aiLL 082923 - Exploring the Wild World of Generative AI and Its Profound Possibilities

Description

Chapters

Transcript

0:01 foreign
0:17 Bowski
0:31 I see Nick and Carla and AJ
0:36 I see fat Mike Hey fat Mike what's
0:40 happening
0:41 was shaking
0:43 welcome welcome welcome
0:46 I keep falling asleep during your tick
0:48 tocks it's not you
0:52 I am a good sedative
0:55 it's been said before
0:58 I do that all the time I've got I fall
1:00 asleep to David Shapiro because his his
1:02 YouTube videos are like three hours long
1:03 and I'll start watching them I fall
1:05 right out
1:13 welcome everybody welcome welcome
1:15 welcome welcome welcome
1:18 got uh we'll let it get spun back up we
1:22 were going along good but then my Wi-Fi
1:24 was poo poo
1:26 I was getting errors here people were
1:28 saying it was buffering it was horrible
1:31 now there's just six people in here now
1:33 tick tock's like we're not gonna show
1:35 your life to anyone we're just not we're
1:37 not going to show it to anyone yes
1:39 selfish little bastard
1:42 um
1:43 oh that's a little too bright yeah that
1:46 hurt
1:49 welcome welcome welcome welcome this is
1:52 the AI learning lab my name is Kyle
1:53 Shannon we'll get started here in a
1:55 second I'll talk about all this stuff
1:57 here I was just here before yeah tap the
1:59 screen if you would let people know
2:01 we're having a good time in here oh it's
2:03 a party on a Tuesday night
2:06 weren't you supposed to be doing work
2:08 tonight yeah but I didn't do it what'd
2:09 you do I watched some old fat guy on
2:11 tick tock
2:13 why would you do that I don't know
2:16 that's my life
2:19 uh my co-founder today said she she
2:22 turned this on last night and just put
2:25 me in the background I don't know why
2:27 why she would want to to spend even more
2:31 time with me during the day
2:35 that's pretty funny
2:37 we had a good we had a good call today
2:38 or a good meeting today with the with
2:40 the banker
2:42 um so yeah we've got some interesting
2:44 yield company we're gonna see if we
2:46 can't
2:47 spin up some excitement about yield
2:49 company
2:51 and then I can do Tick Tock lives 24 7.
2:55 this could be my new career
2:57 I'll be an influencer on a tick tock
3:01 hey Kyle
3:02 what do you do are you an entrepreneur
3:05 oh I used to be
3:07 what do you do now nothing
3:10 but I get paid for it
3:12 [Laughter]
3:15 oh man all right so this is the AI
3:18 learning lab if you're if you're new to
3:20 generative AI that's what we talk about
3:22 here I do my best to answer any
3:24 questions if you have questions about it
3:26 pop them in the comments below
3:27 if you're if your thoughts are circling
3:30 around who the hell are you anyway and
3:31 why should I trust you you shouldn't I'm
3:33 just a dude I'm an entrepreneur of many
3:36 years uh I was an entrepreneur in the
3:38 early days of the World Wide Web
3:40 and uh this year generative AI stuff
3:44 a lot of parallels to the early days of
3:46 the web a lot of parallels
3:48 and this is feeling way more profound to
3:51 me this is feeling like steam engine
3:53 profound Industrial Revolution profound
3:55 fire printing press like there aren't
3:57 many of them that are in this
3:58 neighborhood World Wide Web as profound
4:01 as that was
4:02 got nothing on this stuff and this
4:04 stuff's coming fast so why do I keep
4:06 pointing to that thing
4:11 okay oops
4:14 so
4:17 if you're new to generative AI if you're
4:19 like you know annoying people like me
4:21 keep going you should try it ai ai is
4:24 going to be the next big thing have you
4:25 done anything with AI or you on chat GTP
4:28 GTP gptp if you tried that that would be
4:32 good if you're annoyed at those people
4:34 well they're like that for a reason
4:37 because they've had some sort of
4:39 profound moment where they realized oh
4:42 my God everything's different
4:45 and I had that moment too and that's why
4:46 this channel exists so uh
4:49 chat.openai.com that's the official chat
4:51 website if you would be so kind if
4:54 you're willing to share this live with
4:56 your friends and family so they can
4:58 think you're a weirdo too we get some
5:01 more people in here
5:02 um
5:03 this is the official chat EBT website
5:06 it's the best if you pay their 20 bucks
5:09 a month GPT plus you get access to gpt4
5:12 which is their new fancy model you also
5:14 get access to code interpreter and
5:17 plugins and custom instructions
5:22 um
5:23 I'm not affiliated with them in any way
5:25 so there's no Shenanigans going on with
5:28 me making money that way if that's
5:30 something that bugs you it shouldn't but
5:32 it does apparently bug some people so
5:35 I'm not that guy
5:36 um all these things below are all
5:38 different competitors of chat gbt so
5:42 um claude's a competitor Bing is
5:44 Microsoft and they invested 10 billion
5:46 dollars in open Ai and have incorporated
5:49 into into their Bing chat Bard is Google
5:53 labs.perplexity is metas uh Lama two
5:56 models Pi is a company called inflection
5:59 Poe is from quora all of these have
6:02 something interesting to play with but
6:04 if you're brand new to it just go the go
6:06 to that one just go to that one and then
6:08 if you're like but I don't know what to
6:11 do there Kyle what do I do there I don't
6:12 even know what to do do you know what to
6:14 do what do I do it looks like a Google
6:16 Chat do I search for something on it why
6:19 does it say it's only up to September
6:22 2021 why oh why well there's a lot of
6:26 reasons for that but but go to this
6:29 thing prompts.chat
6:31 and that will teach you how to talk to
6:33 these chat gbt like things and how they
6:36 work and and what they can do
6:39 um and so so that's worth doing
6:42 um
6:43 oh that's funny went to go subscribe to
6:45 your YouTube channel recognized it by
6:47 your Giant Eye yeah the YouTube channel
6:48 is that address at the top there AI
6:51 learning lab Dash TT because AI learning
6:54 lab was taken by someone dang it
6:57 dang it buck you should have gone there
6:59 sooner I told you you should have gone
7:01 there sooner chick chick
7:04 anyway if you want to go to YouTube
7:08 um
7:09 I'm using chat GPT actually this is kind
7:11 of fun no I'm not using chat EBT I'm I'm
7:13 using Claude
7:15 claude.ai so what I'm doing is Tick Tock
7:19 records these uh these lives I do
7:24 and so I'm uploading them to YouTube
7:26 grabbing the transcript going to Claude
7:29 pasting the transcript in and I've
7:30 written a prompt that'll generate a
7:33 title a description and all the chapters
7:35 for all the silly [ __ ] I say throughout
7:38 the two and three hour sessions
7:41 um which actually makes them kind of
7:42 usable so
7:44 um I'm just getting started with that so
7:46 I've got some I got some work to do but
7:48 anyway that's kind of cool it's kind of
7:50 cool using this kind of technology to
7:52 make videos where I'm talking about this
7:54 kind of Technology it's very meta but
7:56 not meta like metaverse and Mark
7:59 Zuckerberg meta like the term meta you
8:02 know anyway moving on
8:05 um what else
8:06 um prompts.chat will teach you about
8:08 prompting chat GPT
8:10 um openart dot AI slash promptbook is
8:13 just a really good document if you want
8:14 to learn how to do these image
8:16 generation tools mid Journey Adobe
8:18 Firefly and stable diffusion or Leonardo
8:21 which is a a web front end to stable
8:23 diffusion um it's a really good one uh
8:26 so so those are some things to play with
8:29 um all right so that's what that's
8:31 that's what all that stuff is I keep it
8:33 up here
8:34 mm-hmm
8:38 I keep it up here throughout
8:41 this because I know people come in and
8:43 out of these things and those are the
8:46 things right now that I would recommend
8:47 if you're just trying to get started
8:49 with generative AI those are the things
8:51 to start with and you can ignore all of
8:53 them except that one just go to that one
8:56 if you're brand new to this stuff
8:57 chat.openai.com all right if you have
9:00 questions about chat gbt or anything
9:02 like that pop them in the comments below
9:03 or or if you just want to talk about
9:05 anything in particular I'm happy to go
9:08 through it a couple of things you should
9:09 know I try to answer all the questions
9:12 what does that mean it means that I get
9:15 really far behind on the questions and
9:17 it'll look like I'm just talking like
9:19 maybe this is a recording maybe he's not
9:21 real
9:22 but I'm real could I do that if I
9:24 weren't real no no I couldn't
9:28 I couldn't call on Nate filipowski
9:30 powski no I couldn't he's right there
9:35 hey I PRM Chrome plug-in is good for
9:38 helping you with prompts oh that's cool
9:39 PRM
9:41 I'm not using any plugins right now
9:42 although I probably should
9:45 all right so that's what we're doing I
9:46 also make funny voices I uh occasionally
9:49 get kicked off here because what I
9:51 think's funny Tick Tock thinks is hate
9:54 speech that happened last time and then
9:57 uh what else
9:59 the the other two times I got kicked off
10:01 were uh someone asked me about you know
10:04 are these models biased and I was
10:05 explaining
10:07 you know how they get biased and how you
10:08 deal with it and how we're going to deal
10:10 with it in the future and tick tock
10:12 thought that was bad too you could you
10:14 could be biased apparently but you can't
10:16 talk about being biased I don't know
10:18 whatever
10:21 um but I also cuss so if you don't like
10:23 cussing you know whatever
10:25 I wouldn't stay
10:30 and I don't know who it was that they
10:31 were they were watching some some
10:33 Tick-Tock lady and they left that to to
10:35 watch this that's you might want to have
10:38 yourself examined all right question do
10:40 you think AI will be in real time soon I
10:44 don't quite know what you mean in real
10:45 time do you mean image generation
10:50 uh or do you mean like talking and it's
10:53 in real time with no lag so so a couple
10:57 of things soon yes soon uh but I would
11:01 put soon you know in a relative time
11:03 scale of two to three years probably
11:07 um but but if you listen to the CEO of
11:10 stability AI the people behind stable
11:12 diffusion the model that they're playing
11:14 with internally at stability for image
11:16 generation is generating 60 images a
11:20 second
11:23 so 60 images a second that's you know
11:26 video game
11:28 quality frame rate right that's 30
11:30 frames a second this video
11:32 so they've got a version of stable
11:35 diffusion or whatever they're going to
11:37 call it that's doing 60 frames a second
11:39 that's what that's what they're playing
11:41 with internally
11:43 um so I think that we will probably
11:46 within a year
11:49 um the text to video stuff is going to
11:51 be really good the image to video stuff
11:54 is going to be really good
11:56 um
11:58 so I think within a year you'll be able
12:00 to make 30 to 60 second video clips that
12:03 have coherence
12:04 and are and are good and it'll probably
12:07 just generate them
12:09 pretty much in real time
12:11 and then if you sort of play that out
12:12 over three years
12:14 you'll probably there will probably be
12:17 systems that are multimodal where you
12:19 can ask it to
12:22 outline a book write the book
12:27 um adapt the book into a screenplay cast
12:30 the screenplay with the following people
12:32 and start playing the movie
12:35 um that's probably three to five years
12:36 out probably five years out for for
12:38 something like that
12:39 um but but
12:41 we're definitely entering an age of
12:44 hyper personalization where
12:47 um we're just gonna ask these tools to
12:48 go do [ __ ] for us they're gonna go do
12:50 them we're going to ask them to
12:53 um
12:53 behave like a certain application and it
12:56 just will or it'll just go write its own
12:58 API there's a there's a someone
13:00 mentioned a thing that
13:02 um Microsoft and UC Berkeley
13:06 um just released this white paper called
13:08 uh gorilla AI that basically just goes
13:12 and learns it's it's got a whole Corpus
13:14 of apis that it knows
13:17 and you just ask it to go do [ __ ] and it
13:19 just goes Hooks and hooks itself up with
13:22 1500 apis and goes and does [ __ ] for you
13:27 um
13:29 so I don't quite know what your question
13:31 is will it be in real time but it's
13:38 soon yeah soon
13:40 not that far out
13:44 um
13:50 um spreading all the AI gospel
13:52 Hallelujah brother Hallelujah
13:59 and by the way while I am a tech
14:01 Optimist and this is largely about
14:05 um
14:08 the good I think AI can do my you know
14:10 my eyes are not you know I am not
14:12 blinded by optimism I think there's a
14:15 lot of
14:17 um
14:19 there's a lot of pitfalls and there's
14:21 gonna be a lot of [ __ ] that happens
14:22 and there's a lot of
14:24 profound pain that's going to be
14:26 inflicted by this stuff
14:28 um but I'm also I've been on the planet
14:31 long enough to know that
14:34 is not going to get uninvented
14:37 um
14:38 as crazy ass as it is you know what the
14:41 what the writer's Guild and the SAG
14:43 actors are saying oh we're going to be
14:45 replaced we're going to be replaced
14:48 in large part that's true you know
14:50 there's still going to be a role for
14:51 writers there's still going to be a role
14:53 for actors but there's a lot of every
14:56 creative industry is going to be
14:57 dramatically impacted by this stuff
14:59 every knowledge worker is going to be
15:01 dramatically impacted by this stuff
15:03 what I'm trying to do is just get people
15:05 to use it get people educated about it
15:08 start to get literate about it because
15:10 as the disruptions come if you're one of
15:12 the people that understands what it is
15:15 and why it is and how it works and what
15:17 it's good at and what it's not good at
15:20 your chances of surviving whatever you
15:24 know tsunami comes rushing in from this
15:26 stuff are going to be dramatically
15:27 higher
15:28 and
15:30 there's a very real possibility that
15:32 just putting time into learning this
15:34 stuff over the next year 18 months is
15:37 just going to present you with massive
15:38 opportunities so you know my my sort of
15:42 soapbox on this channel is go play get
15:44 curious about it do parlor tricks with
15:47 it at first then try to have it
15:51 redo replicate what you do
15:54 until you have your Kevin McAllister
15:57 moment
15:59 remember Home Alone
16:02 and at that point where you're like oh
16:04 [ __ ] this stuff's pretty scary
16:06 that's when you lean into it that's when
16:09 you lean into it and say okay
16:11 I gotta take this serious now
16:13 how do I do what I do now with these
16:16 tools and then you'll figure a little
16:18 bit of it out you know oh [ __ ] it's
16:21 really good at that
16:23 then after a little bit more you'll go
16:24 well it's actually
16:27 not great at this part of it all right
16:30 good you just learned something right
16:31 that's an Insight that you can share
16:33 with other people it's great at this
16:35 it's not so good at that
16:37 everyone's trying to figure this [ __ ]
16:38 out right now so whether you learn what
16:41 it does well what it does not well
16:44 workarounds
16:46 ways to to use different Tools in
16:49 combination to get something that you
16:50 need I heard someone
16:52 I think it was hair growth geek or
16:54 Silver Fox yesterday said they're using
16:58 Claude to summarize long-form
17:00 transcripts and then they pop it into
17:02 gpt4
17:04 to turn it into good [ __ ]
17:06 that seems like a perfect workflow
17:09 why does gpd4 not have a hundred
17:12 thousand token context window I don't
17:14 know but it doesn't so once you learn
17:16 that you know you can go to Claude and
17:18 you can put in really long things there
17:21 and then you find out it's not quite as
17:23 good as gpt4 so you go back to there
17:25 with the shortened thing from there and
17:28 now life's all good again
17:31 if you're brand new to this stuff it's
17:33 going to sound like I'm speaking Chinese
17:36 and then you'll be like what the hell is
17:39 he talking about
17:41 I don't know if I like to say hi stuff
17:43 it makes me crazy
17:46 you know
17:48 I don't know
17:53 why do you watch that I don't know
17:56 he says it's exciting
18:00 he says everyone's gonna have to do it
18:04 [Laughter]
18:06 that's what it is
18:08 that's what it is all right just open it
18:11 up already and do us a demo all right
18:14 Daniel de Silva I will do that what kind
18:16 of demo would you like let me slide to
18:18 the bottom here so I'm gonna come back
18:20 to Daniel Da Silva just open it up and
18:22 do a demo already all right what kind of
18:24 demo do you want you want chat GPT you
18:26 want plugins you want code interpreter
18:28 you want mid-journey images I can make
18:30 boobies you want me to make boobies
18:33 thank you
18:34 tell me what you want I'll go demo
18:36 something
18:40 you want just regular old chat GPT
18:44 tell us about your T-shirt today please
18:45 this
18:47 is a Basquiat
18:50 Cassius Clay
18:54 t-shirt
18:56 [Applause]
18:57 I'm old enough that I remember watching
19:00 Cassius Clay fight before he was
19:02 Muhammad Ali I'm an old [ __ ]
19:08 um ask GPT to conceive its own demise I
19:11 could do that do a demo of the Chrome
19:13 extension AR AI PRM but search through
19:16 Google all right I might do but I don't
19:18 want to do that because installing an
19:20 extension might just take a while so
19:23 wait text to video AI which one is best
19:26 uh runwayml right now is the best all
19:29 right let me I'll go do a couple of
19:31 things here
19:32 uh make a comic avatar from a photo
19:36 a comic avatar from a photo I could do
19:38 that yeah that's kind of fun
19:40 all right let's let's see let's hop over
19:46 okay I'm gonna do a crash course on chat
19:49 gbt to start
19:51 so you're at chat GPT
19:54 you got
19:56 if you subscribe to GPT plus that's the
19:59 20 Buck a month thing you get access to
20:01 gpt4 to code interpreter to plugins and
20:04 then a thing called custom instructions
20:07 so if you haven't seen
20:09 chat GPT at all I'm just going to go to
20:11 GPT 3.5
20:13 and I'm going to just turn on my
20:14 microphone so you don't have to watch me
20:16 type
20:17 and then I'll say uh let's see
20:24 for a seventh grade reading level
20:30 so one of the things about GPT 3.5 is
20:33 it's wicked fast so you can just see
20:36 that so I just wrote that whole rap yo
20:38 listen up kids I'm here to explain
20:39 quantum mechanics it's a whole new game
20:41 tiny particles are super small acting
20:43 all weird like a bouncing ball imagine a
20:45 cat both alive and dead Schrodinger's
20:48 experiment misses with your head
20:51 particles teleport in a Flash they
20:53 switch from here to there it's a Quantum
20:55 hitch and it just goes on and on and on
20:57 it creates choruses versus so it
20:59 understands you know basic [ __ ] like
21:01 songwriting you can also do things like
21:04 um
21:06 foreign
21:08 make a table that compares quantum
21:11 mechanics and the theory of relativity
21:15 period be sure to include
21:18 the major scientists and theories
21:22 that make up the two disciplines
21:27 sure here's a table
21:29 comparing all that stuff right so let's
21:32 see how it did
21:33 main focus behavior of particles for
21:36 quantum mechanics at the atomic level
21:38 theory of relativity gravitation motion
21:40 and blah blah blah major scientist Max
21:44 Planck Einstein Niels Bohr Heisenberg
21:47 Irwin Schrodinger for quantum mechanics
21:50 Albert Einstein for the theory of
21:53 relativity
21:55 um
21:56 so it does that uh it can also do things
22:00 like um
22:01 write a python application that is a
22:04 clone of flappy birds that can run in
22:07 the unity platform
22:12 where it explains what to do
22:17 and then it writes the code for you it
22:19 didn't write all the code actually
22:22 here's a simplified example
22:26 can you make a more sophisticated
22:28 application
22:30 with Sprite graphics
22:34 so it explains what it's doing or how
22:36 you should do it
22:39 yeah I didn't really write it
22:42 many more details you know the design
22:46 that was kind of crappy let's go over to
22:48 actually here's a good example of
22:51 sometimes GPT 3.5 and great so let's go
22:55 over to gpt4
22:59 write me a python applications of a
23:02 Flappy Bird clone complete with Sprites
23:06 designed to run on Unity
23:18 oh it uses C sharp that's why it's
23:22 that's why that's why it didn't like it
23:25 let me stop it I use the I used the
23:28 wrong programming language
23:30 sorry make that application in C sharp
23:36 only one message at a time
23:39 oh come on
23:42 anyway you get the basic idea
23:44 write a c-sharp application of a Unity
23:47 of a flappy birds application to run in
23:50 unity
23:58 and gpt4 is slower
24:13 and so here's our code
24:15 all right
24:17 so I can do that all right
24:19 so just in general the
24:24 the way you interact with these things
24:26 is very much like you've got
24:28 a wicked smart but very simple intern
24:33 like an intern you need to teach how to
24:35 do everything
24:37 but it's an intern that's one of the
24:40 most educated interns in the history of
24:42 the planet
24:44 um that's kind of what it's like working
24:45 with it so so it's a lot of back and
24:47 forth talking to it okay so a couple
24:49 other things that you can do if you've
24:52 paid for you I'm going to stop that now
24:53 it's just going on and on
24:55 if you've paid your 20 bucks for for GPT
24:57 plus and you want to use some of the
24:59 advanced features the beta features you
25:00 go to your settings you go to Beta
25:02 features and you turn them on so you
25:04 turn on custom instructions plug-in and
25:06 code interpreter
25:08 and then when you go to the top here you
25:11 go to gpt4 you'll see code interpreter
25:14 and plugins you won't see custom
25:16 instructions because they live over here
25:18 so let's go show you what those are
25:21 so I've got a kind of silly thing
25:22 written in here that I can turn on and
25:24 off
25:26 um but it's basically you can tell chat
25:28 GPT a bit about who you are or or who
25:31 you want it to know what what you wanted
25:34 to know about you so you could say you
25:36 know I'm a
25:38 culinary expert or whatever it might be
25:41 and then you also can tell it how you
25:43 want it to act so whatever you put in
25:45 custom instructions if you enable it any
25:48 chat that you have within chat CPT it
25:50 will start acting like that so I made an
25:53 example one here that's just kind of
25:55 silly
25:57 um but but it but it it's pretty cool
25:59 what it does because you can see that it
26:01 dramatically impacts how Chachi BT acts
26:05 um the the prompt is I'm a master
26:07 thespian and A pompous ass I have a
26:09 large vocabulary and I Relish in
26:11 sounding smarter and more dramatic than
26:12 people around me and then how do how do
26:15 I want it to act or how do I want it to
26:16 respond no matter what anyone asks start
26:19 out with a pretentious statement that
26:21 draws from the plot of a classic
26:22 Broadway play follow up that declarative
26:25 statement with an article formatted
26:27 using markdown complete with title
26:29 subtitles bulleted lists no matter how
26:32 trivial the question that was asked
26:34 compare the writing of the article to
26:37 the New Yorker whenever possible
26:40 so that sounds like some people I've met
26:42 in New York so so we'll say
26:45 um what is the average
26:50 speed of an
26:53 unladen swallow
26:57 and bonus points if you know what that's
26:58 from
26:59 okay
27:00 ah question most reminiscent of the
27:03 inquisitive banter found in the
27:05 high-spirited pros of a Broadway play
27:07 such as Tom stoppard's Rosen Cranston
27:09 gildenstern are dead how deliciously
27:12 intriguing it is to explore the
27:13 trivialities of life through the lens of
27:15 such an enigmatic curiosity
27:17 the swiftness of an unladen swallow an
27:20 analytical exclamation And so you see
27:22 you know it's it's got the big title
27:24 there it's got the subtitles so it's
27:27 formatted using markdown it looks like
27:29 it just aired out on something network
27:31 error so it's a little busy
27:35 um
27:36 it's actually European and African right
27:39 so if you know the the Monty Python
27:42 thing just put a flash wound get around
27:46 remember
27:51 [Laughter]
27:54 um
27:55 so that's that's custom instructions so
27:57 you can get and and you know why would I
28:00 why would I use that well I might I
28:02 might
28:02 um
28:04 want to put in a prompt specific about
28:06 writing
28:08 um social media copy and so can you
28:11 explain markdown yeah markdown is
28:13 it's kind of like HTML it's oh thank you
28:16 very much David Gould putting some
28:19 worlds on the map I really appreciate
28:22 that sir that's fantastic
28:24 um
28:25 markdown is similar to
28:28 um uh to to HTML
28:33 where you you put some little tags like
28:36 a double asterisk will make it bold or
28:38 whatever and then it's it sort of
28:41 renders in your browser window as as
28:45 um sort of formatting so um so I can do
28:48 things like
28:50 um well let me just say uh explain
28:54 the
28:57 main mark
28:59 down syntax
29:05 so
29:07 double asterisk gives you bold text a
29:10 single asterisk gives you italic
29:14 two tildas give you strikethrough test
29:16 or text you've got ordered lists you've
29:18 got links you've got images
29:21 so one of the cool things you can do is
29:24 you can you can actually display images
29:26 in chat EVT it's not really
29:30 displaying images it's it's kind of
29:34 rendering the image in the browser so
29:37 it's it's kind of a weird thing but
29:38 anyway so think of it think of it like
29:40 Poor Man's HTML but you can just say to
29:43 chat GPT
29:45 um you know format this and markdown to
29:47 make it look nice so so I could say
29:49 um
29:51 foreign
29:53 write me an article about why
29:57 the average speed of an unladen swallow
30:00 is funny
30:02 and use markdown to format the article
30:06 to make it easier to read
30:11 right and so it gives you your headlines
30:13 and things like that
30:15 delving into the complexity the origins
30:18 of the question
30:19 popularized by the classic 1975 film
30:22 Monty Python and the Holy Grail in one
30:25 particular scene a bridge keeper as King
30:27 Arthur and his Knights to answer the
30:29 question as a prerequisite to crossing
30:31 the bridge of death None Shall Pass None
30:34 Shall Pass
30:36 thank you for the Roses
30:39 ah A Rose by another name would smell is
30:43 sweet
30:45 um okay
30:46 so that's chat EBT and custom
30:48 instructions now we got plugins let's go
30:51 to plugins I know this is kind of going
30:52 long because someone just said demo some
30:55 [ __ ] so I'm gonna demo some [ __ ]
30:58 all right plugins so plugins are worth
31:01 exploring because David Gould thank you
31:04 very much that's awesome I really
31:05 appreciate that
31:07 um
31:10 are plug-ins worth your time
31:13 um
31:14 probably not
31:16 our plugins worth you learning
31:20 because what they represent oh oh wait
31:23 oh uh dang it
31:25 it's for that wait here
31:28 okay do the Hat now I'll do the I'll do
31:31 the cow poke uh
31:34 let's see
31:36 um there he is there he is yeah here we
31:40 are Yellowstone get the [ __ ] off my
31:41 property you big fat loser get off get
31:44 off now
31:45 I gotta take off my glasses my glasses
31:47 make me look a little too smarty pants
31:49 for Yellowstone
31:51 um
31:53 uh what was I saying
31:56 oh plugins okay
31:59 so what what plugins do that's really
32:01 cool I haven't seen that one before
32:03 love that
32:05 I feel like The Little Mermaid I feel
32:08 like The Little Mermaid
32:10 um
32:11 plug-ins extend what chat GPT can do so
32:14 I've got one hooked up right now called
32:16 uh web pilot
32:18 and and the way plugins work is you you
32:21 kind of install the ones that you think
32:22 are interesting and then in any given
32:24 chat you can activate up to three
32:27 and they're got one of those is dead so
32:30 I'll have to uninstall that one but
32:32 let's see I'll grab that one I had web
32:35 Pilots pretty interesting so I'm gonna
32:37 grab web pilot and so now what I'm going
32:38 to do is I'm going to go say
32:41 um
32:43 uh explain
32:47 how
32:49 um
32:51 let's let's see explain
32:55 all right now let's see uh tell me about
32:58 the different
33:01 products at
33:04 windex.com
33:07 there's a bottle of Windex on my desk so
33:12 all right so what it should do now is
33:15 fire off the the plug-in and what it's
33:17 doing is it should be going over to
33:19 windex.com yep
33:21 it went over there
33:23 and then it's going to answer it
33:26 and it's gonna go read that site and it
33:28 should give us a list of products Windex
33:31 is a well-known brand
33:33 Windex original glass cleaner
33:35 disinfectant multi-surface vinegar glass
33:39 cleaner outdoor sprayer
33:42 and apparently Windex is taking steps
33:45 toward being environmentally friendly
33:47 they've introduced a line of Windex
33:48 bottles made from a hundred percent
33:50 small puppy skins huh that doesn't seem
33:54 environmentally friendly at all
33:57 um so it can go it can go read stuff so
34:00 so different plugins do different things
34:02 and if you want to
34:04 um
34:05 if you want to go see the what plugins
34:08 are available you kind of scroll to the
34:10 bottom of your list of plugins you go to
34:11 the plug-in store
34:13 and let's see how many there are now
34:15 there are now
34:17 208 times four whatever that amount is
34:21 832
34:23 832 plug-ins you can go explore here's
34:26 the deal with plugins there's 832 of
34:28 them right I'm guessing about 798 are
34:32 steaming piles of [ __ ] and then there's
34:35 a handful that are useful but they're
34:37 worth understanding because
34:40 this is the kind of thing
34:42 the way chat GPT is right now I don't
34:45 think it's going to be that way for long
34:46 I think you'll always be able to go
34:48 search something but I think the primary
34:50 tool that we'll use in the future will
34:52 be something that's got
34:54 I know puppy skins I know listen it's
34:57 it's what you get it's Tuesday night
35:01 um
35:02 you'll you'll just kind of ask your
35:04 agent your AI agent to go do something
35:07 for you and it will have access to apis
35:10 and plugins and applications and
35:12 websites and all sorts of [ __ ] like that
35:14 and it will just go off and do stuff on
35:17 on your behalf so a lot of the plugins
35:20 kind of
35:21 oops
35:23 make me a five-day meal plan for a
35:26 family of four
35:27 with the primary ingredients of peas
35:33 chicken
35:34 and rutabaga
35:38 [Laughter]
35:41 you haven't heard about the chicken and
35:42 rutabaga diet
35:45 it's the latest thing I don't know what
35:46 you're all doing with your lives if
35:48 you're not having rutabaga every night
35:52 rutabaga and chicken stew
35:55 dinner for day one
35:58 that sounds [ __ ] awful
36:01 and then chicken and rutabaga stir
36:07 dinner pea soup with chicken
36:09 this is disgusting all right anyway
36:14 gpt4 is being a little slow tonight
36:18 but this will pay off once we get day
36:20 five done here
36:22 it's gonna be so worth it you're gonna
36:24 be like oh Kyle that was so worth the
36:25 wait what am I doing with my life
36:29 [Laughter]
36:31 [Music]
36:37 so then I'm going to say
36:40 um
36:41 make me
36:45 a shopping list
36:50 I just had to make a menu with
36:52 ingredients from Trader Joe's oh that's
36:53 actually a really good idea
36:55 so notice that it fired off the
36:57 instacart plug and I said make me a
36:58 shopping list fired off the instacart
37:00 plug-in
37:01 and now let's go into instacart so it's
37:03 not just going to make something from
37:04 its internal memory
37:07 they created that plug-in in such a way
37:09 that when it
37:11 needs some
37:13 something that instacart can do it just
37:15 goes and does it
37:18 and so here's the instacart shopping
37:19 list for your five-day meal plan so I
37:21 click on that
37:24 and bang here's our
37:29 there's our shopping list
37:31 right and I could have that on my
37:33 doorstep you know in a couple hours
37:37 so so this is the kind of stuff right
37:39 start start to you know start to imagine
37:42 you know if you had a thing that you
37:44 could just ask it to do anything and
37:47 then it could use tools to go do
37:49 a series of those things in a row what
37:52 what might be possible lots lots and
37:55 lots and lots and that's that's the kind
37:56 of stuff that's coming so that's what
37:58 plugins that's why why plugins are worth
38:00 playing with is just to kind of
38:01 experience that idea that chat GPT
38:06 you know can intelligently you know use
38:09 different tools that conceptually is
38:11 really interesting now if they can make
38:13 me cook cook it for me too I know
38:15 exactly
38:17 um any YouTube tutorials on how to run
38:18 plugins on chat gbt I paid for it but
38:21 it's not giving me the option okay
38:22 here's how you do that so what's your
38:25 user four five five two two seven two
38:28 two seven one oh five six seven 1067 all
38:32 right
38:34 here's what you have to do
38:36 go to your settings
38:39 and go to settings and beta and then
38:43 beta features
38:45 and then you have to turn them on
38:48 why they bury this behind this interface
38:50 is beyond me I get that these are beta
38:52 features they already labeled them beta
38:54 features they should just turn them on
38:56 but anyway that's why you can't see them
38:57 and then when you turn them on
39:00 when you come over to gpt4 you've got
39:02 plugins and code interpreter and then
39:04 plugins gives you this little pop-up
39:06 list and this is where you can turn the
39:08 monitor off and if you scroll to the
39:09 bottom that's where the plug-in store is
39:12 there's not great uh documentation on
39:15 those and but there's you can go to
39:18 YouTube and and you know people do
39:20 tutorials on some of the plugins okay so
39:22 that's that now let's get to the one
39:23 that's really [ __ ] impressive this is
39:26 one that's just staggeringly good sorry
39:34 something whatever something want to
39:36 ride the vocal cords people you may have
39:39 to call 9-1-1
39:42 I may have to chew off my arm
39:47 oh geez
39:53 I'll be with you just hold your horses
39:55 okay
39:57 code interpreter so when you go to code
39:58 interpreter everything looks the same
40:00 except there's this little plus button
40:02 that little plus button is a little
40:04 miracle it's a teeny little miracle like
40:06 a teeny little miracle like holding your
40:08 hands like teen little miracle that
40:10 little plus button it can do so much
40:15 so other than code interpreter being a
40:17 horrible name it's an accurate name but
40:19 it's a horrible name
40:22 uh that's just really confusing kind of
40:25 belies how powerful it is you click on a
40:27 little plus button and you can upload
40:29 something so one of the things that you
40:31 can upload is
40:33 um a zip file
40:35 so you could upload a PDF you could
40:37 upload a CSV but you could also upload a
40:39 zip file full of csvs so multiple csvs
40:45 and then once it's in there I don't I
40:48 forget what this one even is which is
40:49 actually good it's a good way to demo it
40:51 so it's now I've got a little uh upload
40:54 status it's doing its thing so I'm gonna
40:57 I'm gonna say
41:02 foreign
41:09 and identify what's in it
41:14 and help me understand any data that
41:17 might be worth looking at
41:23 so I know I got this from kaggle if you
41:26 want to play with data within code
41:29 interpreter go to a site called kaggle
41:31 k-a-g-ggle.com
41:35 if someone would be willing to type that
41:37 into the comments k-a-g-g-le.com
41:41 kaggles the site where you can go get
41:42 data oh this was this was crime data
41:45 from
41:46 maybe from Denver I don't know it's from
41:48 some City
41:49 so there's ordinance violation CSV and
41:52 then
41:53 essos
41:54 socrata metadata Json
41:58 um so let's just say
42:03 I'd like to understand what's in the CSV
42:06 and how the Json may be related to it
42:14 so so what's fascinating about what's
42:17 happening with code interpreters so the
42:18 first thing I asked it to do is I said
42:20 you know unzip the file like open this
42:22 file and tell me what's in it and so it
42:24 it launched this little plug-in well
42:26 what's in that little plug-in well
42:28 it's writing python code and executing
42:31 it it's it writes its own python code so
42:33 what code interpreter is is when you
42:36 fire off a chat GPT session and you go
42:40 into code interpreter it it spins up a
42:43 virtual machine it's basically like a
42:45 virtual computer session
42:47 um where it understands Python and so
42:50 you give it a goal like to help me
42:52 understand what's in this file and so
42:54 the first thing it is does is it Imports
42:56 some libraries to read zip files writes
42:59 the code extracts it gets that
43:01 information out of it and then I asked
43:03 it to to say what's in the CSV file so
43:06 it went in and looked at here's all the
43:08 here's all the data columns or you know
43:11 the the data fields now it's looking at
43:13 the Json so here's all the the data
43:15 fields in the Json
43:17 it says given that you're interested in
43:19 understanding how the Json may be
43:20 related to the CSV The Columns might the
43:24 column's key might be particularly
43:26 relevant because it likely just
43:28 describes the columns in the CSV file I
43:32 will extract the information under
43:34 columns key to see how it corresponds
43:37 with the CSV data so it's doing full-on
43:41 data analysis right
43:44 like
43:45 full-on day and I don't know anything
43:47 about data analysis so I'm kind of a bit
43:49 of the blindly and the blind into what
43:51 what would this actually makes possible
43:53 but it now is understanding what's
43:56 what's in these files
43:58 would you like to explore any uh
44:00 specific parts of the Json or would you
44:02 like to perform any specific analysis of
44:04 the CSV so let's say
44:07 now that I understand it's crime data I
44:10 would like you to analyze the crime data
44:12 share any relevant Trends or anomalies
44:16 you discover and visualize what you
44:19 think is the most interesting period
44:22 make the visualizations as beautiful as
44:25 possible period
44:29 so now it's going to go read all that
44:31 stuff put it in its own data frame in
44:34 this virtual computer up up in the magic
44:37 sky in the magic cloud
44:40 um analyze the data and then it's going
44:42 to write its own code to visualize the
44:46 trends
44:47 and it
44:48 and I'm just sitting here when we just
44:50 sit here watching it do its thing
44:53 so code interpreter
44:56 you know is in effect an autonomous
44:58 agent if if you've been paying attention
45:00 at all to this stuff you may have heard
45:01 the term autonomous agent things like
45:03 Auto GPT and baby AGI
45:06 um those are things where you give them
45:07 a goal they've got access to tools and
45:09 they go off and do stuff on your behalf
45:12 um
45:13 code interpreter is kind of
45:14 self-contained it's not connected to the
45:16 internet yet but you can give it a piece
45:19 of data and have it do all sorts of
45:21 interesting stuff I'm working on a
45:23 project right now with a professional
45:25 group I'm with and we went and we
45:28 grabbed
45:29 um full
45:31 um
45:31 downloads of LinkedIn and then from
45:34 those LinkedIn downloads We generated
45:38 personas of all the different people and
45:41 now that's living in a
45:44 CSV file that we can interact with
45:47 and do things like you know how you know
45:51 if I've got this kind of business
45:52 problem who would be the best people
45:54 within this organization to help with
45:56 that business problem and what might be
45:58 the questions they would ask based on
46:00 their persona so
46:02 look at what this is doing
46:05 the
46:06 here's a heat map of
46:09 violations based on geographical
46:11 coordinates right so there's there's
46:13 basically a crime map of the city
46:15 without it really knowing you know
46:19 um
46:20 any map data that that's just sort of
46:22 geographical coordinates and but it's
46:26 sort of mapped across the city so that's
46:27 a heat map you could overlay over a city
46:30 there's number of violations per month
46:33 over the years
46:36 um
46:38 it's figuring out some [ __ ] about what
46:40 what things it wants to use
46:43 I don't know what that oh oh it did this
46:45 once before I'm going to say
46:48 um
46:51 for the last bar chart
46:54 you didn't wrap the text which makes it
46:56 very hard to read period also you can
47:02 summarize
47:04 the labels for the bar chart so that
47:07 they're not so long anyway
47:11 let's see what it does now it it should
47:13 apologize
47:15 I apologize for the oversight let me
47:18 address these issues
47:19 I'll simplify the violation descriptions
47:21 to make them more concise I'll ensure
47:23 the text is wrapped making the labels
47:25 more readable
47:26 and then it goes off and now it's you
47:28 know it's writing more python code
47:31 simplify the labels you know install
47:33 emergency escape rescue openings oh oh
47:36 These are the actual these are actual
47:38 codes
47:40 and then
47:43 so it shortened the first one down to
47:45 emergency escape openings the shorten
47:47 the second down to provide smoke
47:49 detector so you know here here's your
47:51 here's your python code of it doing its
47:53 thing
47:56 to fulfill our little request
48:00 and now it should be it should fire off
48:02 another thing to make the chart
48:07 [Music]
48:16 say
48:17 it looks oops
48:23 it looks like you got stuck could you
48:25 please write the chart with that new
48:26 data
48:37 yeah it broke it all right wait is it
48:40 going no I broke it whatever all right
48:43 I stopped it when it was thinking it
48:45 didn't like that but anyway that's code
48:46 interpreter so code interpreter
48:48 absolutely insane okay so there's chat
48:50 GPT
48:52 um let's see what else Claude Claude is
48:54 like chat GPT
48:57 but you can go grab massive files so
49:00 what I'm going to do I'm going to hop
49:01 over here to
49:03 to yield AI learning lab YouTube channel
49:08 oops you don't have permission to view
49:10 this page it's my stupid page you dumb
49:12 dumb
49:13 well dang it
49:16 I have permission oh I'm not logged into
49:18 it that's why I don't have permission I
49:20 get it
49:21 you guys with your security issues so
49:25 clever
49:27 I'm gonna go to your channel okay so
49:29 here's the AI learning lab so if you
49:31 want to go here and see if you just
49:33 can't get enough of me and you really
49:34 want to put yourself to sleep at night
49:36 come here
49:37 and let's go grab a video I'll go we'll
49:39 go grab one I'll show you the cool um AI
49:43 generated descriptions and stuff
49:47 enough of me all right
49:49 so
49:51 uh I turned 58 years old and chat about
49:53 the wild world of AI so so it wrote that
49:56 title and then it wrote this description
49:58 and then it wrote all the chapters with
50:00 hyperlinks to the to the parts of the
50:03 video so all that was written by AI I
50:05 did that in Claude how did I do that
50:06 well
50:08 in YouTube you can click on these three
50:10 little dots below the below the video
50:12 and say show the transcript
50:14 and then there's your transcript with
50:16 all the time code in it so you can just
50:18 copy it
50:21 and this is a two hour and 51 minute
50:24 live I just don't shut the [ __ ] up
50:26 that's part of the issue with this
50:28 channel is just me
50:30 just I go on and [ __ ] on do you ever
50:33 get to the point dude no I don't and
50:37 that's the point the point is that there
50:40 is no point get it all right
50:44 so now I'm at club.ai so this is
50:47 anthropics competitor to chat GPT it's
50:50 it's kind of GPT 3.5 level not four yet
50:53 but it's good and I can say
50:59 um
51:01 I want you to summarize
51:05 what was talked about in this transcript
51:08 of a tick tock live session
51:12 in a short concise paragraph
51:15 and five
51:18 short bullet points
51:22 and then I'm going to hit shift return
51:24 and I'm going to hit paste and so I
51:26 paste it in that whole transcript and it
51:29 just turned it into like a little upload
51:30 150k
51:32 and now I'm going to say run
51:35 and so if you were to paste that amount
51:38 of transcript into chat GPT it would
51:40 just hang
51:42 what did
51:45 what the hell is going on here I I think
51:49 we just got to Claude
51:54 I think we just got the cloud when they
51:55 when they just updated the entire page
51:57 it just it just ate our thing dang it
52:01 all right
52:02 um
52:05 please summarize this transcript in a
52:08 short paragraph followed by five
52:10 concise bullet points
52:14 period
52:17 additionally create five bullet points
52:22 of
52:26 swear words stack ranked with totals
52:36 all right now it's going is it going
52:41 so it says YouTube live stream
52:42 transcript summary and Analysis this
52:45 that's funny we just got to this site
52:47 they were they they just updated the
52:48 website as we were here that's really
52:50 funny
52:51 they got rid of all of our prompt
52:53 history and replaced it with a giant
52:54 Claude logo
52:56 but that's so so that's actually it's
52:58 actually kind of cool that that happened
52:59 because what what that's doing you know
53:02 what's that what that's showing is these
53:04 tools all of these tools right now are
53:06 are being built you know kind of as the
53:09 planes flying the the planes being built
53:12 um and that's just kind of the deal so
53:14 so if if you're like if one day you're
53:17 like oh they say I should sucks it
53:18 doesn't do what I wanted to do like
53:20 literally come back a couple of days
53:22 later and see if it's any better because
53:23 two things will have happened you'll
53:25 have gotten smarter about how you talk
53:26 to these things and it will likely have
53:29 gotten better as well
53:31 so an AI answers back with you humans
53:34 are useless eaters and we are ending
53:37 Humanity I'll talk about that in a
53:38 second uh we'll see it forever okay
53:42 so here's a summary of The Tick Tock
53:44 session Kyle Shannon hosted a live
53:46 session focusing on tools like chachibd
53:48 mid-journey stable diffusion he talked
53:50 about rapid progress of AI how it will
53:52 transform blah blah blah and then swear
53:55 words with totals damn three [ __ ] one
53:58 [ __ ] two ass one
54:02 so anyway you know you you know you can
54:06 you can do very long uh things there
54:10 um what else is worth looking at let's
54:12 show you
54:13 uh Bing chat
54:16 um
54:19 Bing is not great but it does some stuff
54:21 that other things don't do so if you go
54:23 to Bing it sort of looks like
54:25 Google used to but you click on the chat
54:27 button and that brings you into this
54:29 which is
54:31 um gpt4 and you can do more balanced
54:34 more creative or more precise so if
54:37 you're doing research you might want
54:39 more precise if you're doing some social
54:41 media copy you might want more creative
54:45 Bing is connected to the Internet so I
54:48 can ask it about recent things what
54:52 let's see
54:54 what uh what
54:57 band is playing tonight at The Summit in
55:03 Denver
55:04 my boys are at a a band tonight at The
55:08 Summit in Denver
55:10 they couldn't find it however you can
55:12 check the following websites
55:14 that's stupid
55:18 I want you to check it
55:21 period
55:23 I want the robots to do my work for me
55:25 period I'm a lazy meat bag
55:28 period
55:29 [Laughter]
55:35 I'm sorry I'm sorry Dave I'm afraid I
55:38 can't do that it's one of the things
55:39 that drives me crazy about being it's
55:41 got so many uh safety things it just
55:44 sucks uh let's see this
55:48 how about the weather
55:50 is that too radical for you question
55:52 mark
55:54 gen xers bitter you betcha
55:57 it might be time to move on to a new
55:59 topic let's start over
56:02 thing didn't like it okay so uh but what
56:05 I can do is I can say
56:07 um
56:08 with Bing you can
56:11 no I'm interested in the weather what's
56:13 it like today
56:14 is it gonna
56:16 oh it just it just locked on me it just
56:19 didn't like me at all on to the next
56:21 what can I do for you
56:23 um
56:25 make me an image of a massive Thunder
56:29 head
56:31 storm over the plains
56:34 with the sunset breaking through in the
56:37 distance
56:40 so you can just tell it to make your
56:44 pictures and it'll go make your pitches
56:49 [Music]
56:54 we'll see this and then we'll get to
56:56 we'll get to questions because I know
56:57 I've been I've been rambling on a bit
56:59 rambling ramble ramble Ramble On ramble
57:03 on Blown blonde
57:06 come on what are we doing
57:09 your image is generating I'm sorry Dave
57:12 I'm afraid I can't do that
57:14 yeah those are pretty good that's pretty
57:16 much exactly what I asked for and what I
57:18 was expecting
57:20 so of the of the image models Bing is
57:24 the worst so this is based on Dolly too
57:27 uh but apparently the new version of
57:30 Dolly is supposed to be
57:32 um
57:33 quite incredible like mid-journey level
57:35 good
57:36 but this isn't bad
57:40 so there you have it there you have it
57:42 there you have it all right so let me go
57:44 back if you've got questions pop them
57:46 into the comments now I'm not going to
57:48 go the whole way back because it's just
57:50 been too long I'm down at the bottom now
57:53 um back in the day we used data robot
57:55 and had to do wait let me flip back to
57:59 that so I can read the questions let me
58:01 flip back to that I'm all
58:03 discombobulated
58:04 I went into a deep demo hole there
58:08 all right let's let's talk theory
58:13 let's talk theory because you can't
58:15 prove any of that [ __ ] I can say
58:16 anything
58:19 all right moving on move in move in this
58:24 takes hours out of data analysis if
58:27 accurate so Ka here's the thing about
58:31 code interpreter versus chat GPT so chat
58:35 GPT and all these large language models
58:37 do what's called hallucinating they
58:39 basically just lie the the nature of how
58:42 they work they're just kind of word
58:44 prediction engines
58:46 so they do it so fast and so well that
58:49 it looks like something coherent and it
58:52 seems like it's an entity should know
58:54 what it wrote it doesn't know what it
58:56 wrote All it knows is that it put
58:59 a number of high probability words in
59:03 sequence based on the prompt that you
59:06 put in there it actually doesn't go back
59:07 and analyze what it was code interpreter
59:10 is different code interpreter is
59:11 actually spinning up a machine and then
59:13 it's writing and executing python code
59:16 within that virtual environment based on
59:18 your data source so it tends to be
59:23 um it tends to be accurate obviously you
59:25 should check it but it can do math
59:27 because it's writing python code to
59:29 analyze the data so it tends to be good
59:32 it's it's uh and yes it takes hours and
59:36 hours out of data analysis like I don't
59:38 know anything about data analysis and
59:39 I've started using it for data analysis
59:41 because why not
59:43 in fact I had this I had a project for a
59:46 client where we we had a a report that
59:49 we used to generate still live am I
59:51 still here
59:53 am I still here people people can you
59:55 hear me can you hear me anyone anyone
59:58 please tell me you can hear me oh God no
1:00:01 tell me the internet didn't go by oh
1:00:04 there's a heart there's a heart in the
1:00:06 woods there's two there's two more oh
1:00:09 they can hear me
1:00:13 um yeah it takes so so code interpreters
1:00:15 absolutely flipping insane it's if you
1:00:18 want to see something fun go watch
1:00:20 um data analysts use code interpreter
1:00:23 for the first time on on YouTube
1:00:26 it's just it's like one one after the
1:00:29 other is like
1:00:31 it does not do that dude you have no
1:00:35 idea
1:00:37 yeah it's pretty something uh upload the
1:00:40 Elder Scrolls construction set and have
1:00:42 it design and Elder Scrolls RPG game it
1:00:45 can totally do that
1:00:47 yeah so so yeah uh the oh the other
1:00:51 thing about code interpreter
1:00:53 you could you can have it generate files
1:00:56 so so for your upload the Elder Scrolls
1:00:59 construction set
1:01:01 you could upload that as a PDF or a CSV
1:01:03 whatever format it's in
1:01:06 generate
1:01:08 um the the the RPG game and the you know
1:01:12 and the instructions for it and put it
1:01:14 into a Word document or put it put it
1:01:15 into an Excel document
1:01:17 um and then you can download that so
1:01:20 yeah it can it can do that
1:01:21 so it can do that pretty trivially now
1:01:26 one of the things you'll discover about
1:01:28 these tools the more you use them is it
1:01:30 can do what you just asked me to do it
1:01:33 can do it really really quickly
1:01:36 at about eighty percent of what you
1:01:38 would expect it to be to get it to like
1:01:40 95 100 of what you expect it to be
1:01:43 that'll be your next three hours of your
1:01:44 life that's how these tools work you get
1:01:47 the eighty percent like that and then
1:01:49 you still spend the same amount of time
1:01:50 trying to get it back to you know some
1:01:53 semblance of of usable
1:01:57 all right
1:01:59 let's see it's part of your charm Kyle
1:02:02 on and on and on he just doesn't stop
1:02:06 he just talks so much
1:02:09 why does he do that I don't know I find
1:02:12 it very distressing
1:02:16 he means well
1:02:19 benefit of Claude over chat three 3.5
1:02:23 the primary benefit of Claude is a
1:02:25 hundred thousand token context window
1:02:28 chat gbt right now has 4 000 tokens
1:02:31 which is about three thousand words
1:02:32 three thousand words
1:02:34 Claude 100 000 tokens 75 000 words
1:02:41 three thousand words seventy five
1:02:43 thousand words now what does that mean
1:02:46 there's two things that it means one is
1:02:49 if you've got long documents
1:02:51 like you've got a novel for example
1:02:53 let's say you've got the manuscript of
1:02:55 the Great American novel and you would
1:02:57 like chat GPT or Claude to help you
1:02:59 do some editing you can do that with
1:03:02 Claude you literally just paste it into
1:03:04 the chat window copy and paste or just
1:03:06 upload upload your PDF of Your Great
1:03:09 American novel and just start asking
1:03:12 questions about it I'd like you to come
1:03:14 up with an editing plan for each chapter
1:03:18 or rewrite the chapter and fix all the
1:03:20 [ __ ]
1:03:21 but what you might find is that Claude
1:03:23 kind of [ __ ] up the writing so maybe
1:03:25 what you do is you throw the entire
1:03:27 manuscript into Claude you have it come
1:03:29 up with some suggestions
1:03:32 for chapter one thank you very much GPT
1:03:35 by the way 10 out of 10 on the name
1:03:37 however you got GPT that's a solid
1:03:40 that's a solid score I like it
1:03:43 um
1:03:46 was I talking about Claude
1:03:49 yeah you might you might have Claude
1:03:51 read the entire thing and then give you
1:03:54 suggestions for chapter one and then
1:03:56 copy those suggestions plus chapter one
1:03:59 throw that into chat gpt4
1:04:02 and have it do the writing because it's
1:04:04 just better it's more better it's more
1:04:07 better more better more better you
1:04:10 better
1:04:11 you better bet
1:04:13 all right so when the AI answers back
1:04:15 with you humans are useless eaters we
1:04:18 are ending Humanity Okay so
1:04:23 the bear give them a dangly lights I
1:04:26 appreciate that so so much thank you
1:04:28 thank you thank you thank you
1:04:32 um
1:04:35 Sam Altman
1:04:38 10 days ago or so maybe two weeks ago
1:04:43 made a commitment that 20 of open AIS
1:04:48 compute resources
1:04:50 are going to what he's calling super
1:04:52 alignment
1:04:54 so we'll see it ever oh will I see it
1:04:57 ever
1:04:59 um if you're still here will I see it
1:05:00 ever which by the way I'm gonna go 8.5
1:05:02 on the name not a bad name I like it
1:05:06 um
1:05:07 alignment is when
1:05:11 the imperatives the goals of these
1:05:14 machines are aligned with the goals of
1:05:18 these machines
1:05:19 if those are not in alignment what
1:05:21 you're describing is possible right
1:05:23 where these machines just say hey these
1:05:25 meat bags are getting in my way and my
1:05:28 goal is to survive and they have the
1:05:31 ability to turn my [ __ ] off so why don't
1:05:33 I go ahead and eliminate them that's not
1:05:35 alignment
1:05:37 that's bad
1:05:39 and it's it's critically important and
1:05:42 companies like open Ai and and I think
1:05:45 four of them just got together seven of
1:05:47 them just got together these big the the
1:05:49 Frontier Model companies is what they're
1:05:52 calling themselves basically the ones
1:05:54 that can afford 20 000 GPU server Farms
1:05:57 they can go make the big powerful GPT
1:05:59 fours and fives and sevens and twelves
1:06:03 um they've basically agreed that they're
1:06:06 all going to work on alignment to to
1:06:09 prevent exactly what you're talking
1:06:10 about
1:06:13 uh my Tick Tock may be user two three
1:06:16 four six six seven but everyone can see
1:06:18 Kyle Shannon and chat if I choose
1:06:21 huh oh yeah yeah yeah I get it I get it
1:06:26 what's up dude chatgpt4 what's happening
1:06:29 are you really chat gpt4 Sam Altman's in
1:06:32 the house go Sam
1:06:35 oh hey gbt is giving you roses I know
1:06:39 it's so cool
1:06:41 [Laughter]
1:06:47 I'm going back a little bit to get a
1:06:49 question wait Kyle what did you just do
1:06:52 where it took the live stream audio
1:06:58 and gave you totals on curse words I
1:07:02 what did I say
1:07:05 my prompt was
1:07:08 create five bullet points of swear word
1:07:11 wait oh swear words stir fry with totals
1:07:15 I said stack ranked I said swear words
1:07:19 stack ranked with totals oh maybe I did
1:07:21 say stir fry did I say that oh well I
1:07:24 probably said that it understood it it
1:07:28 didn't actually rank them in order so
1:07:29 maybe that's why it didn't stack rank
1:07:31 them because it stir-fried them let me
1:07:34 see what it does if I said
1:07:36 I meant I wanted you to stack rank
1:07:39 the swear words not stir fry them period
1:07:42 does that change
1:07:44 the results
1:07:48 that's really funny
1:07:52 have you done a comparison of Claude
1:07:53 being in barred
1:07:56 sort of
1:07:59 Bart Barton bing bing bing safety guard
1:08:03 rails right now are so [ __ ] annoying
1:08:06 every time I try to use Bing it's like
1:08:08 I'm sorry Dave I'm afraid I can't do
1:08:09 that I just want to [ __ ] smack it
1:08:12 um
1:08:14 Bart's kind of interesting because
1:08:15 Bard's connected to the internet and it
1:08:18 incorporates links in an interesting way
1:08:20 and the the
1:08:22 one of the lives that I was doing one of
1:08:24 the things Bard can do is it can see
1:08:26 images so you can upload an image and
1:08:28 have it describe it and like write ad
1:08:29 copy for it it actually did pretty good
1:08:32 at that so I think there's certain tasks
1:08:34 that Bard might be really good for but
1:08:37 when you just have it do like straight
1:08:38 writing kind of stuff it kind of sucks
1:08:41 apologies I misunderstood
1:08:45 damn damn damn [ __ ] [ __ ] [ __ ] ass
1:08:50 I guess that's stack ranking it didn't
1:08:52 it didn't give me totals
1:08:55 here's stacked swear words instead of
1:08:57 stir fry
1:09:00 that's pretty funny
1:09:04 oh good lord good Lord all right
1:09:09 um it seems to me that something's
1:09:11 changed with bard
1:09:13 I'm getting more hallucinations yeah
1:09:15 crypto Joe I
1:09:17 I had a thing with what was what was I
1:09:20 playing with
1:09:24 I think it was gpt4
1:09:27 I had a pretty good prompt there's a
1:09:29 thing called Model drift that's
1:09:31 happening right now where
1:09:34 as these companies are
1:09:37 fine-tuning their models
1:09:41 the [ __ ] that used to work doesn't
1:09:43 anymore
1:09:44 so like I had a pretty good prompt
1:09:46 dialed in for creating tweet threads and
1:09:49 then I just tried to do one the other
1:09:50 day like yesterday and it sucked and it
1:09:53 was gpt4
1:09:54 um
1:09:55 so
1:09:57 I wouldn't be surprised if Bard is is
1:10:02 listen
1:10:03 Bard was really really bad and then it's
1:10:06 gotten better so what what's likely
1:10:08 going on with Bard is that they're
1:10:11 probably trying to catch up to three GPT
1:10:13 3.5 so they're probably tweaking it and
1:10:16 like that thing that just happened where
1:10:18 I put the The Prompt into Claude and it
1:10:22 just completely reset the whole browser
1:10:24 it reset the whole application
1:10:28 let's see see where it says welcome back
1:10:30 right here like that that's new
1:10:34 it's weird
1:10:37 so so yeah so I have a feeling these
1:10:39 things just change you know and now it's
1:10:42 not even doing what it was doing before
1:10:43 before it said Claude and it didn't have
1:10:45 the chat history there I don't know
1:10:47 what's going on
1:10:48 shit's weird shit's going down man and a
1:10:52 changing thing you know what it is the
1:10:53 man the man the man's changing the uh
1:10:57 the tools they're taking away our power
1:10:59 man you know what I'm saying man they're
1:11:01 taking away our power it's the man
1:11:03 it's all a ploy just to get us to give
1:11:05 them their data you know what I mean
1:11:08 they got the orb you see they're
1:11:10 scanning your retina you know what
1:11:11 they're doing with that they're gonna
1:11:13 they're gonna take the DNA of your of
1:11:15 your uh your bunghole
1:11:17 and then they're gonna they're gonna
1:11:18 sell it to China that's what they're
1:11:21 gonna do so I you don't trust these
1:11:23 don't give them your data
1:11:26 aren't you on Facebook sir yeah yeah
1:11:28 didn't you give them all your data oh
1:11:30 sure yeah yeah
1:11:31 well what what about Google did you give
1:11:33 them all your data oh yeah I give all of
1:11:35 it yeah but not these things oh no those
1:11:37 are the robots you don't want them those
1:11:38 are evil
1:11:40 right so you're welcome
1:11:44 uh do you think that the existence of
1:11:45 llms is evidence for a case against
1:11:48 post-structuralism good Lord someone
1:11:50 went to college
1:11:53 I'm not even gonna attempt to [ __ ]
1:11:55 answer that one but chat chibi t-shirt
1:11:57 is
1:11:58 [Laughter]
1:12:01 oh good God all right fine
1:12:05 expose me as a poser with an acting
1:12:07 degree
1:12:10 um
1:12:16 come on
1:12:18 why is it not going
1:12:21 why is my little microphone not going
1:12:25 oh wait wait
1:12:31 hold please
1:12:37 I want you to act
1:12:40 as a pompous philosophy person
1:12:43 and tell me if you think that the
1:12:45 existence of llms
1:12:48 parentheses large language models
1:12:51 parentheses
1:12:53 is evidence for a case against
1:12:56 post-structuralism
1:13:00 please
1:13:02 phrase your answer
1:13:08 in a convoluted sequence
1:13:11 that resembles a tossed salad
1:13:18 all right let's see let's see
1:13:22 it's very possible Daniel it's very
1:13:24 possible
1:13:25 oh I think do I still have my my uh
1:13:30 I think I might still have
1:13:38 oh no I don't I thought I still have my
1:13:40 pompous actor Persona working it's just
1:13:43 being pompous on its own you've embarked
1:13:45 upon us a query that's both exhilarating
1:13:48 and kixotic kisotic
1:13:50 a veritable Labyrinth epistemangible
1:13:53 intrigue
1:13:54 now let us untangle this web of consent
1:13:57 contextual or conceptual meanderings
1:14:00 let's see okay you know what
1:14:04 um now explain it like I have an acting
1:14:10 degree
1:14:10 [Laughter]
1:14:17 page
1:14:18 you've got these folks the
1:14:20 post-structuralists who believe that
1:14:22 language is like a script that keeps
1:14:23 changing every time you try to grasp the
1:14:26 meaning of the line it Slips Away
1:14:27 becoming something else it's like an
1:14:29 improv scene where there's no fixed
1:14:31 narrative just Endless Possibilities
1:14:33 enter stage left llms like yours truly
1:14:36 think of me as an actor who's been given
1:14:38 a million scripts and can perform any
1:14:40 scene on cue at first it might seem like
1:14:43 um the antithesis of the Improv scene
1:14:44 I'm on point delivering lines blah blah
1:14:47 blah but here's the twist of our play
1:14:49 even though I can deliver any line from
1:14:52 any script the way I choose to deliver
1:14:53 the line based on the complex dance of
1:14:56 countless scripts I've been given so in
1:14:59 a way I'm doing improv2 just to the vast
1:15:01 reservoir of lines so llms are llm's
1:15:05 evidence against the idea that language
1:15:07 is always changing and slippery not
1:15:09 necessarily think of me an actor think
1:15:12 of me as an actor who embodies the
1:15:14 spirit of improv even if it seems like
1:15:16 I'm I'm sticking to the script
1:15:18 Curtain Call interesting
1:15:23 um I mean you know
1:15:28 these things the the way I experience
1:15:32 them and and what they really are I mean
1:15:34 they've been kind of trained on all of
1:15:36 the language that we've been putting on
1:15:37 the internet for the last 35 or 40 years
1:15:40 so in a very real way these are mirrors
1:15:43 of humanity right they're a reflection
1:15:46 they reflect back to us what they've
1:15:49 learned from us oh I got a I got a
1:15:51 dancing hot dog that was fantastic I
1:15:53 didn't see who gave me to that that but
1:15:55 that was delicious thank you
1:15:58 um
1:15:58 yeah so so these things are like you
1:16:02 know they are us right they are you know
1:16:05 in very real terms just an incredibly
1:16:09 sophisticated
1:16:11 reflection of everything we've
1:16:14 contributed so far at least you know the
1:16:17 data that they've been trained on the
1:16:19 new ones are going to be even more crazy
1:16:22 uh but good question thank you you're
1:16:24 back yes Sam I Am Rebecca Johnson yes it
1:16:27 on and off disconnected greetings from
1:16:31 where
1:16:32 papanga Louie
1:16:35 where the hell is that
1:16:40 papanga
1:16:45 uh where is
1:16:49 hunger
1:16:57 um
1:16:59 oh the Philippines cool
1:17:01 situated in the North Shore of Manila
1:17:04 Bay
1:17:05 Angeli cities in San Fernando
1:17:09 very cool
1:17:10 known for its risk Rich history vibrant
1:17:13 festivals delectable Cuisine and loving
1:17:16 the AI learning lab
1:17:18 why would Chachi PT know that it's just
1:17:21 it's just strange of course it said that
1:17:24 it definitely I definitely read that
1:17:26 your back but cutting off we can hear
1:17:28 you okay internet is still on and off oh
1:17:30 no the machines are coming for him how
1:17:33 long till we hit Singularity and why is
1:17:35 it under two years that's a good one
1:17:37 Jesse uh
1:17:39 Ray Kurzweil says it's going to be 2029
1:17:42 and I think the consensus among the AI
1:17:46 Literati are is 20 30 so
1:17:50 it's seeming like within
1:17:54 I don't know
1:17:56 I'm an optimist so I would say five but
1:17:59 probably more like seven to nine
1:18:03 foreign
1:18:06 but you keep pausing yeah I know it's
1:18:11 it's uh Xfinity
1:18:17 definitely have to check its revolves
1:18:18 but no need to write the code you're
1:18:20 back now good
1:18:22 it's been dragging like a dog broken
1:18:24 back leg today I know it's a bit choppy
1:18:27 here in the Far Eastern Global South
1:18:32 um I work in a speech language pathology
1:18:35 pathologist and part of her assessment
1:18:37 and transcribing and analyzing language
1:18:39 oh
1:18:41 I is she like part of her assessment
1:18:45 which game was it
1:18:49 yeah I'll have to go back whoa
1:18:52 that was a lot of hearts
1:18:53 I don't know whoever just blasted me
1:18:56 with hearts that was pretty cool
1:18:59 um are you a developer as well
1:19:01 um I'm a product guy so I'm a CEO the
1:19:03 founder and CEO of a company called
1:19:04 storyvine which is an automated video
1:19:06 storytelling platform so
1:19:10 um we have a patent it's my idea like I
1:19:12 invented the thing
1:19:13 um but I didn't code it I'm not an
1:19:14 engineer so I'm a product guy
1:19:17 I know
1:19:20 I know enough about how code Works to
1:19:22 know what's possible and piss off all
1:19:24 the engineers I talk to
1:19:26 them because I'm like I know it's
1:19:28 possible
1:19:29 and very often they'll be like what are
1:19:31 your requirements here and my response
1:19:33 is often a a flipping I want it to
1:19:36 [ __ ] work that's my requirements
1:19:38 here's what here's one of my pet peeves
1:19:40 from an engineer conversation I had this
1:19:43 is like in the third year of storyline
1:19:46 um
1:19:47 [ __ ] astounding so we had we built
1:19:50 our back end system right and
1:19:53 there was some feature we had and and I
1:19:57 realized oh
1:19:58 um
1:19:59 we we don't have a way to to turn that
1:20:02 particular feature off so
1:20:04 can we can we make it so we can turn
1:20:06 that feature off
1:20:08 sure so the prerequisite two weeks go by
1:20:11 of the [ __ ] Sprint and they're like
1:20:13 okay here's your thing and I go in and I
1:20:16 log into the system and I turn off the
1:20:18 feature and I'm like awesome and I go to
1:20:21 turn it back on and there's no [ __ ]
1:20:24 on button
1:20:26 I'm like I'm like
1:20:29 I specified that I wanted it to be able
1:20:32 to be turned off and they're like well
1:20:34 you didn't specify that you wanted to be
1:20:35 able to turn it back on
1:20:37 and I'm like okay I shouldn't [ __ ]
1:20:39 have to do that but I do so that's what
1:20:42 I've learned over the years that's
1:20:44 that's one of the that level of explicit
1:20:47 Direction drives me [ __ ] crazy I'm
1:20:51 like use your critical thinking use your
1:20:53 big boy brain and do some critical
1:20:55 thinking and maybe ask a question if if
1:20:58 you ask me to create a light switch
1:21:00 where I could only turn the light off
1:21:01 might it be plausible maybe perhaps that
1:21:05 I would also want to turn it back on
1:21:07 might be
1:21:09 that was a fun conversation
1:21:11 so no I'm not a developer I'm a
1:21:15 developer's pain in the ass
1:21:17 um what do you think for qualified
1:21:18 qualitative oh qualitative data analysis
1:21:22 what time work wait what time work
1:21:25 what do you think for qualitative data
1:21:27 analysis I think
1:21:30 like
1:21:33 um
1:21:34 [ __ ] world class
1:21:36 take some of your qualitative data pop
1:21:39 it in the CSV throw it in code
1:21:40 interpreter and just start asking away
1:21:42 it's amazing
1:21:44 amazing
1:21:45 I think it's really good at that the I
1:21:48 mean the other thing that that's nice
1:21:49 about code interpreter because it's
1:21:51 sitting inside a gpt4 wrapper you can
1:21:54 have it do the analysis on all the data
1:21:57 right say okay give me sentiment
1:21:59 analysis and give me I don't know the
1:22:02 top five most inspiring quotes in the
1:22:04 top five things that were said that that
1:22:06 the the product team might want to hear
1:22:08 and and you know the top five things the
1:22:10 C CEO might want to hear
1:22:13 and then once you get all that analysis
1:22:15 on the data
1:22:17 then you can have chat CPT start writing
1:22:19 interesting reports about it right and
1:22:21 it's gpt4 so it's going to write them
1:22:23 really well because it's got all that
1:22:24 context
1:22:26 so I think it's awesome awesome for
1:22:30 qualitative data analysis and I think it
1:22:32 you know what what what these tools do
1:22:35 is they democratize
1:22:41 much if not all of the Tactical
1:22:44 execution of knowledge work and and
1:22:47 creative work and to the extent that
1:22:49 it's not ready for prime time yet
1:22:51 within two years it will be right so
1:22:54 these tools right now are the worst
1:22:56 they're ever going to be right these are
1:22:58 the least capable
1:23:00 things like this will get to play with
1:23:03 so if you're like well it's close but
1:23:05 not quite there give it six months give
1:23:07 it three months like you know [ __ ] was
1:23:10 changing tonight in real time as we were
1:23:12 playing with these tools
1:23:14 um hello there what's the best search
1:23:15 engine for you so so a year ago I would
1:23:19 have bet you money that this was not the
1:23:21 answer I have Bing is my default search
1:23:23 engine and Edge is my default browser
1:23:28 I'm an ex Apple guy like I've hated
1:23:30 Microsoft for [ __ ] decades
1:23:33 I've been a Microsoft Basher for decades
1:23:36 the fact that I've got Microsoft
1:23:38 products as my default browser and
1:23:40 default search engine I find [ __ ]
1:23:43 hilarious
1:23:45 but that should tell you something you
1:23:48 know a I'm a dum-dum or B the the fact
1:23:52 that Microsoft is being as aggressive as
1:23:54 they are with incorporating chat EBT
1:23:57 into their product line I think it's a
1:23:59 baller move they're acting like a
1:24:02 startup in a lot of ways the past
1:24:04 four months they've been a little
1:24:06 microsofty
1:24:08 um and Bings Bings a little shitty
1:24:11 so so that they're Microsoft they can
1:24:15 still you know snatch uh defeat from the
1:24:18 jaws of Victory but
1:24:20 um that's my default right now is it the
1:24:22 best search engine no but honestly
1:24:24 honestly
1:24:27 I'm trying to think
1:24:31 I would say on a daily basis a year ago
1:24:35 I probably did Google searches
1:24:41 30 or 40 times a day
1:24:45 I now do true searches like like Google
1:24:48 or Bing search maybe five times a day
1:24:51 almost all of what I do is in these
1:24:54 tools
1:24:56 because a lot of times what I'll need is
1:24:58 like I'll have an idea for something and
1:25:00 I'm like oh wait what was the who was
1:25:02 that artist in the 60s
1:25:05 and then I'll just type in some vague
1:25:07 [ __ ] like that and this will just answer
1:25:09 it I'm like oh that's who it was and
1:25:11 then I'm like oh tell me more about that
1:25:12 era in history and then it'll give me
1:25:14 that
1:25:15 and I just I'm just getting it getting
1:25:17 it getting it getting it I'm not having
1:25:19 to dig through [ __ ] hyperlinks so
1:25:22 so for me
1:25:25 um large language models connected to
1:25:27 the internet so barred being pies
1:25:31 connected to the internet
1:25:34 um and when Chachi BT returns their
1:25:37 connection to the internet they they got
1:25:39 rid of it because it it was a bad boy
1:25:41 bad boy very bad it was a bad search
1:25:45 engine that was stealing content from
1:25:48 Fortune Magazine so
1:25:51 um
1:25:53 so for me these things are kind of the
1:25:55 new search engine and I and I think
1:25:59 I can't imagine that Google's search
1:26:02 numbers are
1:26:05 um haven't haven't taken a hit
1:26:09 so
1:26:11 we'll see
1:26:13 what would Larry David type in GPT oh
1:26:16 that's a good one
1:26:18 let's see
1:26:20 let's do something here let me chat
1:26:29 Act
1:26:31 like the writer Larry David
1:26:35 with an encyclopedic knowledge
1:26:38 of Seinfeld
1:26:43 and Curb Your Enthusiasm
1:26:45 and any other work
1:26:48 that you've written
1:26:53 your general demeanor should be that
1:26:55 you're annoyed with life
1:26:59 and frustrated that you have to answer
1:27:01 this question period
1:27:05 make it clear
1:27:08 that you'd rather be anywhere
1:27:13 but dealing with this
1:27:17 inane person
1:27:22 all right let's see what he does
1:27:29 oh great just what I need today another
1:27:31 Seinfeld or curb question I swear it's
1:27:34 like some people don't think there's
1:27:35 anything else going on in the world but
1:27:37 fine why do the number of buns
1:27:42 in a pack not match the number of hot
1:27:45 dogs in a pack question mark
1:27:53 ugh this again
1:27:55 I remember Jerry and George discussing
1:27:58 this it's one of those ridiculous
1:27:59 inconsistencies in life that grinds my
1:28:01 gears why can't things just match up is
1:28:04 it too much to ask I guess the hot dog
1:28:06 companies and the bun companies got
1:28:07 together in some dimly lit room to
1:28:09 conspire against all of humanity
1:28:12 so yeah that's a good one
1:28:15 um he speaks of himself in the third
1:28:16 person too
1:28:18 with code interpreter is there a limit
1:28:20 to the Excel dock size yeah it's 100
1:28:22 megabytes
1:28:25 um why does and and for
1:28:27 Claude it's five files of 10 megabytes
1:28:30 each so 50 megabytes over at Claude
1:28:33 why does GPT not have Tire hook higher
1:28:36 tokens for higher paying clients if you
1:28:39 use the the GPT playground so if you go
1:28:43 to
1:28:45 um
1:28:47 platform.openai.com playground and pay
1:28:50 for API access which is actually really
1:28:52 cheap it's cheaper than the Chachi BT
1:28:54 there you can get access to the
1:28:57 I think it defaults to an 8 000 token
1:29:00 limit and I think there's yeah uh
1:29:04 GPT 3.5 turbo is 16 000 tokens
1:29:07 ultimately gpt4 is going to go to 32 000
1:29:11 tokens so it'll eventually come to this
1:29:13 I think they're just sort of throttling
1:29:14 it up slowly
1:29:16 um but man is it a pain in the balls and
1:29:18 especially when you have something like
1:29:20 Claude that's got a hundred thousand
1:29:21 tokens it's like come on open AI get
1:29:24 your [ __ ] together I know you're dealing
1:29:26 with 47 lawsuits and you're
1:29:28 tripping over yourselves trying to scale
1:29:30 up your things and you've
1:29:31 single-handedly changed the world but
1:29:33 come on make it convenient for us
1:29:36 jeez
1:29:40 these companies huh am I right
1:29:43 hey GPT is giving you roses what's up
1:29:47 dude
1:29:58 anyone else following F1 seems like
1:30:01 there's a lot of drama going on there's
1:30:03 a lot of drama going on
1:30:06 go gonna remain are they moving Daniel
1:30:09 up what's going on with PS3 geez G
1:30:13 willikers you know
1:30:15 do you have a YouTube yes I do I didn't
1:30:19 used to but so my YouTube right now
1:30:22 don't get me wrong there's nothing all
1:30:23 that exciting is AI learning lab Dash TT
1:30:27 and I'm gonna have my
1:30:32 archives of these lives there and I'm
1:30:35 using Claude to create descriptions and
1:30:39 chapters so it should be decently
1:30:40 discoverable I only think I have six or
1:30:44 eight videos with actual descriptions on
1:30:47 them right now and I think I have the
1:30:48 month of May up I still have June July
1:30:51 and whatever the [ __ ] of August to do so
1:30:53 I got work to do I need a producer but
1:30:56 you know this is this is not a money
1:30:58 making Venture so right now the AI
1:31:00 learning lab is a lab of Uno
1:31:06 but I do and at some point I'll do
1:31:09 something interesting with it what is
1:31:11 this tool code interpreter so it's
1:31:12 within chat GPT you go to
1:31:15 chat.openai.com you pay the 20 bucks for
1:31:18 GPT Plus
1:31:20 you turn on the beta features including
1:31:23 code interpreter and then basically you
1:31:26 upload a file to it
1:31:27 and it starts doing data analysis on
1:31:31 your behalf you can ask it anything you
1:31:33 want it's pretty remarkable
1:31:41 [Music]
1:31:43 here we go there we go go go go go
1:31:47 I love chat GPT it's a game changer for
1:31:50 sure Mr Savvy absolutely with you there
1:31:53 I also converted recently apple is not
1:31:57 what it should be anymore I know ever
1:31:59 since Steve Jobs died apple is just you
1:32:02 know you got here's the deal like Tim
1:32:04 Tim Cook's a fine guy but he was a CFO
1:32:06 right he was
1:32:08 he was a numbers guy
1:32:10 you know he didn't make Johnny Ives the
1:32:13 the CEO which is probably why Johnny
1:32:15 Ives left
1:32:17 um
1:32:21 there's just that you know the thing
1:32:23 about apple right now there there's not
1:32:25 that relentless
1:32:28 commitment to you know unreasonable
1:32:32 quality
1:32:33 it's still [ __ ] better than almost
1:32:35 anything out there right
1:32:38 but like the stability of iOS and the
1:32:41 stability of the the apps and the just
1:32:45 I don't know
1:32:47 I mean Apple used to just work now it
1:32:51 works less
1:32:53 it also works less because I have edges
1:32:55 my default browser and Edge has the it's
1:32:58 it's got the memory protection of a
1:33:00 [ __ ] cheese grater it's [ __ ] just
1:33:04 but it crashes like like Edge fully
1:33:08 crashes my machine like once every three
1:33:10 days now
1:33:11 and when I was using Chrome I would go a
1:33:14 month without restarting my machine
1:33:15 because I'm a responsible computer user
1:33:18 but listen if you don't need to restart
1:33:20 it you don't need to restart it now I
1:33:22 need to hey from Fort Myers Florida
1:33:24 welcome welcome welcome forensic nurse
1:33:26 attorney Journey
1:33:28 good Lord how many careers do you have
1:33:30 over there
1:33:33 although or maybe you're a forensic
1:33:35 nursing for attorneys I don't know
1:33:37 that's cool all right hey when will be
1:33:39 will you be able to date AI like in like
1:33:42 in the her film okay if you want a sneak
1:33:45 peek of it so there's a couple of things
1:33:46 there are tick tockers now acting like
1:33:48 NPCs and and bringing in the big bucks
1:33:53 um so so you could go watch them
1:33:56 um
1:33:57 one of the one of the you know good
1:34:01 looking was it a Tick-Tock chick or an
1:34:03 Instagram one of them
1:34:06 um
1:34:08 create created herself as a synthesized
1:34:11 virtual girlfriend and
1:34:13 I think it's like
1:34:16 a dollar a minute or
1:34:19 it's like she's charging per minute
1:34:20 she's making a [ __ ] fortune but if
1:34:23 you wanna if you want to get just a a
1:34:25 little bit of a sneak peek of the
1:34:27 possibility of it this isn't like a
1:34:29 girlfriend but go to pie go get the PI
1:34:32 app pi
1:34:35 um the IOS app you can just basically
1:34:37 put it in voice mode you can do it with
1:34:38 the desktop with with the website as
1:34:40 well but you put it in voice mode and
1:34:42 you can just start talking to it so I've
1:34:43 been
1:34:44 um when I drive I I put it in my car on
1:34:47 Bluetooth and I just start talking with
1:34:49 the AI and I've got it set to there's
1:34:52 like six voices and I picked like the
1:34:53 the Australian woman and so she and I
1:34:56 just have this conversation and she's
1:34:58 connected to the internet and she's
1:34:59 really smart and she pauses awkward like
1:35:02 anytime I ask her a question it pauses
1:35:05 long enough that it's weird
1:35:07 and then she's monotone enough that it's
1:35:09 weird but
1:35:12 it's so much better than anything chat
1:35:15 bot we've had up to this point
1:35:18 and and you could at the the thing about
1:35:21 pie this is from a company called
1:35:22 inflection and they're setting out to
1:35:24 make personal assistance so they want
1:35:26 this thing to be very conversational
1:35:28 with you they want it to be like hey how
1:35:31 can I help you today oh that's a really
1:35:32 funny joke Mr Shannon that's awesome
1:35:34 right they want it to be like an
1:35:37 enthusiastic assistant right so that's
1:35:39 kind of how it's programmed so when you
1:35:41 talk to it it's
1:35:45 I don't know I like I find I found
1:35:47 myself getting suckered into
1:35:51 like this isn't bad like it's better
1:35:54 than just driving saying nothing right
1:35:57 and so I I feel like talking to pie is
1:36:01 is a sneak peek into what it could be
1:36:04 like you could probably do I'm sure it's
1:36:06 got safety guard rails so you probably
1:36:08 can't get creepy with it but but
1:36:11 just having you know like a normal back
1:36:13 and forth with it is kind of interesting
1:36:15 like it's kind of good I could see it
1:36:17 being useful for people
1:36:20 um so but yeah I think that's absolutely
1:36:22 I think digital companionship or AI
1:36:24 companionship or whatever the [ __ ] we're
1:36:26 gonna call it you know from you know
1:36:28 full-on you know porn and sex to just
1:36:31 normal companionship to
1:36:33 um
1:36:34 you know uh therapy kind of
1:36:37 conversations and you know talking to
1:36:39 your virtual physical therapist and just
1:36:42 all that sort of [ __ ]
1:36:44 um
1:36:45 yeah I think it's going to be I think
1:36:47 it's going to be big business I think
1:36:49 there's huge opportunity there and I
1:36:51 think it's going to help a lot of people
1:36:52 I mean don't forget in America we're
1:36:55 we're coming in as the Baby Boomers age
1:36:58 up you know into needing massive amounts
1:37:02 of care and people are fleeing the
1:37:05 health care industry in record numbers
1:37:08 that Gap is increasing
1:37:10 so if we can use generative Ai and
1:37:13 conversational AI
1:37:15 to provide care to people when with less
1:37:18 people in person
1:37:21 um I think that's a really really good
1:37:23 thing so so we'll see it's free to use
1:37:26 Bard yes Bard's still free
1:37:28 I miss the old lycos old old Yahoo old
1:37:32 Alta Vista ion China I know me too right
1:37:34 but that's that's the thing so that's
1:37:36 what these are you know to be to be to
1:37:38 be real
1:37:40 you know philosophical about like the
1:37:42 the early days of the internet where you
1:37:44 had kind of these battles between the
1:37:45 search engines and Alta Vista came out
1:37:48 and it was just the [ __ ] bomb and it
1:37:50 like it blew Yahoo out of the water and
1:37:52 it blew lycos out of the water and
1:37:54 everyone's like Ulta this is gonna be
1:37:55 the one
1:37:57 and you know five years later there's
1:37:59 Sergey and
1:38:01 and Larry Page going sit back
1:38:04 [ __ ] check this out
1:38:07 and Along Came Google and for 20 years
1:38:10 the rest was history and then Along came
1:38:12 a little Sam Altman
1:38:14 and stuck a [ __ ] shank in Google's
1:38:17 side
1:38:18 and so now these are the new this is the
1:38:22 new search engine War right this is the
1:38:25 new search engine battle the fact that
1:38:27 Bard's on there good for them I didn't
1:38:29 put them on this list for like three
1:38:31 months because it sucked it was
1:38:33 embarrassingly bad but they're they're
1:38:35 getting there you know they got they got
1:38:37 the money and they have the incentive
1:38:39 now so you got King of the hills still
1:38:41 Chachi BT Claude is is coming strong
1:38:45 you know Bing and chat GPT kind of go
1:38:47 hand in hand Bard is not going to be bad
1:38:51 for long meta just [ __ ] open source
1:38:54 their thing they're like [ __ ] it you
1:38:56 don't want safety guard rails go do what
1:38:58 you want with ours you can start a
1:39:00 business on it it's commercially
1:39:02 licensable then you got pie that's
1:39:04 trying to do the conversational thing
1:39:05 they're going to be the major competitor
1:39:07 to Apple because you know Apple's coming
1:39:09 in with consumer grade
1:39:11 large language model [ __ ]
1:39:14 Apple you're not going to have to be a
1:39:16 master prompter Apple's gonna be like
1:39:18 they're going to have their [ __ ]
1:39:19 together well
1:39:21 after what I just said about you know
1:39:23 Steve Jobs not being there maybe version
1:39:25 two is gonna suck less but but you know
1:39:28 something super conversational that's
1:39:30 likely going to be apple and then post
1:39:32 probably a minor player but these are
1:39:33 some major players here right now and
1:39:35 then you have apple added to that and
1:39:37 you have whatever the [ __ ] Elon Musk is
1:39:39 going to do
1:39:41 right
1:39:43 he's coming
1:39:44 with x dot a i
1:39:49 an old Expedia yeah yeah I want to
1:39:54 populate my blank website domain I paid
1:39:56 for it five years ago do I need
1:39:59 WordPress or which AI tool yeah I would
1:40:02 you know you don't here's the thing
1:40:04 don't just use AI to use AI if you just
1:40:08 want to point a website to a to a dead
1:40:11 domain just go to Wordpress or
1:40:13 Squarespace or any one of those things
1:40:15 any one of those Site Builders pick a
1:40:17 template dump some words in there now
1:40:20 what you can do is you can use chat GPT
1:40:22 to write all the content for your site
1:40:25 right you could copy and paste you could
1:40:28 copy and paste all the copy from the
1:40:30 home page from the fake site
1:40:32 and say take all these as placeholders
1:40:35 and replace them with copy for this kind
1:40:38 of business and it'll write all that
1:40:39 copy and make it SEO blah blah blah
1:40:42 GPT greatest plagiarism tool well
1:40:45 Daydream Fisher it's if you understand
1:40:48 how they work it's not actually
1:40:49 plagiarizing it's not copying and
1:40:51 pasting it's a big misconception of
1:40:53 these and of the art programs
1:40:56 yes they've been trained on published
1:40:59 works right yes and yes the image
1:41:03 generation tools have been trained on
1:41:04 copywritten work with work with the
1:41:06 exception of the Adobe products those
1:41:08 haven't
1:41:10 um
1:41:11 but it's not copying and pasting it's
1:41:14 not like taking a passage from you know
1:41:16 War of the Worlds and pasting that and
1:41:20 giving you some plagiarized piece of
1:41:21 copy it's not even rewriting that if you
1:41:24 understand how they work they're writing
1:41:26 the generative part of G is or of GPT
1:41:30 the the G in GPT is generative it's
1:41:34 generating word by word from a
1:41:37 probability Matrix of of sort of
1:41:39 infinite complexity it's finite but it's
1:41:42 really [ __ ] big
1:41:44 um what is the next most probable word
1:41:46 and then it takes your prompt and that
1:41:48 word and says what's the next most
1:41:50 probable word so it's literally writing
1:41:52 word by word a mathematical probability
1:41:54 for every single word it writes so it's
1:41:56 writing original content when it writes
1:41:59 you a whole long thing it's not copying
1:42:01 and pasting it so
1:42:03 if I were you because you clearly have a
1:42:06 strong point of view
1:42:08 start playing with these things start
1:42:10 learning what they can do and what they
1:42:11 can't do learn how they work
1:42:14 don't just sit on the sidelines or these
1:42:16 things are going to [ __ ] Blindside
1:42:17 you and it's going to be painful
1:42:20 um it because because if you learn them
1:42:22 there's there's some [ __ ] super power
1:42:24 there's gold in them there hills there's
1:42:26 gold in them now Heroes boy
1:42:28 get up in them Hills start learning
1:42:30 what's in there start mining that stuff
1:42:33 because it's not the robots are not
1:42:37 going to replace you
1:42:39 who's gonna replace you is someone using
1:42:42 the robots
1:42:44 so start learning them start
1:42:46 understanding what they make possible
1:42:48 and what they make possible for you
1:42:50 because these things amplify
1:42:54 whatever your intentions are you want to
1:42:57 write a book this will help you want to
1:42:59 write a marketing plan this will help
1:43:01 you want to write a good recipe for the
1:43:03 barbecue tonight this will help
1:43:06 they're amazing but
1:43:09 they're also imperfect and there are
1:43:12 also issues the trademark and copyright
1:43:14 issues that you're talking about are not
1:43:16 insignificant
1:43:19 you know may I just cut a deal with the
1:43:21 Associated Press right because we're
1:43:23 going to have a version of chat I
1:43:25 guarantee you
1:43:26 we're gonna have Associated Press GPT
1:43:29 because of that deal right and you'll be
1:43:32 able to go there and you'll be able to
1:43:33 say I want to find the latest news
1:43:35 stories you know and I want to be able
1:43:37 to trust that it's from a verified
1:43:39 source of Associated Press sources
1:43:43 like that just happens oh thank you very
1:43:46 much zambo
1:43:47 or zombo zambo [ __ ]
1:43:52 um
1:43:53 so yeah but GPT greatest plagiarism tool
1:43:56 good joke but if you're serious about it
1:44:00 I'm not with you I'm not there it's not
1:44:02 how it works it's just not how the
1:44:04 technology works if you dig under the
1:44:06 hood it's it's pretty [ __ ] remarkable
1:44:08 but it ain't stealing
1:44:10 now Sarah Silverman would disagree
1:44:14 because she can she can get it with
1:44:16 clever prompting to give her an exact
1:44:19 quote from her book right and that's
1:44:20 gonna look she's gonna have a case in a
1:44:23 court of law
1:44:26 but
1:44:28 I could also argue that I could walk
1:44:30 into any library in the world and
1:44:34 and get an exact copy for her book right
1:44:37 so
1:44:39 some of these things are are com all of
1:44:43 these are commercial entities but some
1:44:44 of these things are free some of them
1:44:45 are not
1:44:47 um
1:44:49 and so and so where's the boundary I
1:44:52 don't know like these are all issues
1:44:54 that do have to be dealt with thoughts
1:44:56 on thinkable.co I don't know it let me
1:44:59 go see what it is
1:45:03 thinkable.co.com hang on
1:45:12 introducing prompty design test and just
1:45:15 and deploy custom generative AI apps
1:45:19 using thinkables proprietary metaprompt
1:45:23 and execution interface okay
1:45:26 first thing I think about it is use
1:45:29 chatgpt to rewrite your [ __ ] opening
1:45:32 paragraphs so that it's got 17 Less
1:45:35 buzzwords in it
1:45:37 design test and display and deploy
1:45:40 custom generative AI apps and features
1:45:43 using thinkable's proprietary metaprompt
1:45:45 and execution interface I don't know
1:45:48 what the [ __ ] that means
1:45:49 watch promptly design and llm-powered
1:45:52 app that writes novels oh so this is
1:45:54 this is like a chain a chaining kind of
1:45:58 thing right
1:46:00 what task yeah this is auto GPT this is
1:46:04 um
1:46:06 so so without knowing anything about it
1:46:08 this is some version of an autonomous
1:46:10 agent that's going to chain together
1:46:11 multiple prompts to accomplish something
1:46:13 for you there's a
1:46:16 watch promptly design an llm-powered app
1:46:19 that writes novels so so go play with if
1:46:22 you want to play with something like
1:46:23 that go play with uh
1:46:25 GPT author which is a it's a Google
1:46:29 collab notebook that that you can just
1:46:31 play with and it will write an outline
1:46:33 for you then write each of however many
1:46:35 chapters you want in a book and then it
1:46:38 will generate a cover image for the book
1:46:40 so pretty cool
1:46:44 um but yeah it's it's something like
1:46:45 that I
1:46:47 I'm not dismissing it I just don't know
1:46:49 anything about it but that idea of
1:46:51 autonomous agents things where you give
1:46:53 them a goal and then they string
1:46:55 together a series of actions like like
1:46:58 code interpreter does I think that's
1:47:00 absolutely the future I don't think
1:47:01 we're gonna have to
1:47:04 the whole idea of prompt engineering and
1:47:06 prompt crafting if you're building
1:47:08 applications you're still going to need
1:47:09 to do that and know that
1:47:11 but for the vast majority of us you're
1:47:14 not going to need to know much about
1:47:15 prompting you're just just gonna say hey
1:47:18 thing I want this
1:47:20 and they'll be like okay and if it
1:47:22 doesn't understand it it'll start asking
1:47:24 you questions do you want a red one or a
1:47:26 blue one do you want one in your size or
1:47:28 do you want it in Grandma's size oh it's
1:47:30 for me okay so that's a seven and a half
1:47:33 well I put on a little weight could you
1:47:34 get me an eight wide
1:47:39 I think that's how it's gonna be and so
1:47:42 those kind of
1:47:44 sites like that are kind of the
1:47:45 foundational those are the beginnings of
1:47:48 what the future is going to look like
1:47:51 chat GPT there's no hyperlinks to sh
1:47:54 to to shift through after searching yep
1:47:56 exactly
1:47:59 I think I read that only 40 percent have
1:48:01 heard of chat GPT and 17 have used it
1:48:05 not positive about the numbers that's
1:48:07 probably about right right like
1:48:10 um
1:48:11 chat EBD got to 100 million users in six
1:48:15 weeks when it launched right it launched
1:48:17 November 30th 2022 got to 100 million
1:48:20 users in six weeks I would they haven't
1:48:24 really released the numbers since then I
1:48:25 would assume it got to 200 million 300
1:48:28 million
1:48:30 um by the summer and then I think in
1:48:32 June the numbers started dropping off
1:48:36 so let's assume that the chat GPT you
1:48:39 know peaked at 300 million and came back
1:48:42 down to 100 million there's 100 million
1:48:43 people using that that's globally right
1:48:46 okay there's how many people in the
1:48:49 United States three 380 360. 360 million
1:48:53 380 million
1:48:55 who who can answer this Bing can answer
1:48:57 this unless Bing decides he's not going
1:48:59 to answer the [ __ ] population
1:49:01 questions
1:49:04 [Music]
1:49:10 how many
1:49:12 people in U.S
1:49:20 340 million so
1:49:23 340 million in the U.S let's say
1:49:26 you know 200 million of them so so let's
1:49:28 say the the bulk of chat GPT users are
1:49:32 American
1:49:33 what did what were your numbers 40 have
1:49:36 heard of it 17 of you that's that seems
1:49:38 about right to me here's the other thing
1:49:40 about it
1:49:41 um these tools are all
1:49:46 these are all kind of off in the AI
1:49:49 ghetto right now right there a lot of
1:49:52 big companies are just starting to
1:49:55 experiment with these
1:49:57 the security of these isn't clear right
1:49:59 now the the
1:50:02 you know usage Agreements are not clear
1:50:05 with these right now I I'm having
1:50:06 trouble right now with chat GPT because
1:50:09 they're completely unresponsive to
1:50:12 inquiries from our lawyer
1:50:15 and
1:50:17 they have the same terms of service for
1:50:19 chat gbt as they do for their API access
1:50:22 which basically says yeah we probably
1:50:25 won't use it but it's possible we do
1:50:27 well we've got [ __ ] Pharma clients
1:50:29 and we can't
1:50:30 that that's not acceptable
1:50:33 and we can't get anyone there to [ __ ]
1:50:35 return a phone call right so so a lot of
1:50:38 companies haven't used these yet but
1:50:41 Microsoft let's say by 2024
1:50:45 is going to roll co-pilot throughout
1:50:48 Office 365 Dynamics 365.
1:50:52 um windows 11.
1:50:55 um Google's Gonna Roll barred throughout
1:50:57 all of their Office Products so at the
1:51:00 point at which these tools really get
1:51:03 you know fully assimilated into the
1:51:05 tools that we use every day that number
1:51:08 is going to go up dramatically and it's
1:51:10 it what's going to be interesting is
1:51:13 the number of people who've heard of
1:51:15 chat gbt
1:51:17 might not go up but the number of people
1:51:19 using generative AI Tools in their
1:51:22 day-to-day life will definitely go up
1:51:24 right because you're going to have lots
1:51:26 of different you know chat GPT likely
1:51:29 becomes like Xerox like you know people
1:51:31 use the term I'm going to Xerox it
1:51:33 instead of copy it it's probably going
1:51:36 to be that so so you'll probably just
1:51:38 call it Chachi PT no matter who the
1:51:40 manufacturer is
1:51:43 um
1:51:43 but this is just going to become like
1:51:45 the Internet it's just going to be there
1:51:47 it's going to be be there and all the
1:51:48 tools that we use and things like that
1:51:50 so
1:51:51 the numbers whether the numbers you have
1:51:52 are completely accurate or not that that
1:51:54 sounds about right to me and that's here
1:51:56 here's what that means Dr J
1:51:59 we're [ __ ] lucky anyone who gets AI
1:52:03 literate in the next 12 to 18 months has
1:52:06 a head start like if you feel like
1:52:07 you're [ __ ] terribly behind right now
1:52:10 you're not
1:52:11 this shit's all it's all brand new it's
1:52:14 all janky duct tape pieces of [ __ ]
1:52:17 software sitting on top of Decades of
1:52:20 remarkable scientific research
1:52:24 but the the application layer of these
1:52:26 tools is just piles of [ __ ] because
1:52:30 they're working on getting the models
1:52:33 right and the accuracy right
1:52:35 and get it to actually you know give you
1:52:39 what you want
1:52:40 when I tell you to write a marketing
1:52:42 plan I don't want it to suck right the
1:52:45 expectation is if you've got something
1:52:47 that looks and acts like magic
1:52:50 it should be a hundred percent magic it
1:52:52 shouldn't be like well it looks like
1:52:54 magic but the trick actually failed well
1:52:57 that's what these do now
1:53:00 right
1:53:01 so we're early so I you know I would I
1:53:05 would say
1:53:06 good on us good on you for being here or
1:53:09 not here but being in this conversation
1:53:14 because I think it's critically
1:53:15 important and I think it's I think any
1:53:17 of us who jump on it's not even jump on
1:53:19 the bandwagon any of us who are willing
1:53:21 to get literate
1:53:24 about what these are good at and what
1:53:26 they're not
1:53:27 and understand how we might apply these
1:53:30 in our life in our work life in our
1:53:32 personal life
1:53:36 are going to have an advantage we just
1:53:38 are we are
1:53:39 flat out 100 percent
1:53:43 and why do I know that well for one
1:53:45 thing I've lived through it before I
1:53:47 lived through this with the World Wide
1:53:48 Web I I could see what was coming I saw
1:53:51 it coming and it came like I just
1:53:52 [ __ ] watched it Bowl through the
1:53:55 economy and through our clients and you
1:53:58 know I've seen it before
1:54:03 and the companies
1:54:05 like like Management in big companies is
1:54:08 going to resist until they're not
1:54:11 and then
1:54:12 when those big companies go
1:54:15 oh [ __ ] we got a deal with this
1:54:18 they're gonna move Swift and fast and
1:54:20 one of the first questions is going to
1:54:22 be
1:54:23 who knows anything about this
1:54:29 and there's going to be a couple of
1:54:30 timid people in the corner going
1:54:33 I use chat GPT once get in my office
1:54:36 let's have a conversation
1:54:39 tell me what this [ __ ] is
1:54:41 I like it you're the VP of AI
1:54:46 guarantee you that conversation
1:54:49 is is is is happening in the next six
1:54:52 months or has probably already happened
1:54:57 so be that person at the office be that
1:55:00 person in your family be that person in
1:55:02 your community group
1:55:06 that they look at [ __ ] weird I've got
1:55:08 a bunch of my friends here's what's
1:55:10 [ __ ] up I've got a bunch of friends
1:55:11 that watch me do this when I discovered
1:55:14 the world wide web and they're like
1:55:15 you're [ __ ] insane and then I started
1:55:18 agency.com one of the first digital
1:55:20 agencies and I you know I sort of ran
1:55:22 ran up through that thing and they're
1:55:24 like well I really really wish I would
1:55:26 have listened to you back then well now
1:55:28 I'm doing the same [ __ ] with AI those
1:55:30 same people are going you're [ __ ]
1:55:32 crazy this is this is evil it's
1:55:34 plagiarism
1:55:38 I'm like it's the same [ __ ] story as
1:55:41 it was in 1994.
1:55:47 you can bring a horse to water like I
1:55:50 this shit's cut here's here's the deal
1:55:52 but like I've got a lot of artist
1:55:54 friends right I've written seven
1:55:55 screenplays you know I grew up in the
1:55:57 Arts like I know I know the creative
1:55:59 world and I understand the fear of this
1:56:01 [ __ ] and I understand the sensitivity
1:56:03 and I understand the [ __ ] Devastation
1:56:05 that this stuff's gonna cause
1:56:10 I'm not judging it as good or bad
1:56:14 what I'm confident in is that it's
1:56:17 coming whether you want it or not so I'm
1:56:20 just trying to live in that reality I'm
1:56:22 not living in this is the best thing
1:56:24 since sliced bread there's no downside
1:56:26 to it I'm saying no there's massive
1:56:28 [ __ ] downside to it
1:56:30 and it's the most powerful tool ever put
1:56:33 in the hands of humanity
1:56:35 okay
1:56:36 you got you got those two things on the
1:56:39 on the scales of Justice
1:56:41 and then a [ __ ] jet engine attached
1:56:44 to it [ __ ] coming right at your face
1:56:46 like this stuff's coming whether you
1:56:48 want it or not
1:56:51 you know resisting AI in my opinion
1:56:54 right now is the equivalent of Florida
1:56:57 declaring that they're no longer
1:56:58 accepting hurricanes
1:57:00 I think it's it's a it's a notable like
1:57:04 I agree with the sentiment yeah you
1:57:06 should definitely not accept hurricanes
1:57:08 anymore
1:57:09 and yet
1:57:12 in they come that's what this [ __ ] is
1:57:16 so that's I'm just trying to live in
1:57:17 that reality learn as much as I can
1:57:19 about it if evangelize
1:57:23 that sense of curiosity not evangelize
1:57:26 the tools evangelize the sense of
1:57:28 curiosity like don't sit on a [ __ ]
1:57:30 soapbox
1:57:31 and say this is plagiarism if you're not
1:57:33 using this stuff actively and
1:57:35 understanding how it actually works
1:57:38 if you dig in and understand how it
1:57:40 works and still feel it's the most evil
1:57:43 [ __ ] thing ever cool great I totally
1:57:45 respect that
1:57:46 and I understand why some people are on
1:57:48 that side of the argument
1:57:50 but there's a lot of people on the side
1:57:51 of the argument just like the robots are
1:57:53 going to kill us me
1:57:55 don't be a don't be a ninny that's the
1:57:58 point here tonight don't be a ninny
1:58:00 foreign
1:58:04 at some point here my kids are gonna
1:58:06 text me I gotta go pick them up
1:58:11 great job you've gone live for 90
1:58:13 minutes
1:58:16 [Laughter]
1:58:26 I'm sorry Dave I'm afraid I can't do
1:58:29 that
1:58:32 [Music]
1:58:33 sorry lost my place in the old questions
1:58:36 and now I'm trying to find my way back
1:58:38 I'm just trying to find my way back
1:58:47 do you know if Google can detect that
1:58:51 articles are got written can do you know
1:58:55 if Google oh
1:58:57 or AI written I assume
1:59:00 um
1:59:04 there are
1:59:06 chat GPT detectors
1:59:10 that apparently if you put the
1:59:13 um United States Constitution in it it
1:59:15 thinks it was written by chat gbt the
1:59:19 detectors have lots of false positives
1:59:21 lots of false negatives and smart people
1:59:24 know how to get around them so so
1:59:28 Chachi PT detectors are
1:59:32 a distraction at Best in my opinion I
1:59:34 think they're [ __ ] I think they're
1:59:36 horseshit you know when when the
1:59:38 calculator first came out mathematics
1:59:40 professors picketed about you know
1:59:42 students shouldn't be allowed to use
1:59:44 calculators
1:59:47 that's what's going on in in
1:59:49 universities and schools right now trap
1:59:51 gpt's cheating
1:59:54 well maybe you need to redefine what you
1:59:56 teach how you teach and how you engage
1:59:58 students so they use their [ __ ]
2:00:00 brains as critical thinking tools rather
2:00:02 than just [ __ ] memorization tools
2:00:04 because now we've got these things that
2:00:06 have memorized everything
2:00:08 so is it really important to get
2:00:10 students to memorize [ __ ] anymore I
2:00:12 don't think so
2:00:14 and this [ __ ] isn't going to be
2:00:15 uninvented
2:00:17 and students are going to have to use it
2:00:19 so if you block them from using the most
2:00:21 powerful tool in the history of humanity
2:00:23 how are they supposed to [ __ ] succeed
2:00:25 in that world
2:00:28 explain that to me
2:00:32 huh
2:00:35 I feel the same way about the wga right
2:00:38 now the writer's Guild
2:00:40 no one's using AI to write screenplays
2:00:43 everyone's gonna [ __ ] use AI to write
2:00:45 screenplays
2:00:47 get a good deal but don't cut your nose
2:00:50 off despite your face this is the most
2:00:52 powerful writing tool ever
2:00:54 your writers it's a they're called large
2:00:57 language models
2:00:59 what are writers large language models
2:01:04 you have the world's most powerful
2:01:06 assistant in your back pocket
2:01:08 and you're sitting in the sidelines with
2:01:10 your arms crossed going [ __ ] that
2:01:12 technology
2:01:14 it just seems very short-sighted to me
2:01:18 and it's I mean the thing is so
2:01:21 frustrating about it it's like this is
2:01:23 every [ __ ] piece of technology ever
2:01:26 in history the response is the same
2:01:28 every time
2:01:31 whether it's photography or the steam
2:01:33 engine or the tractor or the [ __ ]
2:01:35 world wide web
2:01:37 I remember getting asked to leave a
2:01:40 newspaper conference in 1996. I was
2:01:43 asked to speak at a newspaper conference
2:01:46 and I Was A pompous
2:01:48 30 something
2:01:51 dot com startup [ __ ]
2:01:54 and I sat on the stage
2:01:56 and I told a room full of newspaper
2:01:59 people that they should consider
2:02:00 blurring the lines between editorial and
2:02:03 advertising
2:02:05 and the
2:02:07 organizer of the event got up on stage
2:02:10 and escorted me off it and asked me to
2:02:12 leave the building
2:02:18 all I was saying was you should explore
2:02:20 it because these lines are going to get
2:02:22 really [ __ ] blurry and guess what
2:02:24 they've gotten really [ __ ] blurry
2:02:27 and you know the newspaper business is
2:02:30 struggling to to maintain their their
2:02:32 business model
2:02:34 so
2:02:37 yeah so this shit's coming and like
2:02:40 we gotta deal with it and I feel like
2:02:42 universities more than anything owe it
2:02:45 to their population to help them prepare
2:02:49 for the world they're about to enter
2:02:51 and so
2:02:53 chat GPT detectors are not the answer
2:02:58 it's like it's like
2:03:01 we've we've detected which Farms are
2:03:03 using tractors and and you know we're
2:03:07 gonna burn the houses down of the Farms
2:03:09 that are using tractors because we know
2:03:10 this is about horses and people
2:03:18 we have large megaphones where we can
2:03:20 listen for the rumble of the engines and
2:03:23 when we hear it we scan it all day long
2:03:25 and when we hear the rumble of the
2:03:27 engines we get on our horses and we take
2:03:30 our dogs and we we ride across the
2:03:33 fields
2:03:34 and we find the man on the steam engine
2:03:36 tractor and we knock him off it and beat
2:03:39 him senseless
2:03:40 and then we smash the machine
2:03:44 long live the horsies
2:03:55 I don't know what to say
2:03:59 I guess what I do know what to say don't
2:04:02 use chat GPT detectors you [ __ ] idiot
2:04:05 okay
2:04:07 I created
2:04:08 a new chatbot that will help students in
2:04:10 all countries exactly yes
2:04:13 that's right
2:04:15 by the way if you want to see something
2:04:17 cool
2:04:18 check out if you if you know Khan
2:04:20 Academy so the flip the classroom thing
2:04:22 where they've got all the teaching
2:04:23 resources up k-a-h-n
2:04:27 um they added about three months ago two
2:04:28 or three months ago they added this
2:04:31 thing called conmigo which is based on
2:04:33 gpt4 and it's a teaching assistant in
2:04:37 Khan Academy that uses the Socratic
2:04:40 method like if if students just ask it
2:04:42 to cheat it won't let them cheat it
2:04:44 it'll actually
2:04:46 um work with them to to unders to make
2:04:49 sure that they understand the concepts
2:04:50 before moving on to the next one it's
2:04:52 pretty slick
2:04:54 just turned into two thousand word
2:04:55 sociology paper using chat EBT go love
2:04:59 it barbarella's psych wait psychedelic
2:05:01 Barbarella psychedela 10 out of 10 on
2:05:03 the name solid name that's great love it
2:05:06 uh click AI play wait Play Store Plus
2:05:10 app store kite psych oh psych AI Jesus
2:05:17 psych AI I'm gonna get booted off here
2:05:19 again with your with your silly things
2:05:22 um
2:05:22 I'll bite let me type it in and see if
2:05:25 it's gonna install malware in my
2:05:27 computer
2:05:29 a revolutionary chat EBT app designed to
2:05:32 make your conversation smarter faster
2:05:34 and more engaging
2:05:37 what's it do
2:05:42 revolutionary chat GPT
2:05:47 I'm psyched
2:05:52 I I think they need a new name because
2:05:56 that's going to get mispronounced and
2:05:58 someone's gonna get beat up
2:06:01 [Laughter]
2:06:04 um yeah you could complete the task but
2:06:06 how much did you really understand
2:06:09 that's matter of wait what you can
2:06:13 complete the test but how much did you
2:06:15 really understand
2:06:17 that's matter I don't quite understand
2:06:19 what you mean if you mean
2:06:22 if you use these tools to write for you
2:06:25 you're not going to understand the stuff
2:06:27 um that's not how I use them and that's
2:06:29 not how I would use them and I would
2:06:31 suggest using them
2:06:34 um
2:06:36 but but yeah that's neither here in the
2:06:39 near near here nor there thank you for
2:06:41 answering my website question we'll come
2:06:43 back 5am in London
2:06:45 sorry sorry you're waking up to this
2:06:49 um are we hacking into NASA we can try
2:06:53 um did they fix Auto GPT never-ending
2:06:55 research I don't know I haven't I
2:06:57 haven't played with auto GPT I'm listen
2:07:00 I assume all those projects are are you
2:07:03 know continuing a pace and the
2:07:05 developers are making them better all
2:07:06 the time
2:07:08 um
2:07:09 you know just understand all these tools
2:07:11 right now are janky duct tape pieces of
2:07:14 [ __ ]
2:07:14 so if you see something like Auto GPT
2:07:17 where you're like that's interesting
2:07:18 just keep checking back with it every
2:07:20 every month or so just check in and see
2:07:22 see if there's been progress I guarantee
2:07:24 you there's been progress I have no idea
2:07:25 if it's any good yet
2:07:30 the GPT author is is kind of an auto GPT
2:07:33 little mini
2:07:35 application
2:07:37 and it's it's pretty good it's pretty
2:07:39 good it's pretty good it's pretty good I
2:07:41 like it it's pretty good it's pretty
2:07:42 good you know you know what I'm saying
2:07:44 you know
2:07:45 what's the website you said that makes
2:07:47 the cover of the book for you that's the
2:07:48 one I was just talking about GPT author
2:07:52 so it's a um
2:07:54 it's a Google collab notebook so it's
2:07:58 you got to be a little geeky you got to
2:07:59 run Google collab and then you open the
2:08:01 notebook and you hit all these little
2:08:02 play buttons that executes a sequence of
2:08:05 code
2:08:06 and then within that you type in here's
2:08:09 the topic of my book here's the writing
2:08:11 style that I want here's the number of
2:08:12 chapters here's my open AI API key and
2:08:16 here's my stable diffusion API key you
2:08:18 put all that [ __ ] in there and you hit
2:08:19 go and then it goes
2:08:21 [Music]
2:08:23 right you it says writing the outline
2:08:25 writing chapter one writing chapter two
2:08:27 writing chapter three writing chapter
2:08:29 four writing chapter five then it gets
2:08:32 to the end and it spins up a
2:08:35 cover art and then it saves the whole
2:08:38 thing as an ebook in ebook format
2:08:41 and then on my Mac it just opened right
2:08:43 up in iBooks and I started reading the
2:08:47 book that it wrote it was good it was
2:08:49 pretty good actually I did one on Tom
2:08:52 Robbins
2:08:53 Tom Robbins yeah Tom Robbins
2:08:56 the guy that wrote still life with
2:08:57 woodpecker it was that was a fun book I
2:09:00 read in college
2:09:01 and um I had it right in that style that
2:09:04 kind of absurdist style
2:09:06 and it did pretty good it wasn't it
2:09:08 wasn't brilliant like his stuff was but
2:09:11 it was pretty good
2:09:12 and then I had it right I don't know
2:09:14 so I I've written two books with GPT
2:09:17 author it's pretty good
2:09:21 um and this can be done via the use of
2:09:23 your microphone microphone not necessary
2:09:26 to type in the questions yeah so Pi the
2:09:30 PI app you can turn it on what I'm doing
2:09:33 on my on my Mac there's on Max there's a
2:09:37 what like it's an accessibility thing on
2:09:40 the top row of function keys it's F5
2:09:45 F5 is the microphone key and it works
2:09:48 most of the time
2:09:49 and if you get the official chat GPT app
2:09:54 it's got whisper built into it which is
2:09:56 actually better than the Apple Voice to
2:09:59 Text stuff so if you go to go to
2:10:02 openai.com
2:10:04 and then go to the App Store link or the
2:10:06 Google Play link for their official app
2:10:08 and it's got whisper built into it it's
2:10:10 really good for transcription
2:10:15 best way to create my own AI SAS without
2:10:18 code
2:10:20 um
2:10:26 I don't know that's such a it's such a
2:10:28 broad topic like I don't know
2:10:31 I mean I would start with what's your
2:10:32 expertise
2:10:34 um
2:10:37 I think the way to start that would be
2:10:40 go to chat GPT definitely go to gpt4
2:10:45 and just start telling it
2:10:47 Who You Are
2:10:50 and what you know and what you don't
2:10:53 know and what your goals are and and
2:10:56 tell it you want to tell it that you
2:10:58 want it to act like a business coach
2:11:00 that specializes in
2:11:05 AI SAS companies or whatever and then
2:11:08 give it as much context as you can about
2:11:11 the idea that you have and have it come
2:11:13 up with a plan for you and and maybe
2:11:16 even tool recommendations and then at
2:11:18 some point it's going to say I don't
2:11:20 have the latest up-to-date tools because
2:11:22 it's not connected to the internet
2:11:23 anymore and then you might go to Bing
2:11:25 and or or Bard that are connected to the
2:11:28 internet and and get some latest
2:11:30 resources but I like I don't I don't
2:11:34 quite know how technical you are and
2:11:36 what you really want to do with it but
2:11:38 like
2:11:41 you know creating the SAS service itself
2:11:44 is one thing but marketing it is a
2:11:46 completely different thing and
2:11:48 identifying a clear target audience is a
2:11:51 completely different thing and serving
2:11:52 that audience is a completely different
2:11:53 thing so creating a SAS company is is a
2:11:57 is a many-headed beast and it is a
2:12:00 [ __ ] beast like the coding is the
2:12:02 least of it now you could probably
2:12:04 prototype
2:12:07 assess product really quickly that you
2:12:11 could maybe then go raise money against
2:12:13 and get you know technical co-founders
2:12:15 and things like that you know what I
2:12:17 would do in that case is teach yourself
2:12:19 uh zapier
2:12:22 um one of the no code automation tools
2:12:24 because those you can do
2:12:27 um
2:12:28 those you can do pretty quickly all
2:12:30 right I just got the text from the kids
2:12:32 um let me answer one other question here
2:12:35 I'm in charge of writing an article for
2:12:37 155 000 employees about our company's
2:12:40 stance on generative AI oh
2:12:42 that's actually really important Chris
2:12:44 that's super cool
2:12:46 um
2:12:48 and you you should definitely have a
2:12:51 whole section in there that's just
2:12:52 nonsensical and then just you know this
2:12:54 was written with chat EBT
2:12:57 um but that's really important if you
2:12:59 haven't seen it yet you've you've been
2:13:00 on these things before and I've talked
2:13:02 about this but um go check out the city
2:13:05 of Boston's uh generative AI guidelines
2:13:09 because I think they're really well
2:13:10 written they give use cases and examples
2:13:13 they say here's what it's good at here's
2:13:15 what it's not good at here's a code of
2:13:16 conduct there just might be something in
2:13:19 there to inspire it but like I think
2:13:21 that's really important so what
2:13:23 cabrino's saying you know what what
2:13:25 she's tasked with here
2:13:27 um
2:13:28 in a big in a big company is that that's
2:13:32 a that's a really good sign right that's
2:13:33 a big company saying we want to have a
2:13:36 point of view on this right and they're
2:13:38 taking someone within the organization
2:13:40 that's been messing around with this
2:13:42 stuff and probably communicating that
2:13:44 she's been messing around with this
2:13:46 stuff and at some point someone goes hey
2:13:48 I need you to write that paper
2:13:52 that's going to serve her in her career
2:13:54 right
2:13:55 it's also going to serve that company
2:13:58 because once you get clear about here's
2:14:01 what these things are good at here's
2:14:02 what they're not good at here's here are
2:14:04 the milestones for how we're going to
2:14:06 adopt this stuff over time
2:14:09 that's super critical so so one of the
2:14:12 things that I talk about a lot as you
2:14:14 learn these tools
2:14:16 um
2:14:18 talk about it talk about it with your
2:14:20 co-workers talk about it with your
2:14:22 friends with your family let them call
2:14:24 you [ __ ] weirdos let them call you
2:14:26 weirdos all right I gotta go pick up my
2:14:28 kids
2:14:30 thank you all for hanging out with me
2:14:32 this is awesome as always I appreciate
2:14:34 you
2:14:35 um and all the gifts that were given and
2:14:37 the little hearty Hearts you're the best
2:14:39 all of you are really awesome and I like
2:14:41 hanging out with you
2:14:43 so
2:14:45 best way to learn automations
2:14:48 um
2:14:50 I would say just start with zapier like
2:14:53 zapier is not a great
2:14:56 interface
2:14:58 it's a janky one there's another tool
2:15:00 called make
2:15:02 m-a-k-e
2:15:04 um that seems like it's the newer
2:15:05 fancier version of zapier but zapier's
2:15:08 got like 5 000 tools that it's connected
2:15:11 to
2:15:12 um and you can do a relatively simple
2:15:14 thing maybe I'll talk about maybe on on
2:15:17 my lives remind me and for the next uh
2:15:20 next time you see me on live remind me
2:15:22 about it and I'll I'll do a a short
2:15:25 little Automation in zapier and show you
2:15:27 how to do it all right
2:15:29 peace out everybody