aiLL May 1, 2023 - AI is Accelerating Faster Than We Can Keep Up

Description

Chapters

Transcript

0:00 foreign
0:07 with the AI learning lab
0:10 let's see notifying viewers that I am
0:14 live
0:16 let's see there's a person hello Kyle
0:19 Shannon with the AI learning lab
0:22 logging in late at night well not that
0:24 late it's nine o'clock here in Denver
0:27 hey everybody Welcome in if you have any
0:29 questions about AI jump in ask me I'm
0:31 happy to answer anything I can uh if I
0:35 can't answer it I'll tell you I don't I
0:37 try not to make crap up and I don't know
0:39 what's going on with my hair it's a
0:41 little puffy
0:43 that's a bit of a nightmare so I
0:45 apologize for that but what are you
0:46 gonna do
0:47 what are you gonna hey at least I got
0:49 hair right you know at this age I mean
0:52 my forehead is it looks like I'm getting
0:54 smarter because my head's getting bigger
0:55 but uh anyway yeah so uh let's see
0:59 what's today today's Monday
1:01 um there was some big news item what was
1:03 it let me go see if I can find there was
1:05 an interesting news item
1:08 that I just saw where did I put that
1:13 ask any questions get any questions
1:15 about AI feel free to dump them in the
1:17 chat and I will answer what I can
1:20 let's see what did I oh I know they did
1:28 lamini
1:30 a-i-l-a-m-i-n-i dot a i
1:33 this thing here
1:36 and apparently
1:39 it's really good at training your own
1:41 data which I think I think that's going
1:43 to be I think that's going to be the
1:45 real big the thing that's gonna really
1:49 bring these large language models into
1:52 their power is when we can personally
1:55 train them on our own data and then as
1:57 businesses we can train them on our own
1:59 data data
2:00 um have I have I done anything
2:02 interesting with agent GPT I have not
2:06 um I've played around with it a little
2:08 bit uh you know kind of
2:11 hinting it figuring out how to get it up
2:15 and running but I just I just haven't
2:16 things have been busy at work because I
2:19 have Incorporated AI into my company
2:22 vision and we've been presenting that
2:24 and that's been going really really well
2:26 where clients are like we want to talk
2:28 about this I've been really busy doing
2:30 that
2:31 um but I think I think all of the
2:33 autonomous agent stuff is going to be
2:35 huge
2:36 um and uh generative agents are really
2:39 really interesting and then things like
2:41 Auto GPT agent gbt baby AGI
2:46 all of those things well I mean what's
2:48 amazing is those tools are
2:52 um about a month old I mean you know in
2:55 their in their entirety
2:57 um and and they're they're already
2:59 having a dramatic impact
3:01 um so you know I can only imagine six
3:04 months or a year from now where they've
3:06 actually got those things dialed in and
3:07 and you know they don't they don't stall
3:09 out and they don't go into recursive
3:11 Loops that just end up running forever
3:13 um if you have any questions about AI
3:15 feel free to join hey everybody it's uh
3:17 Kyle Shannon with the AI learning lab uh
3:20 I like talking about AI stuff because
3:21 it's all I can think about anymore and I
3:24 am insufferable
3:26 people ask me about the weather and I'm
3:28 like well you know what chat gbt would
3:30 say about that uh but but it's you know
3:34 the reason I'm so passionate about this
3:36 stuff is I I think this is
3:39 um
3:40 I think it's one of the more profound
3:42 technical advances in humanity
3:46 um
3:47 I was just too early to be part of the
3:49 PC Revolution I was right in the middle
3:51 of the web Revolution and did a lot of
3:53 uh innovating during the mid 90s and
3:56 early 2000s and
3:58 this feels a lot like those days
4:02 um but it is happening significantly
4:05 faster and way more profound in fact one
4:09 of the stats I've said this a couple of
4:10 times on the channel and even on these
4:12 lives but but I think it's it's quite
4:15 important the the World Wide Web to get
4:18 to 100 million users it took eight years
4:21 so I started one of the first digital
4:24 agencies in 1995 I was in it for
4:28 um hi larisha
4:30 larisha that I work with I guess is in
4:32 my life she's sending me gifts
4:35 hi Larry [ __ ] that's really funny
4:37 um
4:39 I started one of the first digital
4:41 agencies we grew really really big
4:42 really really fast I was only in uh it
4:45 was a company called agency.com I was
4:46 only in a part of agency.com for seven
4:49 years it took eight years for the World
4:51 Wide Web to get to 100 million users it
4:55 took chat GPT
4:57 um a month and a half six weeks to get
4:59 to 100 million users so the the speed of
5:03 adoption is uh absolutely insane and oh
5:07 I got a Rose thank you larisha for my
5:09 rose
5:10 I got one rose I don't I don't know I've
5:12 this little my third lives I don't know
5:14 how any of this crap works but if you
5:16 want to send me things I guess you can
5:17 send me things
5:19 hi Tim if you're watching uh let's see
5:22 which AI tools am I excited about at the
5:25 moment I learned love to learn more
5:27 about Auto GPT yeah if you go to well
5:31 there's there's lots of videos
5:34 um
5:36 there's lots of videos on YouTube about
5:38 Auto GPT there's one guy in particular I
5:41 can't remember his name hang on auto
5:46 GPT typing in public is always stressful
5:50 Larissa and I talk about this all the
5:52 time because we're always both typing
5:53 live with people watching us make typos
5:57 um let's see which dude is the dude that
5:59 I like see if I can find him
6:02 e
6:03 let's see auto GPT
6:06 tutorial or maybe it was a lang chain
6:09 tutorial
6:10 because uh Lang chain is within Auto
6:13 gpt2 tour y'all
6:18 uh is this the guy that's appy Dave he's
6:21 pretty good
6:23 that's not who I'm thinking of
6:26 is this the guy
6:31 um
6:32 rabbit metrics no
6:37 well this is boring watching me trying
6:39 to find a video
6:42 but uh anyway
6:43 um the tools I'm excited about
6:46 um
6:47 I just saw a thing about mid-journey
6:48 more of a mid Journey guy love me some
6:50 art I um
6:52 I yeah mid Journey for me is the is the
6:55 the top of the
6:57 the top of the Heap when it comes to
6:58 image generation I did um stable
7:00 diffusion for a long time with my my art
7:03 project and I like I like stable
7:05 diffusion but mid journey is just
7:07 consistently really good art
7:10 um it's a bit of a pain in the ass
7:11 because you have to do it through
7:12 Discord but I'm I'm sure within the next
7:14 year they're going to have a web
7:15 interface or an app or something like
7:16 that so mid journey is where it's at for
7:19 me there
7:20 um
7:21 what am I excited about I mean I mean
7:23 you know quite frankly just the more I
7:27 use
7:29 um chat GPT gpt4
7:31 um just any of the large language models
7:33 but gpt4 in particular is you know
7:36 absolutely mind-blowing it continues to
7:39 be mind-blowing for me
7:41 um so so like that's the thing I'm most
7:43 excited about so if you you don't need
7:44 to get more fancy than that I think
7:46 there's also
7:48 um I had a a question on the on the
7:51 channel today
7:52 um that I made a video about about you
7:55 know someone's like where do you find
7:56 chat EPT when I search for it like
7:58 thousands of apps come up there's all
8:00 sorts of crappy apps that um are using
8:03 gpt3 under the hood
8:06 um that are just you know trying to jump
8:07 on the Gold Rush of people talking about
8:09 chat GPT so they're trying to get you to
8:11 you know Fork over 10 bucks for the
8:12 shitty piece of software so just go to
8:15 chat.openai.com
8:18 um or go to bing.com the chat in
8:21 bing.com is actually egpt4 it's based on
8:23 gpt4 and it's connected to the internet
8:25 so both of those are free and you can
8:28 pay for
8:29 um
8:30 chatgpts plus that gives you access to
8:33 to gpt4 which I do but um you get access
8:37 uh did I get access to the plugins for
8:40 for chat EBT yeah I did
8:42 um I haven't played with them a lot
8:44 because I just haven't had time to go
8:47 build one
8:48 um I have started to have chat cheapyt
8:52 um create a tutorial for me to teach me
8:54 how to create a chat GPT plugin I just
8:57 pasted the documentation into chatgpt
8:59 and told it to create a step-by-step
9:01 plan for me so I'm about halfway through
9:03 that
9:04 um they're very very limited right now
9:06 they're an alpha
9:08 um I've heard that they're they're
9:09 rolling out access a bit more
9:12 um but um the the one that I'm excited
9:14 about I don't have access to and that's
9:16 the the one that's connected to the web
9:18 um where you can go off and search the
9:20 web so I think that that that that I'm
9:23 excited about
9:24 um I think the plug-in architecture is
9:27 going to be huge like I think this is
9:28 totally a game changer it's just really
9:29 really early right now so they're very
9:31 raw and and not a lot of interesting
9:34 stuff that I'm actually interested in
9:36 for work uh more of a mid Journey guy I
9:39 love me some art encourage your viewers
9:41 to share your live and invite more
9:43 friends okay hey if you're viewing
9:47 um share this live and invite your
9:48 friends all right thank you thank you
9:51 for oh wait I think that was from I
9:53 think that's just a I think that's the
9:54 robot telling me how to how to do this
9:56 whatever
9:58 um will I will AI takeover software
10:00 engineering jobs yes
10:03 um absolutely if you you are not using
10:06 copilot on GitHub you should but but
10:09 um it's absolutely transforming uh it's
10:13 transforming programming
10:16 um in in lots of lots of ways
10:18 um
10:19 and and you know I think that
10:24 I think the people that have deep
10:26 engineering experience are are
10:28 absolutely you know always going to be
10:30 needed but but I almost feel like it's
10:32 like the lower level tactical stuff is
10:34 just going to get automated out so I
10:36 would say if you're an engineer get good
10:39 at product get good at sort of the big
10:41 picture and strategy
10:43 um I I think that's that's where
10:45 ultimately we're gonna we're gonna live
10:47 I think a lot of the the Tactical
10:49 stuff's gonna get automated out um which
10:51 for for me as a sort of hacky
10:53 non-technical guy that pretends I know
10:56 what I'm doing with software I'm really
10:58 excited about it real Engineers there's
11:00 there's a lot of change coming
11:02 um let's see
11:04 I can pin a comment
11:06 what's that say I can pin a comment oh
11:08 let's just uh Tick Tock telling me what
11:10 to do will AI take over I already read
11:12 that one um
11:18 myself
11:20 have I seen interesting use cases for
11:22 generative AI working with robotic
11:25 process automation no but I haven't been
11:28 looking for it it's not a world that I
11:29 know
11:30 I know creative and I know marketing
11:33 um I I don't know the the automation
11:35 world and Manufacturing and things like
11:37 that
11:38 um but but you know Tesla's got their
11:41 humanoid robot Boston Dynamics been
11:43 doing it for a decade um I saw an
11:45 interview I think it was Lex Friedman
11:47 interviewed the CEO of Boston Dynamics
11:50 and
11:52 um he said that there are a lot of
11:54 robotic startups right now so I think I
11:56 think large language models and robots
11:59 are gonna
12:00 do this very very quickly and I think I
12:02 think we're going to see advances
12:04 um I think we're going to see pretty
12:06 dramatic advances in uh
12:09 in robotics so yeah
12:12 uh with the strike coming any thoughts
12:14 of the future of screenwriting yeah I
12:16 I've written seven screenplays I know I
12:18 know this world pretty well
12:20 um I think that the
12:22 um
12:24 the uh the writer's Guild just did
12:27 something about trying to block
12:31 um any GPT generated stuff I think
12:36 that's very very short-sighted
12:38 um
12:39 GPT is going to be built into every
12:41 single writing tool out there if it's
12:43 not already it's in notion it's going to
12:45 be an office 365. it's it's likely going
12:48 to be in the
12:49 in the in the um in screenwriting
12:52 software so I think I think just even
12:54 being able to figure out where the
12:56 boundary is between you know what lines
12:58 were created with GPT I just think it's
13:00 I think it's a pointless exercise right
13:02 now
13:03 um
13:06 you know
13:09 we we
13:11 here's the here's the problem with with
13:13 any of the creative work creative arts
13:16 and any of any of knowledge work
13:19 is that we are moving from a relative
13:22 world of scarcity when it comes to
13:26 creative output and knowledge output
13:29 um to a world of abundance that that
13:32 there's just going to be tremendous
13:34 amounts of content and whether it's
13:36 completely
13:38 you know autonomously generated or
13:41 completely human or some hybrid in
13:43 between we're we're going to be able to
13:46 tell less and less that there's that
13:48 boundary there right now because chat
13:51 GPT is the separate tool
13:53 it's you know I I feel like we're in a
13:56 very very short-term you know
13:58 block the block the robots the robots
14:01 are coming for our jobs
14:03 I I don't feel that at all I feel that
14:05 these tools are
14:07 um Talent amplifiers right so if you're
14:10 really good at screenwriting if you're
14:12 really good at storytelling
14:14 and you learn to use AI tools you're
14:17 going to be really good
14:18 um you're going to have your talent
14:20 Amplified using that tool so I think I
14:22 think the the quality
14:24 um and um
14:26 and quantity of content is going to go
14:29 up
14:30 it does make it a lot easier for you
14:33 know completely non-talented people to
14:35 get garbage into the marketplace so I
14:38 think it's going to get noisier but you
14:40 see this with with art you see this with
14:42 mid-journey and with image generation
14:44 tools that
14:46 um really talented people that learn to
14:48 use these tools tend to make way more
14:52 interesting work than less talented
14:54 people right
14:56 um so uh you know and I actually think
14:58 writers you know anyone who's got the
15:00 ability to synthesize ideas generate
15:03 ideas are going to be really good at
15:05 telling these tools
15:06 um how to give them what they want
15:10 um ask any questions you want what are
15:12 my macros I don't have any macros I
15:14 don't really use any macros are you a
15:16 software engineer no I'm a CEO of a
15:18 software company I would call myself
15:19 more of a product guy than well I'm
15:22 definitely not a software engineer um
15:23 I've got an acting degree and a writing
15:26 background I started one of the early
15:28 digital agencies so I did a lot of early
15:30 HTML work but that's not really
15:32 engineering the the we had Engineers
15:34 that did the actual engineering and I've
15:36 got a tech team that that develops my
15:39 current software though I am the one
15:41 that um is the product guy right so I
15:44 have the idea I know what I want it to
15:46 do and I know when it sucks and I I
15:49 think I'm a pain in the ass product guy
15:51 to my Dev team because you know my
15:54 definition of it should work is it
15:56 should work like it's you know it's that
15:58 simple if it doesn't work then it sucks
16:00 um I talk about that a lot
16:04 um and ask any questions you've got
16:06 about AI I'm happy to talk about them
16:09 and what else I'm trying to think what
16:12 else is going on I haven't seen a ton of
16:14 news over the weekend because I was busy
16:17 kind of recuperating from a a last
16:19 minute trip to New York to go out to see
16:22 a client who got very excited about the
16:24 new AI stuff we rolled into my company
16:26 software my company is called storyvine
16:29 and it's got
16:30 um what do we have we have an automated
16:32 uh video storytelling platform so people
16:35 answer questions in an app it goes up to
16:37 the cloud a few minutes later you've got
16:38 a fully edited video and so we are
16:41 rolling a lot of AI functionality into
16:44 that and and actually I'm completely
16:46 rethinking the platform and what it is
16:49 and what it can be and what what it can
16:51 do
16:52 and it seems to be resonating people
16:54 seem to be getting very excited about it
16:55 so that's cool
16:57 um let's see any opportunities to govern
17:00 AI
17:02 I think
17:04 I think that governing AI is is I I
17:08 think there's going to be massive
17:09 Industries built around it I think it's
17:11 absolutely required I think we're
17:13 entering a massive [ __ ] show where where
17:17 governance uh you know open AI is sort
17:21 of trying to govern it and and
17:24 sort of slow roll their releases but you
17:27 know last week or the week before four
17:30 different open source large language
17:32 models launched and
17:33 metas alpaca or metas llama
17:38 um models all got leaked so so those are
17:41 out in the wild and alpacas based on
17:43 llama and so there's a bunch of it's
17:45 it's the it's absolutely a wild west
17:47 right now
17:49 um the the one company in this space
17:51 that I I think is really interesting is
17:54 a bunch of X open AI Engineers that made
17:58 um GPT
17:59 three
18:01 um started a company called anthropic an
18:03 anthropic has what's called a
18:05 constitutional learning model which so
18:08 so the way open AI works is they've got
18:10 this large language model and then they
18:11 use human reinforcement learning so
18:14 humans look at an answer and they say
18:17 thumbs up or thumbs down and then the
18:20 machine learns which answer was better
18:22 right and if you do that over tens of
18:24 thousands of people it gets really smart
18:26 really quick it's one of the things they
18:28 found is you don't need to train these
18:29 large language models a lot before they
18:31 get dramatically better
18:33 um so that's opening eyes thing um the
18:35 the challenge with human reinforcement
18:37 learning is that it's very subjective
18:39 and all the biases human biases
18:41 um get built built into the into the
18:44 reinforcement so so you can reinforce
18:46 bad behavior
18:47 anthropic
18:49 their model is called the Constitutional
18:51 learning model and and the way their
18:53 system works is they write a
18:54 constitution so they write a set of
18:56 rules so first do no evil and you know
18:59 do good for Humanity and whatever
19:00 whatever the rules are and if you go to
19:02 the anthropic site I think they've got
19:04 six or seven of them up there
19:06 um and then and then the way the system
19:08 works is you ask it something it
19:11 generates an answer and then it Compares
19:12 that answer against the Constitution and
19:15 if it if it ranks poorly it'll send it
19:17 back
19:18 um and say no that violates something in
19:20 the Constitution and it'll it'll keep
19:22 iterating until it passes the test and
19:24 then it'll give you the answer
19:26 um I think I think that approach is is
19:27 really really interesting
19:29 um you know it's it's probably not
19:32 perfect right now I think what we'll see
19:33 in the future is we'll see
19:35 um
19:37 there's not going to be a single
19:39 governance model that works for everyone
19:42 so I think what you'll see is you'll
19:43 have lots of different models and you'll
19:45 be able to choose from uh here's a model
19:47 that aligns with my you know values and
19:50 I'm going to use that one and here's one
19:51 that doesn't align with my values and
19:53 I'm not going to use that one I think I
19:54 think that's how we'll start to see
19:56 these things
19:57 the corporate ones are going to be very
19:58 boring and then there's going to be wild
20:00 and zany ones and then there's going to
20:02 be ones that are actually funny and ones
20:03 that are actually smart and interesting
20:05 and ones that are actually dangerous and
20:07 you know things like that all right let
20:09 me see what other questions are here
20:11 um what are the best wait sorry lost it
20:16 what are the best AI software for images
20:18 that
20:20 progress into other images like if you
20:22 can mix images
20:24 um mid-journey right now does really
20:26 good
20:27 um mixing of images
20:29 and
20:31 stable diffusion with dream Booth you
20:34 can train it with your models there's
20:35 also a piece of software called
20:37 leonardo.ai
20:39 um that I really like a lot I think it's
20:41 still in beta testing I have access to
20:43 the beta
20:44 but let me see here the oh Nardo
20:51 um
20:51 this one's really good
20:55 login
21:01 watch as I log in to a piece of software
21:04 don't you hate it when people try to
21:06 show you something in Tech and they're
21:08 like oh it's going to be really good
21:10 and then it takes them forever to log in
21:12 like it is for me right now
21:14 let's see
21:16 AI image generation so this is one
21:19 that um oh yeah that's some disturbing
21:21 stuff I did let's see
21:25 um
21:27 70s muscle here I'll flip just like this
21:31 70s muscle
21:35 car in abandoned
21:40 Factory let's see how this does
21:52 [Music]
21:54 yeah that's pretty good
21:56 yeah so
21:58 so Leonardo is pretty good and you can
22:00 uh oh the reason the person asked was
22:04 um
22:05 how do I go back
22:07 was uh
22:11 you can upload an image so right here
22:14 there's image to image so you can upload
22:15 an image and then base something that
22:18 you do on that image so
22:20 um so most of the most of the image
22:22 generation things are going to have
22:25 um
22:26 are going to have the ability to mix
22:27 images c x a i is what I have no idea if
22:32 anyone knows let me know thoughts on
22:33 auto GPT
22:35 yeah Auto GPT
22:37 agent GPT baby AGI all of these
22:41 autonomous agents are going to be
22:45 huge they're also all
22:48 um
22:49 incredibly new and Incredibly raw and
22:53 um
22:55 there's kind of good news and bad news
22:57 right now the good news is because of
22:59 the popularity of chat GPT the
23:02 development Community is bouncing on
23:05 anything AI so they're developing
23:07 all sorts of new Innovative things these
23:11 autonomous agents are one of them
23:13 they're essentially a month old so even
23:17 though they can do really remarkable
23:18 things so so what they are if you don't
23:20 know what an autonomous agent is is
23:22 rather than you having to prompt a large
23:24 language model and say give me this
23:26 specific answer and then it'll give you
23:27 that
23:28 you basically just give it a goal you
23:31 say I want you to go find the taglines
23:33 of of my 10 biggest competitors and it
23:36 will create a list of tasks and prompt
23:40 itself it will send its own prompts to
23:43 either a large language model like like
23:45 chat gbt or it will go to a Google
23:47 search or whatever it needs to do and so
23:50 it will say okay you know who are the
23:52 top 10 competitors for storyvine and
23:55 it'll go out and find them and then
23:56 it'll go find their tag lines and then
23:59 it'll you know build that into a table
24:01 using a large language model
24:04 um they can obviously do stuff that's
24:05 way more sophisticated than that
24:07 um so they're quite quite powerful and
24:09 so people are using them to do things
24:11 like create website Builders where you
24:14 literally just say make me a website for
24:15 this kind of business and it will go and
24:18 it will write the copy it'll generate
24:19 the code it'll go generate images for
24:22 the design and package it all up and
24:25 launch it and you know potentially even
24:27 go grab you a domain I you know it'll do
24:30 all sorts they do all sorts of crazy
24:32 things because they're so raw and so new
24:36 they're really janky they're really just
24:39 bad software because they're brand new
24:43 um but but what what it's doing is it's
24:46 stitching together a series of tools in
24:48 a way that you don't have to prompt it
24:50 you just give it a goal and turn it
24:52 loose right so that's terrifying also
24:54 because you know imagine
24:58 um
24:58 if if you wanted to say you know go find
25:01 me a bunch of servers that have a
25:03 particular
25:04 um
25:05 you know hack exploit and go exploit
25:08 them and then send me all the URLs where
25:10 I have root access
25:12 you know they'll just go off and do that
25:14 so
25:16 um but they're going to be amazing
25:17 they're going to be amazing there's
25:18 another thing that came out last week
25:20 called generative agents or the week
25:21 before
25:23 and generative agents the the paper
25:25 talks about like the little NPC players
25:27 the non-player characters and games and
25:30 these NPC characters use large language
25:33 models to actually talk to one another
25:35 some sort of you know autonomous agent
25:37 where you'll will each have our own
25:38 agents and we'll just tell the agent go
25:41 do some stuff for me and it will go do
25:42 it and I could see a world where you
25:45 know I'll have my agent talk to your
25:47 agent and our agents will go off and you
25:49 know coordinate things and figure out if
25:51 we've got interest or availability or
25:54 whatever it might be and just do stuff
25:55 for us
25:56 um and they'll get to know our
25:57 preferences and it's it's going to get
25:59 very very interesting thoughts about
26:01 copyright and AI every image is unique
26:03 and and
26:05 irreproducible if you don't know the
26:07 seed yeah so
26:10 um
26:11 yeah
26:12 the copyright and trademark issue the
26:15 the the only thing that I can predict
26:17 right now is that the lawyers are going
26:19 to make a [ __ ] fortune in the next 20
26:21 years that's just gonna make a fortune
26:24 um the lawyers are really excited about
26:26 copyright and trademark right now
26:28 because it's such a mess
26:31 um
26:32 it it's I it's gonna get better you've
26:35 got companies like like Adobe jumped in
26:37 into the image generation thing recently
26:40 with firefly and and you know they
26:43 trained their model on open source
26:45 public donate a public domain images
26:49 um and so the idea with with Adobe is
26:53 that they're going to make it friendly
26:54 for artists where artists can opt out of
26:56 the training set they can opt into the
26:58 training set they're probably going to
26:59 have compensation models I think that
27:01 we're going to see compensation models
27:04 um where if like I'm a photographer so I
27:06 could potentially generate
27:08 um uh you know training sets for my
27:11 photographs and then officially license
27:12 them so if someone liked my photographs
27:14 and they want to generate images that
27:16 look like my sky photos they could do
27:18 that and I would get paid on the
27:20 generation of the image rather than the
27:21 usage right now all the copyright stuff
27:23 lives on the usage side and I think
27:25 that's going to be a train wreck because
27:27 of like all the synthesized music we see
27:29 coming and all the synthesized images we
27:31 see coming I just think it's going to be
27:33 a morass for a lot of years
27:36 but but I think I think probably within
27:39 the next two to three years
27:40 um there'll be systems out there where
27:42 where attribution and compensation are
27:45 going to be handled
27:46 um I also
27:49 you know
27:51 copyright
27:53 I mean this is all it's already not what
27:57 it was intended to be I mean copyright
27:59 was intended to protect people when
28:00 producing work
28:02 was really expensive and really slow and
28:04 required an incredible amount of
28:07 technical expertise to actually make it
28:10 happen
28:11 um that got very much watered down when
28:14 we moved into a digital world when we
28:16 went from analog to digital and then it
28:18 got more watered down when we when we
28:20 moved to the internet where you could
28:22 just immediately distribute things and
28:25 now I think I think kind of the the the
28:27 nail in the coffin to to copyright being
28:30 really relevant
28:32 um is is
28:33 um this AI stuff I think it's just gonna
28:35 it's gonna blow it out of the water like
28:37 I don't know what it looks like I
28:39 understand that people need to get paid
28:40 I understand that you you might want to
28:43 protect something really amazing that
28:45 you put together
28:46 um I just don't know how you protect it
28:48 I don't know how you're protected in a
28:50 world where where
28:51 um you can generate anything instantly
28:53 if you look at some of the songs that
28:54 have been put out like the the one that
28:57 was put out that synthesized Drake and
28:59 the weekend their voice and
29:02 um it was an AI synthesized song and it
29:04 got like you know 600 000 or a million
29:07 or however many
29:08 views on Spotify a lot before they took
29:11 it down I mean
29:12 you know we're gonna start seeing fan
29:14 generated art that Rivals the original
29:16 artist
29:18 um and they don't care if they get paid
29:19 for that they're just doing they're just
29:21 putting stuff out there so I don't know
29:22 how to compete in that world so
29:25 um
29:25 so yeah my thoughts are the lawyers are
29:27 gonna get rich and it will get figured
29:29 out but it's going to be an absolute
29:31 mess for 20 years
29:33 um
29:34 what are some of the predictable
29:35 industries that may come as a result of
29:38 AI technology
29:40 um
29:44 I think we're gonna see a lot of
29:48 incredibly powerful companies that
29:52 historically would have taken
29:54 a hundred million or 200 million or 300
29:57 million dollars to Launch
30:00 will get launched by two or three people
30:03 you know
30:05 over a couple of weeks I think
30:08 incredibly sophisticated disruptive
30:10 companies are are just gonna like spring
30:13 up out of nowhere I I also think there's
30:16 there's a lot of talk right now about AI
30:18 is going to take jobs and there's going
30:19 to be all this disruption in the in the
30:21 workforce I absolutely believe that that
30:22 to be the case I think I think it's
30:24 going to be a bloodbath but
30:28 a lot of those displaced people also
30:31 have access to the tools that displace
30:32 them right
30:34 so what you're going to see is there's
30:36 going to be a lot of companies that
30:37 Embrace Ai and do it right there's going
30:39 to be a lot of companies that resist Ai
30:41 and they're going to get hit like a
30:43 tsunami
30:44 with companies that just come in and and
30:46 absolutely obliterate them I think a lot
30:49 of those companies that come in and
30:51 obliterate the incumbents are going to
30:53 be started by people that got displaced
30:55 out of those those bigger companies I
30:56 think I think the agency holding
30:58 companies are in real trouble
31:01 um not not because you know well they're
31:03 in trouble for a lot of reasons but but
31:06 they they can do good work
31:09 um but I think how they charge and the
31:12 amount of time they take
31:14 um I don't think I don't think they're
31:15 gonna last that long I think they're
31:17 gonna get innovated out really quick if
31:19 they don't if they don't disrupt
31:20 themselves
31:22 um
31:23 what are some of the predictable
31:24 Industries um I think there's going to
31:26 be a lot of Industries around AI safety
31:29 I think there's going to be a lot of
31:30 Industries around identity and
31:33 authenticity
31:35 um so for example my company storyvine
31:37 we we have a technology that allows
31:39 people to share their stories right so
31:41 most of the content produced on our
31:44 platform is authentic human beings
31:45 sharing their stories
31:47 um the the synthesized human thing right
31:51 now you can still tell they're
31:52 synthesized but I would say give it a
31:54 year and it's going to be tough to tell
31:56 if someone is real or not
31:58 um and so I could very easily see us
32:01 doing you know sort of an authenticated
32:03 maybe it's blockchain verified that this
32:06 is a real human and I think authenticity
32:09 is going to go up so I think identity
32:11 and authenticity there's going to be
32:12 lots of stuff there I think security is
32:15 going to be a massive industry that
32:18 comes out of this because the hackers
32:20 have way more powerful tools as do the
32:23 the white hat hackers right the the ones
32:25 battling them so I think those battles
32:27 are going to get really interesting
32:29 um
32:32 boy I I think
32:35 any knowledge work I think consultancies
32:37 are going to face real struggle and real
32:40 change any knowledge work I think any
32:42 knowledge work I think any of the
32:44 creative Industries I think Hollywood's
32:46 gonna get upended and and they're
32:48 fighting it like that I was talking
32:49 before someone asked about the
32:50 screenwriters Guild and the writer's
32:52 thing and they're trying to block the
32:53 use of GPT which I think is really
32:55 short-sighted but
32:57 um I I think I think the creative arts
33:00 and the knowledge
33:02 you know any businesses that rely on on
33:04 knowledge workers are going to be
33:06 seriously upended
33:09 um Blue Willow yeah someone asked about
33:10 uh image Technologies blue Willows one
33:14 that's it's a Discord based thing like
33:15 mid Journey but it's free I think it's
33:17 still free
33:19 um yeah mine has been getting stuck in a
33:21 do do nothing Loop
33:24 um not too consistent that's I assume
33:26 that's in reference to Auto GPT like I
33:28 said it's very very early there's not a
33:29 lot of backstops in there if it gets
33:32 into a bad Loop uh it just loops and
33:35 they they give you very bad answers I've
33:37 had
33:38 I've tried to do I don't know I've maybe
33:41 messed around with it 20 times and I
33:43 think I've had two successful
33:45 completions so two out of 20 10 percent
33:48 it's probably lower than that
33:51 um how do I feel wait sorry things are
33:53 scrolling too fast told
33:55 how do I feel about data security in an
33:57 Enterprise environment well right now
33:59 well so uh two days ago
34:02 um chat GPT announced that you can turn
34:05 off
34:06 um
34:07 sharing of your data and your history
34:08 for their training purposes so that's
34:12 good you will lose your history if you
34:14 do that though so make sure you export
34:15 your data so you have your history if
34:17 you've got any good good prompts in
34:18 there
34:20 um
34:22 openai has also announced I assume
34:25 sitting on Microsoft Azure that that you
34:27 can do Standalone servers
34:29 um so I think I think Enterprise
34:31 Securities major we've got
34:34 um Pharma clients are our primary
34:35 clients at storyvine so this is a really
34:38 big concern especially as we're talking
34:40 about adding generative AI into our
34:42 platform
34:43 um so so we will likely do Standalone
34:46 servers that are that are effectively
34:48 hermetically sealed probably for each
34:50 client but we might just end up doing a
34:52 standalone
34:53 uh thing for our company and then just
34:55 uh you know
34:57 assure security um but but I think I
34:59 think data security in an Enterprise
35:01 environment's huge
35:04 huge Samsung just got a bunch of stuff
35:06 leaked because
35:09 um people put meeting notes and some
35:11 proprietary code in a chat GPT chat
35:14 someone hacked their history and got a
35:16 bunch of that stuff so someone got fired
35:18 there so I think it's huge
35:20 um what about AI content creator
35:22 accounts we probably won't need real
35:24 people to create content anymore well
35:27 to create the content no but to
35:32 instantiate the content to come up with
35:34 the idea to come up with a strategy to
35:37 for every image that you see that
35:39 someone shared on Twitter there's
35:40 probably 50 images they created that
35:42 weren't good enough to make the cut so
35:44 curating content refining the content
35:47 you know you know combining different
35:49 images together either in Photoshop or
35:51 illustrator or whatever it might be
35:53 adding type to it
35:55 um
35:56 my experience with these tools you know
35:59 on the surface you say oh these tools
36:01 these tools can do creative work really
36:03 well
36:04 they're going to replace all the
36:05 creative people my experience has been
36:08 I'm finding myself more creatively alive
36:11 than
36:13 um
36:14 then I've been in a long long time
36:16 because
36:18 what the tools do is they do the low
36:20 level creative work that always got in
36:22 my way before so I would have an idea
36:24 and I'd say oh I want to go do this
36:26 creative idea and then I'd say but I
36:28 need a piece of software that does that
36:29 and then I'd Google that and then I'd
36:31 find
36:32 you know two pieces of shitty software
36:34 that were just ads and then I'd finally
36:37 find a piece of software and I'd go to
36:38 install it and it wouldn't install right
36:40 then it would be buggy and then I'd go
36:41 to YouTube try to figure out how to use
36:43 it and 45 minutes later I don't even
36:44 know what the idea was anymore
36:46 what's happening with AI is I have an
36:48 idea and then I can articulate it with
36:50 chat GPT and I can visualize it with
36:53 mid-journey or whatever tool I'm using
36:55 like like that and and that gives me new
36:58 ideas and so I I'm finding myself being
37:00 way more creatively alive than I was uh
37:03 before it so so I I think I think we're
37:06 actually entering a great Renaissance I
37:09 think we're we're entering an age
37:11 of incredible creative expression but it
37:15 is going to be done through these tools
37:17 it's going to be creative people using
37:18 these super powers to create truly
37:22 remarkable stuff so that's that's my my
37:24 two cents on the creative thing
37:27 um building fintech and want to use Ai
37:29 and ml to prevent fraud and money
37:32 laundering any advice uh yes go do that
37:34 start testing it I mean the thing about
37:38 um
37:38 large language models that
37:41 the the the the
37:42 biggest shift that my sorry
37:46 inarticulate
37:48 um
37:51 the big difference is historically if
37:54 you wanted to do something with you know
37:57 fraud and money laundering prevention
38:00 you would have to explicitly program the
38:03 logic to say you know here are all the
38:05 kinds of money laundering schemes we
38:08 might see and hear all the you know the
38:11 fraud things we might see and let's
38:12 write all these exceptions and you have
38:15 to explicitly write the logic and what
38:17 happens with these large language models
38:19 is you can say act like a you know money
38:22 laundering you know fed
38:26 um who's expert at identifying money
38:29 laundering schemes and go find the
38:31 schemes and now give me 20 ideas for how
38:33 to prevent those schemes now write the
38:35 code to find the schemes prevent them
38:38 and send me a report daily and by the
38:42 way I'm you know I'm gonna run a Cron
38:45 job that we're gonna we're gonna we're
38:47 gonna update that thing every 24 hours
38:50 right so I think your ability to iterate
38:54 um
38:55 features within the company you're
38:57 trying to create
39:00 accelerate dramatically
39:03 your fraudsters and your money
39:06 launderers also have access to these
39:08 tools so they're going to come up with
39:09 increasingly complicated schemes so so
39:12 you know you'll have your fraud
39:14 prevention agent talking to their fraud
39:16 agent and and uh you know they'll be a
39:19 constant War but I think you can iterate
39:22 very very quickly so my advice would be
39:24 just start playing start playing with
39:26 the large language models
39:27 um really explore
39:30 navigating between conceptual things
39:33 like what are the kinds of schemes that
39:36 I might find out there and then how do I
39:38 code against those and let the tools
39:40 help you come up with those and give
39:43 those to your engineers sorry late to
39:45 the party can I tell us can you tell us
39:47 what your background is how you got to
39:49 where you are today yes I have a degree
39:51 in acting
39:52 I moved to New York New York City out of
39:55 school started and ran a theater company
39:57 for four and a half years uh wrote seven
40:00 screenplays in a one-man show after that
40:02 and then in the mid 90s
40:05 um I was doing desktop publishing as my
40:07 day job and
40:09 um I stumbled on this thing called the
40:11 world wide web and I thought Hmm this
40:13 thing's interesting and I had an idea
40:15 for an online magazine and started one
40:18 of the first digital magazines a Zine
40:20 called Urban desires and
40:23 um and then from that met a guy that I
40:26 ended up co-founding a company called
40:28 agency.com with which was one of the
40:30 first digital agencies
40:32 and so built a lot of the first websites
40:34 for the Fortune 500 so one of the
40:36 reasons I started my Tick Tock Channel
40:38 um at the ripe old age of 57 that I am
40:41 uh
40:43 last thing on the planet I thought I'd
40:45 do start a tick tock Channel
40:47 um
40:48 but I think that what's happening with
40:50 AI is akin to to
40:52 [Music]
40:53 um
40:54 what I discovered with the World Wide
40:56 Web that that it's going to change
40:58 everything and I felt that in 94 when I
41:01 discovered the web
41:02 and I feel that way more profoundly now
41:05 I think that I think the changes that
41:07 are coming with AI are
41:09 um significantly
41:10 um more profound and they're coming way
41:13 way way way way faster
41:15 um than any of the any of the internet
41:18 stuff that that happened over the past
41:20 30 years
41:21 um the past decade I've been founder and
41:24 CEO of a company called storyvine which
41:26 is an automated video storytelling
41:28 platform and then I would say the last
41:30 year I've been pretty active in AI stuff
41:34 and in the past six months I've just
41:37 been absolutely obsessed with it I
41:39 cannot
41:40 um I cannot uh I can't get my head
41:43 around how fast it's moving and how
41:45 profound this stuff is which is why I'm
41:47 doing this account and even when I'm
41:49 doing these lives I don't know why I'm
41:50 doing a live
41:52 um let's see AI to grammarly
41:55 um to remove AI detection yeah again I
41:59 think all this AI detection [ __ ] is
42:01 going to be [ __ ] these large
42:02 language models are gonna literally be
42:04 everywhere I think this is a very short
42:06 window where everyone's panicking and
42:07 they have their panties and a Twist
42:09 about yeah it was written by a robot
42:11 it's not real writing and there's some
42:14 truth to that but these tools are just
42:16 going to be like having a spell check or
42:18 having grammarly you know plug-in in
42:21 your writing where it checks your you
42:23 know these things that check your
42:24 spelling and check your um your grammar
42:26 Rai right so the generative AI stuff is
42:29 just a slight variation on the theme
42:31 um these tools are going to be be
42:33 everywhere
42:34 um so you will likely always be able to
42:37 find a tool that will be able to skirt
42:39 around AI detection but I don't think I
42:43 don't think AI detection you know
42:45 whether or not something was generated
42:46 with AI is going to be all that relevant
42:48 and issue a year from now I just don't I
42:51 you know I think this is going to be
42:52 ubiquitous this is it's going to be just
42:54 like having a word processor or a
42:56 calculator right back when calculators
42:59 were invented everyone freaked out about
43:01 no one's going to need to know math
43:02 anymore because I got these fancy
43:03 machines what are we gonna do
43:06 so
43:08 all right uh image generation and AI
43:11 takes a lot more skill if you're trying
43:12 to create something specific exactly so
43:15 so the idea that it's one of the one of
43:18 the my pet peeves is that people talk
43:20 about these machines like they're just
43:22 behaving on their own Auto GPT will just
43:25 behave on its own so I guess I should be
43:27 a little careful where I tread here
43:29 but but
43:30 um
43:32 a creative act is an intentional act
43:34 right you intend to create something
43:36 specific you know for a project or or
43:40 for a goal or
43:41 I mean you might just be creating stuff
43:43 for for fun and that's fine
43:46 um but yeah those people that can say
43:48 hey I want an mg speeding down a road
43:50 you know in the in the highlands of
43:53 Scotland
43:55 um and they can turn that around quickly
43:57 those people are going to be more
43:58 valuable so absolutely I think there's
44:00 absolutely skill
44:02 um you see it all the time you see
44:04 people put stuff on Twitter that's just
44:06 all the same and then you see someone
44:07 else that does something you're like oh
44:08 my god well that probably took them
44:11 you know a month of sort of you know
44:14 thousands of images to kind of get to a
44:16 look and feel and figure out how to do
44:18 that I know I know someone that I that I
44:20 work with that is brilliant at this
44:23 stuff it took him a month or two to
44:25 figure it out
44:27 um what are the top three threats to AI
44:30 um an EMF would be a big one if all of
44:32 our computers stopped working it would
44:33 shut it down pretty quick
44:36 um
44:39 I would have said
44:41 you know prior to um
44:44 prior to chat GPT I would have said
44:46 something about government intervention
44:48 or or monopolization
44:51 but I think I think the fact that chat
44:53 CPT went to 100 million users in a month
44:56 in in a month and a half
44:59 um has forced a lot of companies out of
45:01 the woodwork and it's forced a lot of
45:03 Technology into open source and even you
45:06 know stability just released their their
45:08 large language model open source and
45:11 their image model is open source I have
45:12 a feeling stability is doing that
45:14 because
45:15 they probably have an inkling they're
45:17 getting the pants suit off them by Getty
45:20 um for their for their training models
45:21 so if they don't settle that that could
45:23 put them out of business so I have a
45:25 feeling companies like stability
45:27 understand that
45:30 that they might not survive litigation
45:34 and so I think they're putting things
45:36 out in the open source to get them out
45:37 there and you know the fact that meta's
45:40 large language model leaked and is out
45:42 in out in the open now it's in the
45:44 development Community
45:45 um the fact that there were four large
45:47 language models
45:49 um released last week the fact that some
45:51 of these image models are open source
45:54 um I don't think that governments are
45:56 going to be able to get their hand
45:57 around it
45:59 their hands really around this stuff
46:02 um it's possible but I don't see that as
46:04 a threat I see a bigger threat being
46:06 government incompetence
46:09 um kind of regulating the wrong things
46:12 um and and you know just doing stupid
46:15 [ __ ] like you know you can't use chat
46:17 tpd to write
46:19 um
46:20 okay
46:21 what are some of the other threats I
46:23 mean
46:27 if
46:30 there's enough of a negative kind of
46:32 panic out there because of Hollywood
46:34 that you know AI is gonna
46:37 destroy Humanity
46:40 that you know if someone takes one of
46:44 these Auto GPT things and kind of turns
46:46 it loose in the right way and someone
46:47 ends up you know dying or you know
46:49 people end up dying because of it I
46:51 could see there being some kind of panic
46:53 where where a lot of this stuff gets
46:56 locked down temporarily but I just I
46:58 don't think it's gonna stick so
47:01 um
47:07 I don't know I don't know I
47:11 where where I'm trying to live is in the
47:14 place that
47:15 any individual who gets curious and
47:19 explores AI right now is going to have a
47:22 disproportionately positive impact on AI
47:25 moving forward and so I'm trying to get
47:28 as many people up to speed and
47:30 enthusiastic about trying it and at
47:32 least understanding it right now is
47:34 possible because I think it's going to
47:35 precipitate lots and lots of change and
47:37 if you're on the inside of it using it
47:40 um
47:40 you can help precipitate that change in
47:42 a positive way rather than just a
47:44 destructive way right if we just leave
47:46 it up to the capitalists they're just
47:48 going to cut staff there's another model
47:51 that says hey with the same staff we can
47:53 do way more work or way better work with
47:56 the same amount of staff so there's
47:57 there's lots of different ways to
47:59 approach these tools so I don't know if
48:01 that answered your question
48:04 um
48:05 you articulated Tech frustration for
48:07 creatives perfectly by the time I
48:09 figured out the ideas gone yeah exactly
48:11 exactly that thing where you're like
48:13 you've got an idea and then you're just
48:14 trying to like just jump the hurdles to
48:17 be able to even get to a like The
48:19 Whisper of an articulation of that idea
48:21 and then it's gone so yeah absolutely
48:24 that's that's where I mean these these
48:27 generative AI tools are just a [ __ ]
48:29 superpower do I know Howard cash I do
48:32 not
48:35 um
48:35 there's uh it happens don't take it okay
48:38 hi cheers hello
48:41 um age is just a number thank you mine's
48:43 57.
48:45 it's a big one
48:47 [Music]
48:48 um thanks for your content interesting
48:50 background thank you
48:53 thanks to Mom and Dad oh I guess that's
48:55 your your name all right okay it just
48:58 I'm gonna love it let's see what else
49:01 good point about stupid government
49:03 Regulators yeah I mean like if you watch
49:05 those Tick Tock things that like the the
49:07 the Congress people didn't even know how
49:10 Wi-Fi worked much less how Tick Tock
49:12 worked much less how incredibly
49:15 transformative powerful AI tools work
49:17 like
49:19 it's [ __ ] terrifying we did um the
49:22 everyday AI newsletter that I do every
49:24 week we did it one of the
49:27 um one of the topics this week was
49:30 um this military tool where they're
49:32 using large language models to do
49:34 military logistics and it was just
49:36 absolutely terrifying but I'm sure the
49:38 military's got you know AI tools that
49:42 make what we have look comically
49:45 um simple let's see they're a very
49:47 knowledgeable AI developers all of a
49:49 sudden
49:50 um where have they been all this time
49:51 well I mean so so so listen when the
49:55 oops
49:56 when the uh when the World Wide Web
49:58 first came onto the scene the internet
50:00 had been around for a number of decades
50:02 same with AI ai's been around for a
50:05 bunch of decades
50:06 um
50:08 a couple of things have happened so so
50:10 when gpt3 came out
50:12 um in the documentation of gpt3 were the
50:15 instructions for how to build chat GPT
50:17 and Sam Walton assumed that someone in
50:20 the development company would build it
50:22 and they didn't and then
50:24 um Microsoft was about to announce that
50:26 they were going to make this billion
50:27 dollar investment into open Ai and Sam
50:30 Altman said to his Dev team you've got
50:31 two weeks put together chat GPT so they
50:34 built that in two weeks so if you feel
50:36 like chat GPT is kind of like
50:39 uh like a thin piece of software it's
50:42 because it is and you know Sam Walton
50:43 has said it's it's a bad piece of
50:45 software
50:46 um but but
50:48 it got to a hundred million users in a
50:52 month and a half
50:53 and so what that's meant is that the
50:56 entire development Community has just
50:59 flocked to
51:01 um this generative AI stuff right so so
51:03 the tech had been there for a long time
51:05 and a lot of developers have been
51:07 developing things with gpt3 and and you
51:10 know using a lot of the tools that are
51:12 sitting underneath it for for like a
51:14 year and a half before before chat GPT
51:16 came out
51:17 um but but because it got so popular
51:21 that's the thing that's generated all
51:23 this Buzz that's the thing that's forced
51:24 Google out of the shadows and and sort
51:27 of force them to put I mean they're the
51:28 ones that invented the Transformer that
51:30 all of these generative Technologies are
51:32 based on
51:34 um and they were sort of keeping it
51:35 behind the firewall and they were sort
51:36 of forced out in the open um AWS has now
51:39 been forced into the game Apple's got
51:41 something uh some announcement coming in
51:43 June that I assume is going to have some
51:44 AI stuff with it
51:46 um you know meta has been forced out of
51:48 the out of the Shadows with this stuff
51:51 um so so I think that that the developer
51:53 Community
51:55 um is is
51:57 um is just pouncing all over it the
51:59 other thing about these large language
52:00 models is
52:01 with a hundred lines of code you can
52:04 generate applications that are
52:06 incredibly sophisticated right so so I
52:09 think you've got even Junior developers
52:11 um are able to whip whip up some pretty
52:14 remarkable stuff very very quickly so
52:16 that's where I think they're I mean you
52:17 know part of it is just there's there's
52:19 all this buzz and fervor but but it's
52:21 because the tools are that good and and
52:24 uh there was all this you know
52:27 visibility because of chat GPT
52:30 um how is it possible to technically
52:31 lock it to lock what to to lock
52:35 um large language models well you could
52:37 I I you can lock it if you run a large
52:40 language model locally on a machine
52:41 that's not connected to the internet you
52:43 can now do that with like with alpaca
52:45 and and some of the things that are
52:46 built on oops my machine went to sleep
52:48 uh alpaca and some of the some of the
52:51 things that are based on
52:53 um llama and and some of these other
52:55 open large language models they're
52:57 they're getting them I forget what they
52:59 call the process what's it called
53:01 um
53:02 I don't know where they shrink it down
53:04 not miniaturization but quantization
53:08 quantumization I don't know
53:10 um
53:11 I heard of one large language model that
53:14 um they've got down to a five gigabyte
53:16 install so a five gigabyte install you
53:19 can just put that on a laptop and run it
53:20 so if you're just running a large
53:21 language model locally there's no
53:23 there's no real risk you're not sending
53:25 your data out to anyone so if that's
53:27 what you mean lock it down you know you
53:29 run it on a virtual machine or you run
53:31 it locally
53:33 um let's see please go live again soon
53:37 okay I will all right
53:40 um I've been at this for about an hour
53:41 now if anyone has any other questions
53:43 jump in there pop them in
53:46 otherwise isn't it kind of frightening
53:48 that even expert AI researchers don't
53:50 really understand how AI works yeah it's
53:51 frightening
53:53 um
53:55 I mean they they understand technically
53:58 how they work they don't quite
53:59 understand
54:01 how these emergent behaviors are
54:04 happening
54:05 so they don't really understand
54:08 you know
54:10 they shouldn't be as good as they are
54:12 right so so that's a little
54:15 disconcerting but
54:17 the thing that the Transformer did the
54:20 Transformer technology was allowed
54:22 machine learning neural networks to be
54:26 trained on massive data sets
54:28 and so what you've got with GPT three
54:31 and four is you've got a tool
54:34 that is essentially learned
54:36 all of the internet like like think
54:39 about from like the mid 90s until
54:42 September 2021 think about all of the
54:45 content that has been published in
54:48 Wikipedia and every blog and every
54:52 non-paywald news site and and every
54:55 website anything that can be crawled
54:58 um has essentially been crawled and it
55:01 knows all of that right so so when you
55:03 have that amount of stuff
55:06 um even though it's doing something that
55:08 is you know
55:11 um
55:12 probability based
55:14 um it's it's a massive massive uh data
55:18 set that it's learning from and then the
55:20 other thing that happened which I find
55:21 this fascinating
55:23 so they had gpt3 for a while and it was
55:25 okay
55:26 and then they put together this code
55:29 model I forget what it's called codex I
55:31 think or something like that what what
55:32 what GitHub co-pilot is based on they
55:35 basically trained up you know these
55:37 large language models on all of the
55:38 computer code right
55:40 so over here it learned all of the words
55:43 and over here it learned all of the code
55:45 and a funny thing happened was once it
55:47 learned all of the code it got a lot
55:49 better at the words now my thought there
55:51 is that it learned logic with the
55:53 programming and it could now take those
55:56 patterns of logic and apply those to
55:58 language right because language in a lot
56:00 of ways is human programming language
56:03 right it's you know when we talk to each
56:05 other it's it's in code there's there's
56:07 lots of different things but language is
56:08 really powerful for us as humans so
56:12 um so they saw these emergent behaviors
56:14 where the language model got better when
56:17 it got better at programming so
56:19 um I find it a little scary that I don't
56:21 understand or that they don't understand
56:23 how that actually works
56:26 um
56:27 um but but I don't know these tools are
56:30 growing so fast they're they're going to
56:32 very quickly dramatically outstrip our
56:37 capability as humans to understand them
56:39 so
56:40 I think you know
56:42 we're lucky to be able to play with them
56:44 I feel incredibly lucky right now
56:47 um
56:51 my comments about the government could
56:53 clock the AI Trend I don't think they
56:56 can I don't think they can either
56:58 um no I think because everything's open
57:00 source now I don't think the government
57:01 could get their hands around this
57:03 anymore
57:04 wait till it's used to hacks on nuclear
57:06 code and then see how fat well
57:08 yeah but so so that's the thing is is
57:11 like so so even if someone hacks the
57:13 nuclear code and something truly
57:15 horrific happens and the government says
57:17 we gotta lock this [ __ ] down
57:19 um the fact that so many of these things
57:21 have already been open sourced means
57:23 that they're open source they're they're
57:24 out there
57:25 um they can make them illegal and and
57:27 things like that but it's not gonna it's
57:29 not gonna stop people from using them so
57:32 um I I think the big commercial
57:34 companies are probably going to get
57:37 sacked with regulations right I think I
57:39 think open Ai and Microsoft and
57:41 whoever's using these things you know
57:43 Google might get them locked down but
57:45 the open source stuff they won't be able
57:48 to lock it down
57:50 um is this that Brokeback Mountain [ __ ]
57:52 I don't know what you're talking about
57:55 [Music]
57:56 um
57:57 [Music]
58:00 yeah but uh thanks for your hard work
58:03 you're welcome
58:05 um
58:08 how fast do I think the progress is to
58:11 AGI I don't know I mean
58:15 I think that you know there's there's a
58:18 Microsoft paper that was released a
58:20 month or so ago that that says gpt4
58:23 shows is showing signs of AGI
58:26 um signs of it uh it's it's a really
58:28 long paper
58:29 um it's it's worth exploring if you
58:32 don't have ADD like I do
58:35 I'll just rely on summaries of others
58:38 um
58:40 the the problem with with the term AGI
58:43 is that there's not an agreed to
58:45 definition of what it actually means
58:48 um I I think that we're past the Turing
58:51 test at this point I I some of my
58:53 interactions with chat GPT
58:56 um
58:57 it's actually the thing that makes it
58:59 makes it clear that it's not a human is
59:01 how how fast and how good it is like no
59:04 human could be that fast
59:07 um
59:07 but but AGI you know artificial general
59:11 intelligence is is very undefined right
59:14 now I think this combination of
59:18 autonomous agents and memory the ability
59:21 for these agents to act on their own
59:23 behalf remember what they did
59:26 learn from what they did
59:28 um I I think it's going to accelerate
59:30 quite quickly I mean I mean also the
59:32 fact that so many developers are doing
59:35 so much work on this technology it's
59:38 it's going to come
59:39 um quite quickly I think that Ray
59:42 Kurzweil predicted
59:44 um 2029 and I think a lot of the AI
59:48 scientists have now
59:51 um shortened their prediction to 2030 so
59:53 Ray Kurtz file has been at 2029 for a
59:56 long time the AI scientists are now down
59:59 to 2030.
1:00:00 um I'm guessing it could come faster
1:00:02 than that I I think I think we're going
1:00:04 to start to see exponential Tech
1:00:05 development
1:00:08 um
1:00:09 what's my next move for a product or
1:00:11 service to monetize it given this uh AI
1:00:14 technology for me personally
1:00:16 it's rolling AI
1:00:19 um
1:00:20 into my product but doing it in a way
1:00:22 that it it dramatically expands what
1:00:25 what my product is so I've got a video
1:00:27 technology company that that makes it
1:00:30 automates video production and allows
1:00:33 our clients to to have people tell their
1:00:35 stories really powerfully we have an app
1:00:36 that guides an end user through telling
1:00:38 their story goes up to the cloud a few
1:00:39 minutes later you got a fully edited
1:00:40 video
1:00:41 and with generative AI dramatically
1:00:44 expanding the output side of what our
1:00:47 company does but also dramatically
1:00:49 expanding the input side of it so for me
1:00:51 personally it's about rolling this stuff
1:00:54 both technically and conceptually into
1:00:57 what my company's all about because if I
1:01:00 don't the value of what we do now is
1:01:02 going to drop to zero very very quickly
1:01:04 and it's terrifying
1:01:08 um
1:01:09 oh this is great so I found myself
1:01:12 thanking chat GPT so I find myself
1:01:14 thanking it I find myself getting a
1:01:18 little guilty if I ask it to do too much
1:01:20 like if I ask it to rewrite something
1:01:21 for like the sixth time if if it were a
1:01:25 human I would just not I wouldn't have
1:01:27 asked them the third time to rewrite it
1:01:28 you know so
1:01:30 um
1:01:31 there there's something about
1:01:34 um talking to chat GPT as if it's like a
1:01:38 creative collaborator
1:01:40 um where you where you talk to it like a
1:01:42 human and you give it more context
1:01:44 um it I think it generates better work
1:01:46 um whether or not you actually need to
1:01:47 thank it I don't think it means anything
1:01:49 to it
1:01:50 um but I think if if there's something
1:01:52 about the tone of how you write
1:01:55 um it's doing predictive you know
1:01:57 answering on the other side of it so the
1:01:59 more context that you give it
1:02:02 um the more it colors what you get back
1:02:04 so maybe it gets a little more pleasant
1:02:06 in its responses to you as well because
1:02:08 it because you're doing that so I do too
1:02:11 like I you know I think we
1:02:13 anthropomorphize a lot of things anyway
1:02:15 we certainly do it with pets right
1:02:18 um so I think it's going to be really
1:02:19 easy to do it with these tools
1:02:21 especially if you see some of the some
1:02:22 of the tools that are you know
1:02:24 synthesizing you know people's faces and
1:02:27 voices and putting a chat GPT behind it
1:02:29 and if it does it quickly where you're
1:02:31 just talking to someone and it's talking
1:02:32 back to you
1:02:33 um yeah we're gonna pretty quickly get
1:02:35 to the someone asked that before about
1:02:37 how close to her do I think we are I
1:02:40 think we're probably
1:02:42 hell of a lot closer than we were than
1:02:45 we were a year ago I mean
1:02:47 I would say probably within
1:02:52 within three years we'll have robots
1:02:54 that are not as weird as they are now
1:02:56 and within five we'll probably have ones
1:02:59 that are that are pretty remarkable and
1:03:02 I think probably within
1:03:04 two years we'll have virtual you know
1:03:06 on-screen
1:03:07 you know like you're chatting with
1:03:09 someone on FaceTime but it'll be a bot
1:03:10 and it'll just be like talking to
1:03:12 someone I think that's probably within
1:03:14 two years have I used the Chrome
1:03:16 extension web chat GPT there's a bunch
1:03:19 of extensions out there um I've used
1:03:21 some of them my problem is things are
1:03:23 moving so fast I like I installed like
1:03:26 every time I install something it shows
1:03:28 up as a feature in in some tools so I'm
1:03:31 just I'm just trying to slow roll what I
1:03:33 install
1:03:34 I'm not located in New York I'm an ex
1:03:37 New Yorker I miss New York
1:03:39 um I will definitely go back to New York
1:03:41 but I'm not there right now I'm in
1:03:43 Denver it will collapse the stock market
1:03:45 I think it's going to have dramatic
1:03:47 impact on everything I don't know if
1:03:49 it's going to collapse the stock market
1:03:50 it could it could
1:03:52 um expand it but I don't know AI will be
1:03:56 doing the trading I agree with that it
1:03:57 to a lot of degree
1:03:59 a lot of those speed trading systems are
1:04:02 all AI anyway like that's the thing none
1:04:05 of this AI stuff is new like ai's been
1:04:07 around for decades right what's new is
1:04:09 us as real people now have access to
1:04:12 these tools that are you know
1:04:15 otherworldly powerful
1:04:19 um
1:04:20 AI to analyze technical analysis for
1:04:22 stock trading yeah if you want to if you
1:04:23 want to play with that
1:04:25 um go to gpt4 3.5 or 4 whatever chat GPT
1:04:29 three or four
1:04:31 um tell it to recommend a stock trading
1:04:34 site that does paper trading right where
1:04:36 you can do it without real money I mean
1:04:38 if listen if you got lots of money and
1:04:39 you want to go blow it do that
1:04:41 um but but go to a site that's got paper
1:04:43 trading that has Bots
1:04:45 and then either go find a bot or two
1:04:47 that you like copy and paste the code
1:04:49 out of it
1:04:50 um put it into chat GPT and have it you
1:04:53 know you can have it Mash two Bots
1:04:55 together or you can just say hey cheche
1:04:57 EBT give me five different strategies
1:04:59 for trading and then and then say write
1:05:01 me a bot for this particular site you
1:05:04 know in the following code and it'll do
1:05:05 that and then you can go play with it
1:05:07 and try it so yeah I think you can
1:05:10 um you can absolutely use AI to do
1:05:12 technical analysis
1:05:15 um and and you know and you know buy and
1:05:17 sell triggers and stop losses and all
1:05:19 that sort of happy horse crap I'm
1:05:21 horrible at that stuff so I'm not going
1:05:24 to touch that oh I would do it with um
1:05:27 with paper trading right uh but but with
1:05:29 real money it's just it's ridiculous for
1:05:32 me to do something like that advice for
1:05:33 someone experienced in server
1:05:34 development for 20 years next career
1:05:36 move I think right now uh you know the
1:05:39 if if you're in if you're in server
1:05:41 development I mean there's
1:05:45 like
1:05:46 the need to have service they need to
1:05:48 have computers running you know isn't
1:05:50 going to go down it's going to go up
1:05:56 um just get um
1:05:57 get as knowledgeable as you can about
1:05:59 this AI stuff just learn as much as you
1:06:02 can in your field or in something you're
1:06:04 interested in even if it's not in your
1:06:06 field just start getting
1:06:08 um as educated as you can with it um
1:06:11 with things like security and coding I I
1:06:14 would I would start digging deep into
1:06:15 autonomous agents I think things like
1:06:17 server maintenance and and
1:06:20 um keeping like devops kind of stuff
1:06:23 keeping
1:06:24 um libraries up to date all of that
1:06:26 stuff is going to go from really
1:06:28 complicated hard-coded systems that you
1:06:31 either have to pay for or build yourself
1:06:32 to just have some version of Auto GPT
1:06:35 just go do it for you right or an
1:06:38 autonomous agent could do it for you
1:06:40 so if you understand those tools and can
1:06:42 go build those tools
1:06:44 um uh you're gonna you're gonna be in
1:06:46 high demand I think
1:06:47 um so
1:06:49 um just get get going
1:06:51 hang on
1:06:54 um yeah we'll do the trading I'll be
1:06:56 doing the learning it will collapse the
1:06:58 stock market
1:06:59 let's see I got chat 250 to write a
1:07:01 piece with extreme emotion and emojis
1:07:04 yeah it'll write with Emojis it's pretty
1:07:05 good it'll write tables too if you
1:07:07 haven't gotten chat GPD to write a table
1:07:09 for you it's pretty pretty amazing
1:07:12 um you do realize that the more people
1:07:13 that hand over their thinking process to
1:07:15 AI the quicker we go no I don't agree
1:07:17 with that and I don't I'm not handing
1:07:19 over my thinking process to AI I'm
1:07:21 handing over the crappy low-level
1:07:23 tactical execution [ __ ] that I hate
1:07:27 that gets in the way of my creative
1:07:28 thinking I'm giving that to AI I'm
1:07:31 taking the high-end strategic stuff so I
1:07:34 you know
1:07:35 any any anyone who
1:07:38 it's so easy right now to to give into
1:07:41 the robots are going to kill us
1:07:44 um I think it's actually a tougher place
1:07:46 to go to go to the to the Light Side
1:07:49 come to the light come to the light it's
1:07:51 going to be okay
1:07:54 um I like my new t-shirt thank you oh
1:07:57 it's got crap all over it what's on
1:07:58 there
1:08:00 dinner
1:08:01 [Laughter]
1:08:04 um thank you for your advice how about
1:08:05 digging into the development of chat gbt
1:08:07 apis and other type of apis
1:08:11 um yeah I should do that I should do
1:08:12 some stuff on apis I do I do a lot of
1:08:14 work well not a lot of work I've built a
1:08:16 number of tools using zapier um doing
1:08:18 automations no code tools where I'll
1:08:21 take inputs from a form and then
1:08:24 um take those inputs roll them into a
1:08:28 prompt for chat GPT send it off get the
1:08:31 result back and then send out an email
1:08:32 so I've built I built a number of
1:08:34 content generation tools with that
1:08:37 um but yeah I mean listen the apis for
1:08:39 these things have been out for years so
1:08:41 so a lot of developers have been using
1:08:43 them so this isn't as new as it seems um
1:08:46 just what's new is our access to it um
1:08:48 like you're thinking very professional
1:08:49 and pleasant to listen to thank you
1:08:51 thanks for your time I've been using a
1:08:53 chat gbt to use the API and it works
1:08:55 well yeah the apis are clean the apis
1:08:58 um for
1:09:00 um DaVinci 003 are quicker than chat GPT
1:09:03 because jet gbt is overloaded I think
1:09:06 that's going to balance out pretty soon
1:09:07 I still don't have access to the 4.0 the
1:09:11 gpt4 API which is bumming me out but
1:09:14 whatever
1:09:14 uh
1:09:17 um I'd make a great host for an AR AI
1:09:19 maker type show thank you very much I
1:09:21 appreciate that I'm gonna have to do
1:09:23 something with the hair
1:09:24 [Laughter]
1:09:25 by virtue of asking your questions to AI
1:09:27 it begins to analyze human thought
1:09:29 process yeah I think I think rather than
1:09:31 it's it's not really analyzing anything
1:09:34 um it I think
1:09:36 I think what these GPT tools are
1:09:38 actually doing is reflecting Humanity
1:09:41 right it's been trained on most of the
1:09:43 crap we've written for the past three or
1:09:45 four decades so and and you know way
1:09:48 before that right obviously all the
1:09:50 historical stuff that's there
1:09:52 um the hair is perfect yeah the hair is
1:09:54 special the hair looks like I could run
1:09:56 a jet ski off it
1:09:58 um all right I gotta run
1:10:01 thank you all for your time and hanging
1:10:03 out and uh yeah we'll talk soon I'll
1:10:06 make another video or two but thanks
1:10:08 everybody take care