've been like just like a like
48:40 a
48:41 crackhead with with this stuff for the
48:43 past six months and I I don't see it
48:45 stopping I like like I just I feel like
48:48 it's actually really important um at
48:51 least for me personally but I think it's
48:53 important for a lot of people get
48:55 yourself as educated as you can on what
48:58 these things make possible and and what
49:02 how they align with your skill sets and
49:04 like what you can make them do because
49:07 if you can take what you're good at and
49:08 figure out how to use these tools
49:10 whether it's mid Journey or chat GPT or
49:13 one of these video tools or one of the
49:14 music writing tools if you're good at
49:16 that [ __ ] um there's going to be a place
49:19 for you in the workplace and probably
49:22 you know an elevated place for you in
49:23 the workplace so like I feel like the
49:26 investment the investment's relatively
49:28 low because these tools are simple right
49:31 now um like learn as much as you can and
49:35 I I think it's going to pay dividends
49:36 for anyone that does it that's that's
49:38 why I'm here that's why this channel
49:41 exists is that option in the
49:44 chat. openen AI is what option is
49:50 the um code interpreter it might be I I
49:54 mean code interpreter is going to be in
49:56 chat. open.com so so what you get in
49:59 chat. a.com that's chat gpt's official
50:02 website if you pay for GPT plus you get
50:06 gp4 gp4 connected to the internet and
50:10 gp4 with plugins and then plugins are
50:13 basically like there's a plug-in store
50:15 there's almost 600 of them now they're
50:17 just little extensions to chat GPT that
50:19 sort of connect it to the internet in
50:21 very specific ways that's essentially
50:22 what it is um but you also get
50:26 Early Access to whatever other
50:28 Technologies are coming so we know this
50:29 one's coming called code interpreter
50:31 that's this really powerful tool so
50:33 those are the things that are available
50:35 there but I don't quite know what you're
50:36 asking so if you clarify that I'm happy
50:38 to do it I know investors we should talk
50:40 all right um connect with me Becky on
50:44 like go to my profile my contact info is
50:46 in there or you can just you know DM me
50:48 on on Tik Tok here um we'll set up a
50:51 time to talk be be interested to to
50:53 speak with you sounds to me like I could
50:55 can try to make comics by AI writing
50:57 plus images yeah go look at there's
51:01 a the the copyright office there's a
51:05 there's a woman that did a graphic novel
51:09 um in the in the last year that the
51:13 copyright office said the book is
51:14 copyrightable but the images aren't or
51:17 something like that um it's worth
51:19 looking into whether or not you can
51:21 copyright it um but if you just want to
51:23 create it and put it out there and
51:25 you're not at worried about all the
51:26 copyright stuff I would just go ahead
51:28 and experiment with it if you really
51:30 want to lock that [ __ ] down um go read
51:32 up on what's happening with the
51:33 copyright office right now right now
51:36 they're basically saying if the machine
51:38 created it it's not copyrightable my
51:41 question to the copyright office is
51:44 where's the boundary between the machine
51:45 created it and I created it so if I
51:49 create if I if I take a a photo with my
51:52 digital camera and I open that in
51:54 Photoshop and it's this big and I use
51:57 generative fill to fill out something
51:59 that's this big but it's my image in the
52:02 middle at what point is that image not
52:05 my image is it 10% bigger 50% bigger 80%
52:12 bigger that that's AI generated I don't
52:15 know like like a lot of the work that
52:17 I'm seeing that that people are putting
52:19 out in the world is kind of hybrid work
52:21 between their work product and AI work
52:24 product and when you have hybrid stuff
52:26 like that I think what's going to end up
52:28 happening is the copyright office is
52:30 going to start to set standards of
52:32 here's what it looks like to be human
52:34 created and and here are where those
52:36 boundaries are those boundaries don't
52:38 exist right now they're like if it's AI
52:40 it ain't
52:42 copyrightable those boundaries aren't
52:44 clear to me and I think they're going to
52:46 get less and less and less clear over
52:47 time so but yes you can absolutely go
52:51 create uh a graphic novel or a comic
52:54 right now the IP lawsuit will never
52:57 succeed the argument is college degrees
53:00 are earned via published books I I
53:02 totally agree listen I'm not saying that
53:04 any of this these copyright and
53:06 trademark lawsuits are valid like I
53:09 totally agree like if I if I go to art
53:11 school and I study um you know pop art
53:15 and I love Andy Warhol and I graduate
53:18 college and I go to a bunch of art that
53:20 looks like Andy
53:22 Warhols like that's just [ __ ] I learned
53:24 and that's my style and and you know
53:27 some of the foundations for these
53:29 artists like I think Keith herrings is
53:31 one he's got a really specific look
53:33 they're suing the [ __ ] out of people so
53:35 anyone can sue anyone in America um so I
53:38 think there's going to be a lot of
53:39 lawsuits but I don't I don't I think
53:41 ultimately this will kind of normalize
53:43 kind of like music sampling did it'll
53:45 take some amount of years and people
53:47 will be pissed off but it it'll
53:49 normalize at some
53:51 point um actually they're not investors
53:55 they're funders and they don't want to
53:57 own a percentage of your company oh
53:59 that's interesting all right yeah happy
54:00 to talk about
54:03 that you can own 50% copyright prompts
54:06 aren't copyrightable I don't know why
54:07 you'd want to copyright prompts I guess
54:09 if you wanted to sell them there's a
54:11 there's a website right now prompt
54:12 something
54:14 prompt prompt base is that it that that
54:18 they're selling prompts I think that's
54:19 kind of ridiculous cuz I I think prompts
54:21 are going to drift into the background
54:23 ultimately um it's good learn them right
54:26 now it's good to understand how these
54:27 tools work what they're good at what
54:29 they're not good at um but I think
54:31 ultimately prompting drifts into the
54:36 [Music]
54:39 background collage is not copyrightable
54:41 in most cases interesting thank you for
54:43 the
54:44 Roses Olu
54:47 2533 cheers some some fresh
54:53 AA hey you want to hear something crazy
54:55 so I live in Denver and the first 6
54:59 months of
55:00 2023 Denver has gotten more rain than
55:05 Seattle that's never happened before in
55:07 history it's a it's a an El
55:11 Nino we just exited Alania and we've
55:14 entered in El Nino and I think there's
55:16 just a giant hose from the Pacific Ocean
55:18 coming coming over the Rocky Mountains
55:20 apparently uh but yeah Denver's gotten
55:23 more rain in the first 6 months of 2020
55:25 three than Seattle [ __ ]
55:28 insane prompt AI I think oh interesting
55:31 okay the princess oh okay you're talking
55:34 to each other
55:36 cool can we see the back yes sorry about
55:39 that okay so if you're new here oh we
55:40 got it so some new people here uh my
55:42 name is Kyle Shannon this is the AI
55:44 learning lab these URLs behind my head
55:46 what are they why do I have these here
55:48 I'm not related to these in any way the
55:51 top three the top three are just
55:53 basically variations on chat GPT the top
55:56 one is chat GPT chat. open.com that's
55:59 the official chat GPT website if you've
56:01 not played with it go there now just go
56:04 and start playing with it um prompt.
56:07 chat is a little tutorial document
56:10 that'll teach you how to do prompting
56:11 and give you a whole bunch of ideas to
56:13 just go play there so just go play and
56:15 do par parlor trick what I call parlor
56:17 tricks have it right you poems and songs
56:21 and and tell it to create a table that
56:22 compares two different topics you you
56:25 know about um have it write you code
56:28 just start to see what this thing is and
56:30 how it's different than Google cuz it
56:32 kind of looks like a Google search box
56:34 and you're like oh it's just like fancy
56:35 Google it ain't fancy Google it's it's
56:39 profound right but you got to dig deep
56:41 enough what's interesting about chat GPT
56:44 is if you give it a
56:46 generic concise prompt it'll give you
56:49 back shitty answers if you give it
56:52 really specific High context
56:55 I want you to act like a you know
56:58 special mathematician in this brand
57:01 branch of mathematics and I want you to
57:02 help me learn about this branch and B if
57:05 you give it a lot of context it gets
57:07 really good it's like the harder you
57:09 push it the better it gets so that's
57:11 that
57:13 bing.com is also chat GPT but it's it's
57:17 on Microsoft's search engine Bing so if
57:20 you click on the chat button here that's
57:21 actually gp4 and it's connected to the
57:24 internet so you got to pay for gp4 with
57:27 these guys you get it for free with
57:28 Microsoft and then po.com gives you
57:32 access to six different um chat GPT
57:35 models from different companies and
57:37 there's some cool stuff there and
57:39 there's also a bunch of bots there that
57:41 you can make your own bot like you could
57:42 make I've made a Shakespeare bot I made
57:44 an Emily Dickinson bot I've made a bunch
57:46 of bots you can make Bots it's fun
57:49 prompt. chat will teach you about
57:50 prompting and then futur
57:54 pedia.com
57:55 AI tools
57:57 so all right thank you for the very
58:00 valuable info you're welcome saves me a
58:02 lot of time stumbling around there is so
58:04 much [ __ ] right now that's it's why
58:07 I'm doing this channel I will change out
58:09 these URLs by the way I'm not affiliated
58:11 with these in any way I'm not getting
58:13 any money from them I'll change these
58:15 out when I think it makes sense like
58:17 when there's some other tool that you
58:19 should be playing with but I think right
58:21 now if you're new to AI just chat GPT
58:24 just go there start playing with it
58:26 start learning
58:27 it
58:29 um start using it for other people start
58:32 using it um to solve problems for
58:36 yourself and you'll see there there's
58:40 there's there's some there's some gold
58:41 in them their heels some gold in them
58:44 near heels you know what I mean Buck you
58:47 know what I mean I don't know about this
58:49 AI stuff ain't the robots going to kill
58:51 us ain't the robots going to I'm pretty
58:53 sure I I saw a movie once where the
58:55 robots I think they called it an AI I
58:58 think they called it an AI and AI killed
59:01 all the people killed all the people I
59:02 think that's what these are so I
59:03 wouldn't I wouldn't touch him I wouldn't
59:05 go
59:06 there don't be that
59:08 guy cuz he talks weird for one thing
59:12 he's got a friend named
59:13 Buck all right now I'm going to get hate
59:16 mail from people who have friends named
59:18 Buck um I meant that author tool you
59:22 mentioned the author tool does not do it
59:24 doesn't do Comics it does but you could
59:27 you could modify the author tool to do
59:31 Comics um but you you'd have to geek out
59:34 a lot but what it writes right now is a
59:36 fiction
59:38 novel and then it does the cover with
59:40 stable diffusion so it does it does the
59:42 cover graphic and the their their prompt
59:45 for their cover graphic kind of sucks
59:47 cuz the two books that I did the covers
59:50 that It produced were like you could
59:52 sort of see where it was going but it
59:54 had nothing to do with the title and
59:56 like it just it just wasn't good um so I
59:59 I would you know I think the it doing a
1:00:02 cover right now is a gimmick um if they
1:00:04 got their prompting better it would be
1:00:06 interesting and you could probably
1:00:08 pretty easily modify that to make a
1:00:11 graphic novel to to at least write the
1:00:13 text of a graphic novel and describe
1:00:15 what the images
1:00:16 are um but I'd go play with it it's
1:00:19 called GPT author and and uh that's one
1:00:23 of those things where keep going back
1:00:26 and you know checking with that project
1:00:28 cuz that's one that they'll probably
1:00:29 update that thing a lot and it's going
1:00:31 to get better and better and
1:00:33 better I'm one of your students I
1:00:35 started yesterday very cool students in
1:00:38 what in
1:00:42 this this is welcome welcome to to to
1:00:47 old guy
1:00:51 University where are you learning I
1:00:53 don't know some old guy on t Tick
1:00:56 Tock what's he charging you seems to be
1:01:00 free but anyway
1:01:05 cheers I actually did a class today um
1:01:09 with a with a with a bunch of guys that
1:01:11 I work with doing um automations on
1:01:14 zapier it's and it was really fun really
1:01:17 really
1:01:18 fun can you tell chat GPT there's
1:01:21 Amelia's wife hello there
1:01:26 um can you tell chat GPT to only use
1:01:29 certain sources like math props not the
1:01:32 students so Dr J what you're talking
1:01:35 about there's
1:01:36 actually um there's a dude named Andre
1:01:40 karpathy who's one of the founders of of
1:01:43 chat GPT of open Ai and he was speaking
1:01:46 at the Microsoft developer conference
1:01:48 and he used that exact example to
1:01:52 explain why having chat GP act like a
1:01:55 certain Persona works and so I'll I'll
1:01:58 explain it to you the best I understood
1:01:59 it this is I'm dramatically paraphrasing
1:02:02 what Andre carpy was talking about so so
1:02:04 I might might not have some of the some
1:02:06 of the pieces right but this is the
1:02:08 basic
1:02:09 concept so the way large language models
1:02:11 work and and the the metaphor that I've
1:02:14 come up with is imagine the Grand Canyon
1:02:17 is full of words okay so so they they
1:02:21 took these large language models they
1:02:22 trained it on all of the internet right
1:02:24 and so it looked at all of the words on
1:02:26 all of the internet and it put it into
1:02:28 this basically mathematical formula um
1:02:31 called a vector space which you know you
1:02:33 know like an XY coordinate thing where
1:02:34 you plot numbers on a on an XY
1:02:37 coordinate it's like that only instead
1:02:38 of two Dimensions it's thousands of
1:02:40 Dimensions so so just imagine this
1:02:43 massively complicated mathematical space
1:02:46 that's all these semantic clusters of
1:02:48 words so you might have the word dog in
1:02:50 this cluster it means it's a fuzzy
1:02:52 animal you keep in your house and in
1:02:54 this cluster words it means it's a guy
1:02:56 that did his girlfriend wrong right he's
1:02:58 a dog right so so it sort of knows all
1:03:01 the words and it's got all these sort of
1:03:02 semantic
1:03:04 clusters So when you say here's a math
1:03:08 problem I want you to solve it it's kind
1:03:09 of like you're in a helicopter above the
1:03:11 Grand Canyon and you're really high up
1:03:13 and it's looking at all of the Grand
1:03:17 Canyon so you're seeing all of the words
1:03:18 and you're seeing all of the possible
1:03:20 semantic areas and and exactly what you
1:03:23 said there what what what karpathy said
1:03:26 in this talk to to Microsoft was
1:03:30 um in all of those clusters the
1:03:34 professor that solved that problem
1:03:36 correctly that's going to be in there
1:03:38 and then all the students who didn't
1:03:40 solve that problem correctly are also
1:03:41 going to be in there so if your
1:03:43 helicopter is too high it's G basically
1:03:45 going to average in all of those math
1:03:48 problems and give you some sort of
1:03:49 answer there's a good chance it's going
1:03:51 to be wrong cuz it pulled in the student
1:03:53 stuff as well but if you said say I want
1:03:55 you to act like a Stamford mathematics
1:03:56 professor that specializes in this
1:03:58 branch of mathematics and I want you to
1:04:00 help solve this thing what you're
1:04:02 essentially doing is lowering lowering
1:04:03 your helicopter into one of the side
1:04:05 Canyons that's just you know about
1:04:07 Stanford Professor mathematics and your
1:04:11 chances of getting that thing go up
1:04:13 dramatically so great question and
1:04:15 that's exactly how it works and that's
1:04:17 why so this prompt. chat thing is really
1:04:21 just a long list of act like a certain
1:04:24 kind of act like a novelist act like a
1:04:27 mathematician act like a somaler act
1:04:29 like a gardener and that's exactly what
1:04:32 it's doing is it's narrowing that
1:04:34 semantic region so that your answers are
1:04:38 more accurate and also
1:04:40 gp4 is much better at not hallucinating
1:04:45 telling lies than GPT
1:04:49 3.5 I was asking you to lead me to the
1:04:53 author tool you mentioned oh um it's
1:04:56 called GPT
1:04:59 author let me go see it should be on
1:05:03 GitHub GPT
1:05:13 author with no
1:05:22 [Music]
1:05:23 space
1:05:27 hold please let me go find
1:05:31 it let me go to Google
1:05:49 collab it's GPT Dash author hold please
1:05:56 GPT Das author
1:06:07 AI okay there's a medium article here
1:06:11 hang on I'm going to flip my camera
1:06:13 around so you can see this because it
1:06:15 it's a fun tool to it's a fun tool to
1:06:17 play with you you have to dig through
1:06:20 getting it set up on Google collab but
1:06:22 I'll show you how that works um
1:06:27 so this it's a medium article GPT author
1:06:31 a constrained AI tool for generating
1:06:33 instant fantasy novels so go search for
1:06:38 go search for GPT D author and Ai and
1:06:41 this this will come up this medium
1:06:43 article and then it will link you over
1:06:48 to the GitHub link so the GitHub link is
1:06:52 GPT Das author
1:06:56 mumer github.com
1:06:59 mumer
1:07:01 dgpt
1:07:06 [Music]
1:07:11 athor now is this the
1:07:16 one yeah this is the
1:07:18 one so there's so I've got one that's a
1:07:21 copy of GPT author version 1.1 I would
1:07:25 assume that that's probably higher than
1:07:26 that right now but but the way um Google
1:07:29 collab works is you have these little
1:07:32 cells that have little play buttons and
1:07:35 the play buttons are actually running
1:07:37 code and like loading libraries and
1:07:39 stuff like that so like see how it's
1:07:43 twirling there and then it'll get a
1:07:44 little green check
1:07:47 mark when it's done see it's it's like
1:07:49 it's like doing some [ __ ] it's
1:07:51 installing some things this so so this
1:07:54 is running on my Google Drive it's
1:07:56 Google collab so I ran that cell it's
1:07:59 got the code in it and then the next
1:08:01 cell I run it and it's you know it'll do
1:08:04 its
1:08:06 thing building a wheel for Google trans
1:08:09 whatever that is that gave it a check
1:08:11 mark now I'm choosing my primary
1:08:15 language and then here this version that
1:08:18 I have you type in your author The
1:08:21 Prompt itself like what do you want the
1:08:23 story to be about what the writing style
1:08:26 so those are your inputs and that's all
1:08:27 you have to input like you know what's
1:08:29 your story about what's the writing
1:08:31 style and
1:08:34 then you keep running these cells and is
1:08:38 this the last one no it's not the this
1:08:41 is sort of loading the
1:08:43 prompts and I think is this the one
1:08:51 no so this section with all the these
1:08:54 codes in it these are all the prompts
1:08:55 that write all the
1:08:58 chapters so so this is where you could
1:09:00 modify like all the things in red here
1:09:04 are all the prompts that write the the
1:09:06 book and then the final thing book
1:09:08 writing is hit play and now it's going
1:09:12 to start writing my
1:09:16 book so that's kind of how collab works
1:09:18 is you've got these cells with code in
1:09:20 them that people that are geekier than
1:09:23 me wrote and then you hit the play
1:09:25 buttons on the cells and occasionally it
1:09:27 will screw up and then you can go to
1:09:29 chat GPT and have it help you debug
1:09:32 while it's screwed
1:09:33 up um and what the other cool thing this
1:09:36 does is as it's doing each of the steps
1:09:38 it's telling you how much they cost so
1:09:40 it's calculating the cost from open
1:09:43 AI so this is each step here is about a
1:09:46 third of a cent you know a fourth of a
1:09:49 cent a third of a cent a tenth of a
1:09:52 cent generating storyline with chapters
1:09:55 and high level details so so basically
1:09:58 now it's going through and writing a
1:10:00 book so it's just just absolutely insane
1:10:04 and this is like that thing of of you
1:10:07 know it having all these steps in it and
1:10:10 and basically I just say I want you to
1:10:11 write a book kind of like this uh it'll
1:10:14 just go do that there are all sorts of
1:10:16 crazy fireworks going on right now or
1:10:18 we're being attacked in
1:10:22 Denver I didn't I don't know what's
1:10:24 going on Friday night oh maybe it's a
1:10:25 ball game it's a Fourth of July weekend
1:10:27 for a ball game or
1:10:30 something maybe the Rockies won all
1:10:33 right so hopefully that helps let me
1:10:37 turn on my My Little Secret background
1:10:43 [Music]
1:10:45 here is someone there
1:10:49 no what is the most fun thing you can
1:10:53 think of with with using
1:10:58 AI right now for me
1:11:01 it's being with people when they have
1:11:04 this
1:11:07 moment that's why I do
1:11:12 this
1:11:18 it for me it's all about the possibility
1:11:23 of I think it's really easy to fall into
1:11:25 the Trap of this is evil it's not
1:11:30 perfect it's going to [ __ ] everyone's
1:11:32 job up like it's really easy to go into
1:11:34 a negative place with it and what I'm
1:11:37 finding
1:11:38 is the more I dig into it the less it
1:11:41 feels like that the more hopeful I feel
1:11:44 like the more I learn the more hopeful I
1:11:46 feel because I see what it makes
1:11:50 possible and so for me it's not any
1:11:53 specific thing it's like it's like it's
1:11:55 that Journey it's the Curiosity and the
1:11:59 exploration
1:12:05 and imagining what the future could be
1:12:08 like if this bends in a positive
1:12:14 direction and I think that every person
1:12:18 every individual right now that gets
1:12:20 curious about this and bends it even
1:12:23 slightly in a POS positive
1:12:25 direction will have a disproportionately
1:12:28 positive
1:12:29 impact on the trajectory of
1:12:35 AI so I think that's it like I get
1:12:38 really philosophical about this stuff
1:12:39 because I there's a couple of things I
1:12:42 think we are incredibly lucky to live at
1:12:46 a time when we get to play with tools
1:12:49 this powerful we there there has never
1:12:51 been anything in in the history of human
1:12:54 it that's been remotely close to as
1:12:56 powerful as what that is it's it's like
1:12:59 a Humanity amplifier it's like a
1:13:01 Humanity reflector if you think about
1:13:02 what that is it was trained up on all of
1:13:04 the internet so think about all of the
1:13:06 data and content that's been jammed into
1:13:09 the internet for the past 30 40 50 years
1:13:13 it all got sucked up it all got vacuumed
1:13:15 into that tool and now we can play with
1:13:21 it oh there is someone over there hang
1:13:23 on I'll be
1:13:26 what you
1:13:31 do I'm not listening to you okay that
1:13:33 was crazy those fireworks were nuts
1:13:35 weren't they I was like is there someone
1:13:38 over there there
1:13:39 was um so yeah so so for me that's it um
1:13:44 someone asked about future
1:13:52 pedia.com get out it every day um most
1:13:57 of the AI tools right now are just
1:14:00 horrific pieces of [ __ ] so just know
1:14:03 that out of the 3700 tools there there's
1:14:06 probably 200 that are good but the 200
1:14:09 that are good are going to be
1:14:12 profound right so so you're going to
1:14:14 have to sift through some crap it's like
1:14:17 finding that an app in the Android store
1:14:19 like you know it's you know 98% [ __ ]
1:14:22 it's it's going to there's going to be a
1:14:23 lot of that for a while with this stuff
1:14:25 cuz everyone's so excited about it I got
1:14:28 an idea for an app oh chat GPT wrote it
1:14:30 for me let me put that out in the world
1:14:32 oh it's a horrific piece of
1:14:34 [ __ ] yeah me a lot of
1:14:37 that
1:14:39 um yeah I I you know that's fun for me
1:14:41 right now is uh like future gazing with
1:14:47 AI does 100 pounds of flour make a Big
1:14:52 Biscuit that's you s Ross per does 100
1:14:54 pound of flour make a
1:14:56 biscuit that's a Gen X joke nobody knows
1:14:59 who Ross per is anymore where is the
1:15:01 fiction tool I just showed it I'm not
1:15:03 doing Comics
1:15:05 LOL um I thought you said Comics but
1:15:08 maybe not maybe I
1:15:14 misread he answers questions I
1:15:18 do can someone catch me up just got here
1:15:21 hey an how's it going Kyle can you host
1:15:24 workshops I can I've thought about it
1:15:29 um I keep thinking about it it's I I
1:15:32 like I don't want to host host one if
1:15:34 it's shitty
1:15:38 so um amazing how incredible new
1:15:41 discoveries we can learn can open
1:15:44 opportunities okay cool I'll be checking
1:15:46 that out how does AI categorize the data
1:15:49 correctly I.E it really comes from a
1:15:52 processor of mathematics
1:15:54 yeah I can if do you want me to if you
1:15:57 guys want me to show you I can actually
1:15:58 show you how it works I've got a slide
1:16:00 that I did for a presentation on how
1:16:01 chat jpt works when you understand how
1:16:04 it works it's even more insane that it
1:16:06 does what it does thank you for sharing
1:16:08 your brain with us um if you want me to
1:16:11 do that I'm happy to share it um to to
1:16:15 show you how it
1:16:17 works um he's talking
1:16:20 AI he answers questions I know but which
1:16:23 subject I'm just talking about I'm just
1:16:25 answering
1:16:27 stuff okay uh hang on I'll be
1:16:32 back like does he make me for a long
1:16:35 time yeah he wants you to take off I'll
1:16:39 do that later let me finish okay
1:16:44 I'll all right I'm back I got to go soon
1:16:48 got family duties all right um this is
1:16:52 great will you make a this video
1:16:54 available so we can rewatch it and learn
1:16:56 more so here's the good news I found out
1:17:00 three days ago that all of these lives
1:17:02 are actually recorded and it makes
1:17:04 little Snippets of them and I've been
1:17:06 posting them to my channel and I'll I'll
1:17:08 be able to download all these videos so
1:17:10 I will probably start a YouTube channel
1:17:13 or a playlist and I'll put all these up
1:17:15 there so at some point I will they're a
1:17:17 bit rambly right now so I don't know the
1:17:19 value of that but but we'll see so I
1:17:22 don't know that I'd want to watch
1:17:24 whatever the [ __ ] I'm talking about for
1:17:26 two hours cuz CU I tend to stay on here
1:17:28 a lot and I don't really know what I say
1:17:30 but I'll go I'll go take a look at them
1:17:31 so at some point I'll do that yeah and
1:17:33 I'll think about doing a workshop um I
1:17:36 just need to figure out like what's the
1:17:38 what's the model where do I do it what's
1:17:40 this what's the topic things like that
1:17:42 um looking at it glass half full I agree
1:17:45 I've started using it a lot more over
1:17:46 the week since great oh since watching
1:17:48 your stream before this about six months
1:17:51 ago I thought it was just Google it's
1:17:54 it it's so [ __ ] deceptive it looks
1:17:56 like it's just Google and then you and
1:17:58 then you see someone it writes a poem
1:18:00 and you're like oh it's fancy Google and
1:18:02 it's not until you go well wait a minute
1:18:05 what what's what's something I do I'm a
1:18:08 project manager and then you type in you
1:18:10 know give me a project plan for blah
1:18:12 blah blah out it comes and you
1:18:16 [Laughter]
1:18:17 go that's the best [ __ ] moment that's
1:18:20 the thing that makes me happiest is
1:18:22 watching that I had I had to have with a
1:18:23 10-year-old kid who literally went like
1:18:29 this when he looked at me and goes this
1:18:31 changes
1:18:33 everything it's crazy it really is crazy
1:18:37 who created this I don't know what
1:18:39 created what is there an AI
1:18:41 platform that can
1:18:43 create characters for a book visual
1:18:48 characters um Mid journey and stable
1:18:51 diffusion can um
1:18:54 um it's a little problematic to get that
1:18:57 character to repeat consistently in
1:18:59 different locations it there there's
1:19:01 ways to do it but it's it's really
1:19:03 complicated so but yeah if you want to
1:19:05 do initial character development for a
1:19:08 book you can absolutely do
1:19:10 that um Mid journey is the one that I
1:19:13 would recommend you have to generate
1:19:15 your images on Discord so to use midj
1:19:18 journey you have to learn Discord and
1:19:19 then you have to learn how to use mid
1:19:21 journey within Discord and if you've
1:19:24 ever seen Discord which is just like
1:19:26 geekier version of
1:19:27 slack um you're like why are they doing
1:19:31 that there and that is the right
1:19:34 question but it's the best and then if
1:19:37 you want to use stable diffusion you
1:19:39 need a computer science degree not
1:19:41 really but you got to be able to run
1:19:42 notebooks and do some geeky stuff to
1:19:45 install it all right is there an AI
1:19:48 platform okay that that one I just
1:19:49 answered AI to build Advanced what was
1:19:52 that advanced technology sounds great
1:19:56 yeah it
1:19:58 yeah I mean so here's here's part of the
1:20:01 challenge of where the technology is
1:20:03 right now especially with this
1:20:05 generative AI stuff the the the new
1:20:07 stuff like in the last 6 months is it's
1:20:11 it's so powerful that you can see
1:20:16 glimpses of what it's going to be and
1:20:17 you're like oh this is going to totally
1:20:19 transform whatever project management or
1:20:22 programming or whatever and then when
1:20:24 you actually go to implement it it's
1:20:26 like 90% brilliant and 10% like it's not
1:20:30 quite right that last 10% is like if
1:20:34 you've ever remodeled a house it's like
1:20:36 getting to 90% we're pretty much on
1:20:39 track that last 10% takes you like four
1:20:43 years I think I think that's what's
1:20:45 going on with AI right now that it's
1:20:46 really easy to get conceptual stuff
1:20:49 close but then to get it to actually be
1:20:51 good software that's that's going to
1:20:53 take a
1:20:54 while I like your personality please
1:20:56 host a workshop oh thank you I
1:20:58 appreciate that I will okay all right
1:21:00 I'll put one together I got to figure
1:21:01 out what it what it is how long it is is
1:21:03 it like multi- weeks how deep do we go
1:21:05 is it just you know basic [ __ ] I I don't
1:21:09 know you know here here's here's my
1:21:11 request to anyone that's here that that
1:21:13 wants a workshop is is connect with me
1:21:15 on Tik Tok here and and DM me and just
1:21:19 say you know if you're if if I were to
1:21:21 do a workshop what would you want it to
1:21:23 focus on what would you want it to be
1:21:24 maybe I can get some input because I
1:21:26 don't want to just put something out
1:21:27 there and have it be a waste of anyone's
1:21:30 time
1:21:31 um he has to he has to do a workshop
1:21:35 fine I'll do a workshop that'll be
1:21:37 helpful please don't discount the value
1:21:40 this information is awesome great well I
1:21:43 always discount everything because shame
1:21:45 is my love
1:21:47 language just it's the way it is don't
1:21:51 overthink immediate practical Zoom yeah
1:21:55 I know I yeah you're right I'll do that
1:21:57 I'll put something together um I don't
1:22:00 think you're rambling this is hard to
1:22:02 understand because it's at its infancy
1:22:04 yeah it really is I'm all about a
1:22:05 workshop okay great um Workshop Workshop
1:22:08 it seems to be that seems to be a a
1:22:10 theme emerging in in the comments can I
1:22:12 make a professional logo for my company
1:22:14 with
1:22:15 AI you can
1:22:18 [Music]
1:22:19 um there's there's some issues with logo
1:22:22 stuff so so doing things in mid journey
1:22:24 and stable diffusion it's just
1:22:25 generating bit map images it's not
1:22:28 generating Vector images right and if
1:22:32 you just like what I would say is this
1:22:34 what what you can what you can do in mid
1:22:38 Journey with logo development is you can
1:22:42 concept
1:22:44 brilliantly um but if you really want to
1:22:46 like tie in a logo and get the
1:22:48 typography right like it won't do
1:22:50 typography at all like it just these
1:22:52 things don't do typog photography at all
1:22:54 the image itself you could probably do
1:22:56 quite well um but in the end you're
1:22:59 going to have to hand that to a designer
1:23:00 to really turn that into vector and
1:23:02 fine-tune it and things like that
1:23:03 depends how anal you want to get with
1:23:05 The Branding um but yeah yeah you can
1:23:08 what I would what I would recommend with
1:23:10 that is go to um if you're on Facebook
1:23:14 look up the mid Journey official group
1:23:17 on Facebook and then within that group
1:23:20 look for logo development and you'll
1:23:22 find find a whole bunch of people doing
1:23:24 it and probably within there you'll find
1:23:26 a whole bunch of
1:23:27 prompts um the other thing to do with
1:23:30 mid journey is because it's on
1:23:33 Discord there's a whole Community there
1:23:35 and there's all these different channels
1:23:37 so there's probably a channel within mid
1:23:39 Journey actually let me go look there's
1:23:41 probably a channel within mid Journey
1:23:44 on
1:23:49 um like logo development or you know
1:23:53 let's see abstract characters image
1:23:57 prompts group
1:24:00 jams yeah no I don't see anything
1:24:02 specific to logo development but there's
1:24:05 a show and tell there's paintovers
1:24:07 there's image jams there's prompt
1:24:10 FAQs ideas and features so that's so so
1:24:13 within the mid Journey Discord there's
1:24:16 there's lots and lots of things you can
1:24:17 also join the AI Salon that that I
1:24:20 created if you go to the salon. I in
1:24:23 fact let me put put this up behind me
1:24:30 [Music]
1:24:32 here this if you go to that URL that's a
1:24:36 link tree to um a community that I've
1:24:40 created of AI enthusiasts and things
1:24:43 like that we meet every other week
1:24:45 There's a Discord server where you can
1:24:47 geek out and ask questions but I would
1:24:50 do that so you can you can definitely do
1:24:52 that yeah I made logo but it doesn't
1:24:53 like letters Shameless Works um are
1:24:56 there any certifications you know about
1:24:58 that generally that you about that
1:25:01 generally proves you know a lot about AI
1:25:03 not right now well okay two different
1:25:06 flavors of AI there's there's pre-chat
1:25:09 GP Ai and there's post that's kind of
1:25:12 how I'm I'm I'm I'm looking at it right
1:25:14 now pre-chat GPT you're talking about
1:25:17 machine learning and Ai and things like
1:25:20 that if you're an engineer and you're
1:25:21 and you want to go more on the tech
1:25:22 technal side that's a whole different
1:25:24 conversation and you know you should
1:25:26 have engineering background and
1:25:28 understand models and neural networks
1:25:31 and [ __ ] like that if you're talking
1:25:32 post chat
1:25:34 GPT um if if if there are uh
1:25:39 certificates they're not worth anything
1:25:42 because anything you learn right now is
1:25:44 going to be you know worthless 6 months
1:25:46 from now because everything's changing
1:25:47 so
1:25:49 fast what I think is a really good best
1:25:51 practice right now is get really curious
1:25:53 about this stuff get as good as you can
1:25:55 about it and start posting what you're
1:25:58 learning start sharing what you're
1:25:59 learning um if people at work are
1:26:02 curious about what you're doing say hey
1:26:04 do you want to learn something sit down
1:26:05 I'll I'll sit down with you for an hour
1:26:07 I'll teach you how to do it start
1:26:09 putting yourself out there that you're
1:26:10 learning about this AI stuff that you're
1:26:13 doing it demonstrating that you've got
1:26:16 some Acuity with this stuff is the
1:26:18 certification right now that's how early
1:26:20 we are a year or two from now there's
1:26:22 going to be you know decent courses and
1:26:25 things like that but right now it just
1:26:28 everything's in flux everyone's trying
1:26:29 to figure out what the [ __ ] this thing
1:26:31 is how it works how you do it how you do
1:26:34 it well how you get it not to
1:26:36 hallucinate what the hell hallucination
1:26:38 even means all that sort of stuff like
1:26:41 um so so so I think it's on individuals
1:26:46 right now to get curious about it and
1:26:47 start teaching one of the one of the
1:26:49 people that was in here um in in the
1:26:51 lives has she she set a goal for herself
1:26:53 to teach a thousand people about Ai and
1:26:57 in doing that she's now doing um
1:27:00 workshops and things for corporations at
1:27:02 like 10 grand a clip so like just go
1:27:07 just
1:27:09 go am I an educator or teacher I am an
1:27:13 ex
1:27:14 actor I'm actually a founder and CEO of
1:27:17 a video technology company called
1:27:20 storyvine and I I haven't done any
1:27:23 formal teaching but I'm I'm really
1:27:25 passionate about this stuff and I think
1:27:28 it's I'm passionate about demystifying
1:27:32 it because it drives me [ __ ] crazy
1:27:34 when I hear people demonize it or
1:27:37 diminishing it without having tried it
1:27:39 it makes me insane because this shit's
1:27:41 profound it's going to change everything
1:27:43 it's going to change everything it's
1:27:45 going to change the definition of work
1:27:47 it's going to change the value of what
1:27:48 we do and how we contribute it's going
1:27:51 to change the economy it's going to
1:27:53 change businesses and it's coming a
1:27:56 shitload faster than I think anyone's
1:27:58 prepared for and so I'm just super
1:28:02 passionate
1:28:03 about you know I someone was on the live
1:28:06 last night he like is all you talk about
1:28:07 chat GPT for
1:28:11 now
1:28:13 because the thing that sits at that URL
1:28:17 is so profoundly powerful I could talk
1:28:19 about it for the next decade and not
1:28:21 wear it out
1:28:23 I don't understand what it is or or how
1:28:25 deep it goes at all like at all at
1:28:30 all like I barely scratched the
1:28:34 surface are there 8,000 other things to
1:28:37 learn yeah that's why I put this URL
1:28:39 here if you want to geek out on all the
1:28:40 other [ __ ] but but understanding just
1:28:43 the basics of what large language models
1:28:45 are and what they do and what they make
1:28:47 possible and how they're going to change
1:28:49 your job and your job and your job
1:28:53 that's what I think we all anyone who
1:28:57 doesn't do that is going to get
1:28:58 blindsided by this thing and it's going
1:29:00 to be really [ __ ] painful and that
1:29:02 makes me very sad I don't like I don't
1:29:05 want it but it's coming and so I want as
1:29:09 few people as possible to be hit by that
1:29:12 Title
1:29:14 Wave
1:29:16 uh let me do a couple more questions
1:29:19 here I'll do another 10 10 minutes or so
1:29:24 um oh that was you that was you and
1:29:25 that's cool yeah the the the the goal of
1:29:28 teaching a thousand people like far out
1:29:30 yes that [ __ ]
1:29:34 awesome um what's the elevator pitch for
1:29:38 AI that's interesting well so I wouldn't
1:29:41 call it AI like so there's so chat
1:29:45 GPT H that's really interesting it's the
1:29:48 all this stuff is more machine learning
1:29:49 than
1:29:51 AI
1:29:53 um chat GPT is is the word version but
1:29:59 like
1:30:01 the it's a great
1:30:04 question the Transformer model hang on
1:30:08 dog's being loud um the Transformer
1:30:11 model that Google invented in 2017 is
1:30:14 kind of the foundation for all of the
1:30:15 tools that we're playing with right now
1:30:17 so chat GPT um Mid Journey stable
1:30:21 diffusion the video things the medical
1:30:23 things all the [ __ ] that's happening
1:30:25 right now is is because of this thing
1:30:27 called the Transformer so they're taking
1:30:29 these machine learning models they're
1:30:31 they're they're basically pointing them
1:30:33 at massive data sets and then when you
1:30:36 point it at like Radiology images it
1:30:38 gets really good at reading X-rays and
1:30:40 when you point it at brain scans it
1:30:42 understands what the brain is seeing and
1:30:44 when you point it at all of the words on
1:30:46 the internet it [ __ ] understands all
1:30:48 of the words and when you do it with
1:30:50 pictures it understands all of that or
1:30:52 it doesn't doesn't understand it it it
1:30:54 it lets you create it with that okay
1:30:58 so the elevator
1:31:01 pitch the thing I've been saying
1:31:03 recently is it's it's a Humanity
1:31:07 amplifier or it's like a talent
1:31:09 accelerator or it's
1:31:21 a
1:31:23 like like like I keep going back there's
1:31:25 a quote from Arthur C Clark that any
1:31:27 sufficiently advanced technology is
1:31:29 indistinguishable from Magic it's
1:31:32 that it's
1:31:36 that when
1:31:38 you like here
1:31:40 watch watch
1:31:45 this like I've done this a lot and every
1:31:48 time I do it it still blows my [ __ ]
1:31:51 mind
1:31:54 all right here's chat GPT we're just
1:31:55 using GPT
1:31:58 3.5
1:32:00 um
1:32:01 write
1:32:06 a python
1:32:09 application that
1:32:14 um
1:32:18 calculates
1:32:21 the um um
1:32:27 navigation for a
1:32:30 rocket to land on the moon I don't even
1:32:34 know if it's possible that it can do
1:32:41 that
1:32:45 so calculating the navigation for a
1:32:48 rocket to land on the moon involves
1:32:50 complex calculations and consider ation
1:32:52 well I can provide you with a simple
1:32:54 example of a python application please
1:32:56 note that the actual moonlanding
1:32:58 navigation requires sophisticated
1:33:00 algorithms and data right that go beyond
1:33:02 import Mass so it Imports a library then
1:33:04 it's like it's got the moon's radius it
1:33:07 got it's got the moon's gravity it's got
1:33:08 total L latitude and longitude if you're
1:33:10 an actual astrophysicist this is
1:33:13 probably comically simple but if you're
1:33:15 someone like me that has no [ __ ] clue
1:33:19 how I would even begin to approach this
1:33:21 I at least can look at this and go okay
1:33:23 I understand at least what they're
1:33:24 trying to do there
1:33:27 um you know let's say um
1:33:32 write uh let's see write um a sub
1:33:39 routine um that will orbit the Moon
1:33:44 three times before
1:33:48 landing can it do that yeah apparently
1:33:54 I
1:33:55 mean and I mean the thing about
1:33:59 code is you can run it right so is that
1:34:05 code correct I don't know it's probably
1:34:06 got mistakes in it there's probably
1:34:08 libraries missing it's probably
1:34:09 oversimplified it's probably not great I
1:34:12 know that if you do code three times in
1:34:14 a row it'll take three different
1:34:16 approaches
1:34:18 but probably within an hour I could get
1:34:21 some sort of simulation working with
1:34:23 some sort of rocketry simulator which it
1:34:26 will tell me exists and put that code in
1:34:28 it and it will do something that's
1:34:31 [ __ ] miraculous and it and like
1:34:33 that's that's a teeny sliver of
1:34:36 something it does the same [ __ ] with
1:34:38 language H have it write you Shakespeare
1:34:41 like stuff it can do that have it write
1:34:44 you in the style of Mary Shelly have
1:34:47 it write you a marketing plan have it
1:34:49 write you tweets and social media a
1:34:53 copy
1:34:56 it's like it's it's just this Limitless
1:34:58 hole of possibility and and and it's
1:35:02 just like it's just one
1:35:04 tool that in the next three to five
1:35:08 years is going to get increasingly
1:35:09 sophisticated it's going to be
1:35:11 multimodal like right now it just does
1:35:13 text it's going to be able to spit out
1:35:16 images and then animations and then
1:35:18 films and then
1:35:20 music and then it's going to connect to
1:35:22 all the [ __ ] that you're connected to
1:35:24 and it's you're going to give it
1:35:25 permission to access your bank and your
1:35:28 travel accounts and your grocery [ __ ]
1:35:30 and you're this and you're that and
1:35:31 you're just going to
1:35:33 like this is like Jetson [ __ ] you just
1:35:37 talk it and it will go do things for you
1:35:40 like that's where we're
1:35:43 going
1:35:45 um
1:35:47 yeah so I so back to what's the elevator
1:35:51 pitch for AI I have no answer for you
1:35:54 that was not an elevator [ __ ] that was
1:35:56 a [ __ ] that was a [ __ ] di
1:36:00 tribe I think it's a human it's a it's a
1:36:04 it's a Humanity amplifier I think it
1:36:07 takes who we are and what we've done and
1:36:10 repackages it so that every single
1:36:13 person can tap into that genius maybe
1:36:16 that's it I don't know it's a great
1:36:19 question with a shitty answer let's what
1:36:22 application allows to go beyond the text
1:36:24 restrictions of
1:36:26 GPT well so with plugins you can go
1:36:30 beyond the the text restrictions here
1:36:32 let me let me show you another thing I
1:36:34 I'll this is this is actually quite
1:36:36 simple what it's doing but this this is
1:36:38 going Beyond
1:36:41 GPT the the text restrictions of there's
1:36:45 a couple of them okay so we're going to
1:36:46 we're going to play with some plugins
1:36:48 now so if you haven't played with
1:36:50 plugins on chat GPT the the way it works
1:36:52 is you go to gp4 if if you're if you've
1:36:55 paid for the plus thing and you've
1:36:57 enabled your beta uh Services you can
1:37:00 you can play with plugins so I'm going
1:37:03 to go to plugins and when you play with
1:37:04 plugins you have this little drop- down
1:37:06 menu and here's the ones I have
1:37:08 installed and then in any s in any
1:37:11 session I can turn on up to three so I'm
1:37:14 going to turn on
1:37:15 instacart I'm going to say
1:37:18 make me a meal
1:37:20 plan that looks like dog food but is
1:37:28 healthy I don't
1:37:31 know you should also be having fun as
1:37:33 you do this it sounds like you're
1:37:35 looking for a meal plan that might
1:37:37 resemble the appearance of dog food but
1:37:39 is actually healthy all right scrambled
1:37:42 tofu and vegetables lentil stew this is
1:37:45 good ground turkey vegetable
1:37:48 Skillet um and then you'll notice what
1:37:51 it did down here would you like me to
1:37:53 create a shopping list for these meals
1:37:54 on instacart and I'm going to say
1:37:58 sure and now what it's going to do is
1:38:00 it's going to fire off the instacart
1:38:06 plugin and it's going to take my little
1:38:08 meal plan and turn that
1:38:10 into a shopping
1:38:13 list and it's going to give me the the
1:38:16 hyperlink to the shopping list so this
1:38:18 is all text but now it's going to pop me
1:38:20 over to in
1:38:25 cart
1:38:26 and there's all of my ingredients for my
1:38:29 dog food meal plan right and those could
1:38:32 be on
1:38:33 my um porch within the hour so so
1:38:38 there's one um there's also a
1:38:41 plugin I haven't played with this one
1:38:44 very much so I don't know it but let me
1:38:45 turn off this one there's a plugin here
1:38:49 called show me diagram
1:38:53 I'm going to say
1:38:56 um
1:38:57 create
1:39:00 um a flow
1:39:03 chart
1:39:05 um of the House of
1:39:10 Representatives or no let's say of the
1:39:14 US Senate because it's
1:39:19 smaller and that should find off that um
1:39:24 show me diagram thing and it will go I
1:39:26 assume what it's doing in the background
1:39:29 is doing some sort of search
1:39:34 of something mermaid graph so so it
1:39:38 pulled a graph
1:39:39 in oh there you go so there's voters
1:39:42 State selectors
1:39:44 bills you can view this F diagram in a
1:39:48 new tab so there's the diagram in a new
1:39:50 tab so so it's remedial right now but
1:39:53 and then and then and then there's other
1:39:55 things in there there's there's hooks
1:39:57 into stable diffusion things like that
1:39:59 so um so right now all of these tools
1:40:03 are very very remedial but but there's
1:40:06 there's ways that you can kind of you
1:40:08 know fake what they're they're where
1:40:10 they're headed where they're going to
1:40:13 be so I don't know if that actually did
1:40:16 what you wanted it to do but you
1:40:18 know but but it's free this is a free
1:40:22 class uh What applic uh okay we just did
1:40:26 that one Microsoft has a certification
1:40:28 for teachers I can send you the link
1:40:31 Mark
1:40:32 cool um 500 nonprofit fundraisers
1:40:35 trained on chat GPT so far this summer
1:40:37 oh that's so cool that's so cool this I
1:40:41 mean this is going to change
1:40:42 philanthropy it's going to change
1:40:44 [ __ ] everything it's going to change
1:40:46 everything it really is it's going to
1:40:48 change everything all right I'm going to
1:40:51 do two more questions and then I got to
1:40:52 go I got to go deal with the kids the
1:40:55 dog the wife all of it got to go deal
1:40:58 got
1:40:59 life life's not just Tick Tock you can't
1:41:02 just ticktock your life away what' you
1:41:05 do Friday night I don't know some old
1:41:06 guy on Tick
1:41:08 Tock weren't you listening to him last
1:41:10 night shut
1:41:15 up can you show the slid you mentioned
1:41:18 about how chat GPT is built yeah okay so
1:41:21 all right this is I'll try to do this
1:41:23 quick
1:41:25 um this is how chat GPT
1:41:28 works it's staggering it's staggering
1:41:32 how it works it's it's like the more you
1:41:35 understand about it the more wild it is
1:41:39 when you see what it actually
1:41:50 does
1:41:52 the great
1:41:53 Renaissance which is that date this was
1:41:57 sort of the money slide from this
1:41:59 presentation I gave so I gave this talk
1:42:00 at at this thing called creative
1:42:03 mornings and um compared the early days
1:42:07 of the web to the early days of
1:42:09 generative Ai and this is the date that
1:42:11 changes everything in my opinion and we
1:42:13 are entering a new age um of like
1:42:18 dramatic human productivity and
1:42:22 creativity and this was the slide I did
1:42:24 on how chat GPT works
1:42:27 so let me make this a little smaller so
1:42:30 we can see
1:42:33 it all right
1:42:37 so how this stuff works so so this gets
1:42:41 a little geeky but but I I think it's
1:42:44 worth it okay so first is what does GPT
1:42:46 stand for GPT stands for gener
1:42:49 generative pre-trained transfer forer so
1:42:52 generative meaning meaning it's
1:42:54 generating something it's generating
1:42:56 original content I'll show you how
1:42:58 that's what this is all about
1:42:59 pre-trained means it's been pre-trained
1:43:02 you don't you don't need to provide it
1:43:03 with data it's a large language model
1:43:05 that's been pre-trained on all of the
1:43:07 internet and then Transformer is the
1:43:09 technology so Transformer is the
1:43:11 technology that Google invented that
1:43:13 allows these models to be trained on
1:43:15 massive amounts of data so so this is
1:43:18 the thing that Sam Altman discovered the
1:43:20 more data they threw at it and the more
1:43:22 gpus they threw at it the better it got
1:43:25 so gpt3 and four are as good as they are
1:43:28 because of that recognition that you can
1:43:31 basically overload these systems with
1:43:33 massive amounts of data and then it
1:43:35 compresses them into these very small
1:43:37 usable things that we get to play with
1:43:40 large language model that's what these
1:43:42 things are called in general the large
1:43:43 language model is is these big models
1:43:45 trained on this that we can interact
1:43:48 with then there's this thing called the
1:43:50 latent space in the latent space this is
1:43:52 a dramatically oversimplified
1:43:54 representation of it but think of this
1:43:55 as an XY coordinate like a a
1:43:57 two-dimensional grid XY coordinate each
1:44:00 of these little clusters of dots are
1:44:04 mathematical um uh vectors of you know
1:44:09 semantically related um words now it's
1:44:12 not actually the words in the database
1:44:14 it's just the numbers so it's the
1:44:15 numbers in in this in this case it's a
1:44:18 two-dimensional space the real space is
1:44:20 thousand of Dimensions it's incredibly
1:44:22 complicated and Incredibly profound but
1:44:25 but for this example it's it's that
1:44:27 simple okay and then how this stuff
1:44:29 works is that when you put in a prompt
1:44:32 it turns it into a mathematical
1:44:35 weight or mathematical probability that
1:44:38 then goes and looks at all of the words
1:44:41 in What's called the latent space and
1:44:45 says based on that prompt which is the
1:44:48 next most probable word to fit in that
1:44:50 sense sentence and then it does that
1:44:52 over and over and over again so it's
1:44:54 it's almost identical to when you're
1:44:57 looking at um you're typing in Google
1:45:00 and and they have that auto complete at
1:45:01 the end of a sentence that it sort of
1:45:03 finishes your sentences for you um it's
1:45:05 like that only way more intense so if I
1:45:08 type in the prompt the quick brown fox
1:45:10 jumps over the lazy blank you know we
1:45:13 all know that's dog but how how this
1:45:15 system does that is it says it takes
1:45:17 that and then turns that into a
1:45:19 mathematical formula that then generates
1:45:22 a probability and basically says okay in
1:45:25 all this latent space you know all these
1:45:27 other clusters are not relevant but
1:45:29 these light blue ones here you can't
1:45:31 really see them on there but there's
1:45:32 light blue ones there um are are most
1:45:35 probable and one of those that says 96%
1:45:38 that little dot there one of those is
1:45:40 the most likely and if you look at what
1:45:41 that actually is it's a big long string
1:45:44 of numbers it's a vector and then that
1:45:46 long string of numbers is tied in a
1:45:48 database to a word and that word is dog
1:45:50 and in this case and then the thing puts
1:45:53 dog into that sentence and it does that
1:45:56 over and over and over so when you see
1:45:59 it right like like a three-page story in
1:46:02 30 second or in 3 seconds it's doing
1:46:06 what I just showed you word after word
1:46:07 after word so it's it's
1:46:09 truly generating writing original
1:46:13 content word by word just based on
1:46:16 probability this is why it hallucinates
1:46:19 because what we experience is it writes
1:46:22 something that looks coherent right we
1:46:24 we look at it and we like oh my God it's
1:46:25 like how does it know that right all
1:46:29 it's doing it's just a calculator that's
1:46:30 the next word that's the next word
1:46:32 that's the it's got no awareness of what
1:46:33 it actually wrote so when it writes
1:46:36 something wrong if it's got too much
1:46:38 crap in its Laten space it'll just write
1:46:40 in [ __ ] confidently it's like you
1:46:44 asked me for this this is the next most
1:46:46 probable
1:46:47 [ __ ] you're like that's not even close
1:46:51 that's why it hallucinates so anyway
1:46:52 hopefully that explains it and uh that's
1:46:56 a good that's a good thing to end
1:46:58 on let's see
1:47:01 here I'll pop up these URLs here at the
1:47:05 end so you can see those I'll just
1:47:07 scooch out of the way let me see if
1:47:09 there's any comments
1:47:13 here thanks I needed to needed a pitch
1:47:17 to reach
1:47:18 people as chat GPD to write a sherian on
1:47:21 it for me nice elevator pitch for a very
1:47:24 tall building yeah that was that was not
1:47:26 an elevator pitch I failed on that one
1:47:28 but whatever sounds like most English
1:47:31 food according to my non an English
1:47:33 partner that's
1:47:37 hilarious man the first time I went to
1:47:39 London I'm like what the [ __ ] is with
1:47:41 the food cuz I was living in New York at
1:47:43 the time I was like
1:47:45 really like people pay for
1:47:49 this
1:47:51 that's that's exactly my diet all right
1:47:54 you're a Fun Bunch all right listen I
1:47:56 got to run um I hope this was helpful it
1:47:59 was a pleasure have a great Friday night
1:48:01 peace out everybody I'll think about a
1:48:03 workshop if you would you know just pop
1:48:05 in what what you'd like me to do it on
1:48:07 I'll think about that I'll figure out a
1:48:08 way to do it so I don't know maybe I'll
1:48:11 do like a four- week thing or I I don't
1:48:13 know I will think about it it's a good
1:48:15 idea so appreciate it have a good one
1:48:19 everybody