aiLL April 30, 2023 - The Future of AI is Here Faster Than We Realize

Description

Chapters

Transcript

0:03 hello hello hello I guess I am live
0:08 it's the AI learning lab turn it on the
0:11 light here
0:13 let's see does that help
0:16 um
0:16 [Music]
0:24 hello hello hello
0:26 it's AI learning lab Kyle Shannon here
0:29 talking about things Ai and how yeah a
0:34 little bit
0:35 how is everybody
0:37 if you've got any questions put them in
0:39 the chat I might even bring if if you're
0:42 interested in having a conversation I
0:45 can bring people up on stage or just
0:46 learn that
0:48 so we can try that if you have any
0:50 questions about AI that's what this
0:52 channel is all about I'm happy to answer
0:53 any questions
0:55 um
0:56 I don't know much but I'm learning so
0:59 that's what we're looking to do here
1:01 hope everyone's well
1:03 happy what day is today Sunday happy
1:05 Sunday everyone
1:07 um
1:08 getting ready to uh to put some work in
1:11 on the everyday AI newsletter and
1:15 and talk about that one of the things I
1:19 talked about yesterday was uh how I see
1:24 um the current state we're in it with AI
1:27 as having very significant parallels
1:31 in
1:32 um
1:33 in the early days of the web so in the
1:35 early days of the web the internet had
1:36 been around for decades and then this
1:38 world wide web thing came along I
1:40 discovered it in 1994 and end up ended
1:43 up starting one of the first online
1:44 magazines and one of the first digital
1:47 agencies and this feels very much like
1:50 we're in the in the same kind of uh
1:53 same kind of era where where massive
1:57 change is coming not a lot of people
2:00 know
2:01 um what or how right now because there's
2:05 so much excitement there's a lot of
2:07 people acting like experts and there are
2:09 very few of them who actually are so one
2:12 of the things I try to do on this
2:14 channel is is talk very straight about
2:17 um what I know what I don't know
2:19 um and uh and just trying to keep up
2:22 with everything quite frankly because
2:23 the you know in comparing where we are
2:26 with AI to
2:29 um to the mid 90s with the with the
2:30 internet and the World Wide Web
2:33 um
2:34 things right now are moving way way
2:37 dramatically faster and I think the um
2:40 the implications even though the
2:42 implications for the popular internet
2:45 um are pretty profound um they're
2:46 nothing like what's going on here so
2:48 with that if you have any questions ask
2:51 me questions let me see if there's any
2:52 here I haven't looked yet
2:54 I someone just got the number one
2:56 glitter badge good Lord congratulations
2:58 that's exciting uh no one's asking any
3:00 questions if you have any questions let
3:01 me know I can also bring people up on
3:03 stage if you want to chat happy to chat
3:05 about AI other stuff
3:07 um you know I could talk about what can
3:09 I talk about I've started a lot of
3:11 companies I can talk about that I can
3:12 talk about writing uh but but uh the
3:15 thing this channel is all about is about
3:17 learning about AI generative Ai and
3:20 what's going on so if you've got any
3:21 questions ask away this is only my
3:23 second live so I don't know what the
3:24 hell I'm doing here I am I am making it
3:26 up as I go so so
3:29 um hopefully we can have some fun uh
3:31 talking about AI on a Sunday Sunday
3:34 morning I'm in Denver here so let me
3:36 know where you are and I'll give you a
3:38 shout out
3:39 and what else
3:42 um
3:43 trying to think I had a really good week
3:44 last week it's taken me about three
3:47 months to incorporate I just did a video
3:50 on this it's taken me about three months
3:51 to incorporate
3:54 generative AI tools with my current
3:56 company which is storyvine it's a it's
3:58 an automated uh video creation platform
4:02 and I finally came up with a vision for
4:04 how to incorporate all this stuff and
4:06 one of the first conversations I had
4:07 with was with one of our biggest clients
4:09 which is a big Pharma company and they
4:12 got so excited about it they're like fly
4:13 out look we should talk about this in
4:15 person
4:16 um so so that's really exciting uh it's
4:18 taken me a while to get there though
4:19 gpt4 worth it yes absolutely worth it it
4:22 is it is markedly better at a lot of
4:25 things coding reasoning creative writing
4:29 Nuance then GPT 3.5 is that said GPT 3.5
4:34 is very very fast and it's much cheaper
4:38 so if you're doing API calls and things
4:40 like that
4:41 um there's reasons to use 3.5 or even
4:43 three
4:45 um just just because of things like
4:46 speed and and cost but if you're just
4:49 talking about chat GPT is it worth
4:51 paying you know the 20 bucks I mean
4:52 there's a couple of things
4:54 there's there's an app and a website
4:56 called po po is the app Poe and po.com
4:59 poe.com
5:01 um it's made by uh the company quora the
5:04 question and answer company and
5:07 um there you can use multiple models so
5:09 you can use GPT 3.5 you can use gpt4 you
5:13 can use anthropic's Claude model uh and
5:16 maybe some others you can also create
5:18 your own chat bot um I made eight or
5:20 nine chat Bots like uh I am Shakespeare
5:23 if you go to po.com I am Shakespeare you
5:26 can interact with a chat bot that
5:28 responds in iambic pentameter and writes
5:30 like Shakespeare it's pretty fun
5:32 um so so Poe is a good one
5:35 um how can I create an app from the
5:37 scratch so so
5:39 if you're not um a programmer a coder
5:42 I'm not um I understand code and and I
5:46 have coded in the past but like way in
5:48 the past and and never uh
5:51 of of any um significance because I have
5:55 ADD and I just don't have the attention
5:57 span to really do coding
5:59 um but you can use no code or low code
6:02 tools
6:04 um to sorry this thing is teaching me
6:07 how to use it well whatever I can't
6:09 figure it out
6:10 um you could use a low code tool like
6:13 zapier which allows you to
6:17 um to string together a set of actions
6:20 it's an automation tool
6:22 um and and with that I've built a number
6:23 of apps now
6:25 um where I create a Google form uh ask
6:27 people a couple of questions in a Google
6:29 form I then take that and and
6:31 incorporate their answers into a prompt
6:33 send that to open ai's API and get back
6:37 um responses and it's it's pretty cool
6:39 you can email them or you can create a
6:41 Google doc or something like that so
6:42 that's how you can create apps from
6:43 scratch there's a there's a no code if
6:46 if you want to create true apps there's
6:48 one called Bubble which is still pretty
6:51 technical you know all of these things
6:53 are a little janky and Technical even if
6:55 they're if they're no code or low code
6:58 um but but yeah I would just say just
6:59 start playing with it figure out what
7:01 you want to do all right
7:03 are there best practices I found with
7:05 prompting
7:06 um yes
7:08 many a lot of the things that you'll see
7:10 are um
7:12 The Prompt chat gbt to act like a
7:16 certain thing so act like a
7:19 um a coding expert or act like a
7:22 marketer or act like a you know
7:24 controller or whatever it might be
7:27 and then you can ask it questions uh and
7:30 and it will it's it's answers will
7:32 improve dramatically so that that's
7:34 that's that's just a little trick but
7:37 here's here's the biggest thing that I
7:38 found is that
7:41 um
7:42 shifting your mental thinking your
7:44 mental thinking shifting
7:47 um how you think about prompting from
7:50 what we do with Google right now where
7:52 Google is kind of this single
7:56 um prompt single response you know find
7:59 me whatever cheap flights and then the
8:01 list of things comes up
8:03 if you treat chat GPT like that it's not
8:06 as good if if you just say you know I
8:08 want this and it gives you that if you
8:10 shift your mental model to go this is a
8:13 conversation it's called chat EBT think
8:15 of it like a conversation and think of
8:17 it more like less like this is a tool
8:20 that's going to give me the exact
8:21 answers and more like you're working
8:23 with like a really smart intern like
8:25 like the smartest intern you've ever
8:26 worked with because
8:28 it essentially knows all of the internet
8:30 right so it knows lots and lots of stuff
8:31 but if you think about it like that so
8:33 you ask it for something and it doesn't
8:35 give you what you want you say hey
8:36 that's not quite right do it like this
8:37 oh do it like that you know what I
8:39 really want is this can you turn that
8:41 into a table so so almost treat it like
8:44 you know a co-worker that you're working
8:46 with remotely and um and that makes a
8:49 big big difference so hope that helps
8:53 let's see there's nothing to learn
8:56 it does everything for you I completely
8:58 disagree with that it does a lot of
9:00 things for you but the intention of the
9:02 user the intention of the prompter is
9:04 everything and then the curation of the
9:06 answers is even
9:08 um more important
9:09 um so so you know when someone shares a
9:12 piece of artwork on Twitter that they've
9:13 done I guarantee you that there's been
9:15 you know a hundred images that they
9:17 tossed in the trash uh before that one
9:20 so
9:21 um so this is there this is very much
9:23 listen these large language models check
9:26 EPT and these image tools and the video
9:28 tools and the music tools all this sort
9:30 of stuff they're just tools right
9:33 we get to choose how to use them so the
9:35 intention of humans and then what we
9:37 choose to curate and refine and put out
9:39 to the world that's
9:41 um
9:42 that's ultimately uh what this stuff's
9:44 about so
9:46 um they are incredibly powerful tools
9:47 that said
9:49 are there portals in space and time uh
9:52 if you watch the Skinwalker Ranch show
9:55 you would certainly think so there's
9:57 something above the triangle area
10:00 um and and you know our large language
10:03 model is going to help us solve that and
10:04 get there I don't know maybe uh let's
10:07 see where are we here are there portals
10:09 is face and time you guys are funny
10:13 um can you please llm also in these
10:17 models
10:18 are independent later layers and what
10:21 they do I don't quite know what that
10:22 means if you could ask that again could
10:23 I please explain an llm
10:26 so llm stands for large language model
10:28 I'm assuming this is what you're asking
10:31 um large language model is is what llm
10:34 stands for
10:35 and it basically means
10:37 a an AI system that uses the Transformer
10:42 architecture so the the Transformer
10:45 architecture was invented by Google in
10:47 2017
10:48 it is the big breakthrough that all of
10:52 um the kind of modern AI stuff that
10:55 we're talking about and playing with
10:56 right now is based on and basically what
10:58 it allowed is it allowed for these large
11:01 for these AI systems these machine
11:03 learning algorithms to be trained on
11:05 massive data sets so so gpt3 was
11:09 essentially trained on all of the public
11:12 internet right so all of Wikipedia all
11:15 of you know websites that are kind of
11:17 accessible and freely accessible without
11:19 being behind you know passwords and
11:21 paywalls and things like that
11:23 um so it is remarkably deep in the
11:26 amount of knowledge it's been trained on
11:28 now
11:29 um they stopped training at September
11:32 2021 so chat GPT is is only good up to
11:35 there if you use the GPT chat within
11:40 Bing that's connected to the internet
11:42 and there's plugins coming for chatgpt
11:44 that are connected to the internet
11:45 that'll fill that Gap that time Gap
11:49 um and then
11:50 um and then the large language model is
11:53 basically just that that underlying
11:55 model that when you're interacting with
11:56 chat GPT and you ask it a question it's
11:59 then you know going into its database of
12:02 the knowledge of all of the words that
12:04 it knows
12:06 um and it's it's figuring out what is
12:08 the most likely answer that you want
12:11 based on your prompt so so that's that's
12:13 what they are so the llm is the is the
12:15 underlying layer and then there are lots
12:19 of applications going to be built on top
12:20 of that so you ask about the layers so
12:22 at the at the core level you've got
12:25 um
12:26 all of these words in context of you
12:31 know what they're written in all the
12:32 meaning of all the words basically goes
12:34 into this database and then there are
12:36 applications that sit on top of that so
12:37 if you use Jasper or copy AI those are
12:41 an application layer above the large
12:44 language model which is GPT and
12:47 basically there's a series of prompts
12:49 that are pre-written for you so if if in
12:52 Jasper you use the blog writer
12:55 um they have prompted chat GPT to to be
12:58 really good at writing blogs for that
12:59 particular tool so it's a layer on top
13:01 of the core large language model
13:04 now the large language models there are
13:06 lots of them last week there were four
13:08 of four open source models were were
13:10 announced or released
13:12 um there's the Lambda model which came
13:15 from meta
13:16 um there's
13:18 um uh stability just announced theirs
13:20 Google's got barred open AI has uh chat
13:24 or has GPT three and four three three
13:27 point five and four
13:28 so I asked you yesterday but didn't
13:32 uh
13:34 what did you ask me yesterday you didn't
13:36 answer and today too all right hang on
13:38 let me go see what you asked me
13:41 how can I create an app from scratch I
13:43 did answer that use things like zapier
13:46 low code or no code tools like zapier
13:49 um
13:51 z-a-p-i-e-r-dot-com or bubble if you
13:53 want to create an actual application
13:54 zapier is an automation tool for
13:56 creating web-based
13:58 uh tools uh you might be able to create
14:00 an actual application with it I think
14:02 with bubble you can do actual apps do I
14:05 think time travel is possible I don't
14:07 flip a no I have a degree in acting
14:10 um
14:13 I can pin a comment now are there
14:16 portals in space and time
14:19 all right ask any questions I think I've
14:21 got all the questions how hard is it to
14:23 build an extra later layer
14:26 for example
14:28 um
14:28 crypto GPT
14:30 uh if you're a programmer it's not hard
14:32 at all uh I I think that layering
14:35 different tools is is going to be uh
14:39 incredibly powerful if you look at if
14:41 you if you haven't
14:43 um
14:44 watched videos on or or looked into
14:46 these things called autonomous agents
14:48 you should they're incredibly new so
14:52 they're very very raw and and they don't
14:54 work all that well right now but they're
14:55 they're very interesting so one of them
14:57 is called Auto GPT there's another one
15:02 called Baby AGI
15:04 and what these tools do is they stack
15:06 together different different tools to
15:09 create powerful things and they also
15:12 prompt themselves so you say to Auto GPT
15:16 um go find me the I don't know the
15:18 profitability of the top five
15:21 um you know Global corporations in 2022
15:24 and it will give itself a task list it
15:27 will say okay go find the revenue you
15:29 know go find the top companies
15:31 um you know buy revenue or whatever and
15:34 then it will say go go find their
15:36 profitability and then it will say make
15:38 a table with that so it will actually
15:39 prompt itself and it will figure out do
15:42 I need to do this on Google do I need to
15:44 send this to Jet chat GPT
15:46 um so
15:47 a lot of developers are playing with
15:49 these autonomous agents right now
15:51 figuring out what combinations of tools
15:53 they can put together the other thing
15:55 that's happening is um
15:57 chat GPT has introduced a plug-in
16:00 architecture so I've got access to the
16:02 plugins right now they're they're very
16:05 very thin and and what I would call very
16:07 Alpha meaning you can't play with a lot
16:09 of them they're hard to discover they're
16:11 hard to share I think you can only share
16:13 a plug-in with up to 15 people or
16:15 something like that but what it allows
16:18 is if you have one of these plugins
16:20 turned on for example
16:22 um let's say there's there's one of the
16:24 plugins is for OpenTable you could say
16:26 to chat GPT I want to have a good steak
16:29 dinner on Saturday night you know for a
16:31 date and it will go find you
16:34 a restaurant that serves good steak
16:37 dinners and even book you a
16:40 um a table using the OpenTable plug-in
16:43 right so so by adding plugins to these
16:46 models what you're going to see is is
16:48 lots of combinations that create very
16:50 sophisticated applications
16:52 hope that helps let's see
16:57 um have I done any movies or television
17:00 work a little bit I was mostly a stage
17:02 actor but
17:04 um I was I was white boy number two in
17:07 New Jack City I was I was like a kid
17:11 from Connecticut buying drugs in Harlem
17:12 that was fun
17:14 it was fun to shoot I also
17:17 um my best friend Steve session
17:19 um he and I met filming an episode of
17:22 America's Most Wanted he was the Killer
17:24 and I was the best friend of the Killer
17:28 and I remember one of my lines was uh
17:31 Adam about this gun thing
17:33 right before you shot someone in the
17:35 back
17:36 um let's see
17:38 how do you generate wait
17:42 that's nice I remind someone of their
17:45 homeroom teacher in high school I I
17:46 guess that could be a good or bad thing
17:50 um shouldn't there be more custom data
17:52 sets data sets like Bloomberg GPT would
17:55 that bring more value yes I think
17:57 private data sets uh are gonna be are
18:01 going to be where the primary value is
18:02 because I think the large language
18:04 models are going to be ubiquitous
18:05 they're going to be everywhere and so I
18:07 think being able to search on data that
18:10 you trust and data that is uh
18:13 specialized to what you want to solve I
18:15 think it's going to be quite powerful so
18:16 I think Bloomberg GPT is is really
18:19 interesting I was not a teacher ever
18:21 um
18:23 foreign
18:27 sorry going through here
18:30 um
18:33 um how do you generate actually
18:35 wait how how to do it to generate
18:40 actually applicable code I assume you
18:42 mean chat GPT
18:45 um if you so so this was one of the
18:47 first breakthroughs I had
18:49 um with chat GPT
18:51 was
18:53 um I had heard it was good at code and I
18:55 said write me a python script that takes
18:58 an input and does something with it and
19:00 so it generated all of this code and I'm
19:03 like I realized after it generated the
19:05 code I don't know how to run python code
19:07 so then I told chat gbt how do you run
19:10 python code and it did uh it it said uh
19:16 if you're on a Mac you go to the
19:18 terminal and you type in this command
19:20 line I'm like I don't want to use
19:21 command line can can I do it on a
19:24 website it said oh sure here's some
19:25 websites that can run python code and
19:27 then it told me how to do it paste it in
19:28 the left-hand window hit the play button
19:30 and it'll generate the code so I copied
19:32 the code pasted it in the window hit the
19:34 play button and it gave me an error and
19:36 I'm like here's the error it gave me and
19:37 it said well you need to install this
19:39 library and I said how do you install
19:41 that library and it said
19:44 um you click in the lower left hand
19:45 corner of the window and then you search
19:47 for the library and you install it and
19:48 within 90 minutes I had written a very
19:51 sophisticated python application that
19:54 was functional
19:55 having never written a line of python in
19:57 my life so so basically you just you you
20:00 use chat EPT to teach you how to
20:03 generate applicable code and even how to
20:06 run it
20:08 um the question gets mixed in the
20:09 Transformer I I don't quite know
20:11 technically how all the stuff works the
20:12 Transformer
20:14 um I don't quite know how it technically
20:17 works but what happens is
20:18 the data that you put into the large
20:22 language model I assume via the
20:24 Transformer gets embedded and basically
20:26 what embedding does is it turns every
20:28 word into a vector and a vector is think
20:32 of it just like an XY chart when you
20:34 plot a point on an XY chart that's a
20:36 vector so every word gets plotted in not
20:40 two-dimensional space but like
20:42 thousand-dimensional space it's
20:43 incredibly complicated mathematical
20:45 space but all of these words are sort of
20:47 plotted
20:48 in groupings of semantic meaning so if
20:51 there's multiple meanings for a word it
20:54 might live in multiple places within all
20:56 these word clouds so so when you tell
21:00 chatgpt to act like a certain thing it's
21:02 actually kind of narrowing What's called
21:04 the latent space it's it's narrowing
21:06 sort of the word choices or the semantic
21:08 meaning choices
21:10 um within this thing so it's fascinating
21:12 how it works
21:14 all right did I teach in school no
21:17 um
21:17 see what else we have have I done any
21:19 movies covered that one
21:21 what is embedding and how does that tie
21:23 into the llm so that's what I just
21:24 described so embedding is taking a
21:27 document or a series of documents or a
21:29 series of data and encoding them into
21:32 these it's basically it's wild it's a
21:35 string of numbers and the string of
21:36 numbers represents some position
21:39 in in What's called the latent space so
21:42 this multi-dimensional
21:43 thousand-dimensional
21:45 um space of meaning with all the words
21:48 and then when you search what it's doing
21:50 is it's literally looking for what's the
21:52 next most probable word in the answer
21:55 based on all of the stuff that was
21:56 submitted
21:58 and it just does that over and over and
22:00 over again so when you see it respond
22:02 with a whole paragraph it's literally
22:03 you know typing that out word by word
22:06 just using probability of of what's the
22:08 most likely word to come next it's wild
22:11 the the more you understand this the
22:13 more remarkable it actually is auto GPT
22:16 is very good it's it it's new
22:20 um but it's it's quite profound what
22:22 opportunities do you see with AI
22:24 websites tools and that everyone is
22:27 missing and not paying attention to
22:30 um
22:30 all of this stuff is quite new
22:33 um but I think
22:35 I think the idea that you have to
22:39 learn how to code a website that you
22:42 have to learn how to design a website I
22:44 think that's going to be a thing of the
22:46 past relatively quickly
22:48 um and and similar with with software
22:52 certainly simple software more complex
22:54 software I think is still going to
22:56 require a fair amount of engineering
22:58 but I think we very quickly move into a
23:01 world where with any kind of knowledge
23:03 and creative work I did a video on this
23:06 called are you a virtuoso or are you a
23:09 conductor
23:10 and what I meant by that is if if you
23:12 just do one specific thing like I design
23:14 images or I write you know copy or
23:17 whatever it is
23:19 um those individual tasks are going to
23:21 get automated very very quickly and so I
23:24 think the skill moving forward is
23:27 you know what's the strategy what am I
23:28 trying to accomplish
23:30 um what tools do I need to accomplish
23:32 that what does good look like what does
23:34 success look like all of that remains
23:36 really important and I think that
23:38 becomes our role and then what you will
23:40 do is you will have tools out there
23:42 where you say you know design me a
23:44 landing page or design me a a an
23:48 application and it will just design it
23:50 and then you say make it function like
23:52 this and it will just do that and if you
23:54 look at these Auto GPT tools I even
23:56 heard a concept of
23:58 you know the idea of a single use
24:00 application where you ask the tool to do
24:02 something it figures out if there's an
24:04 app for that if there's not an app for
24:06 that it will just build you one and then
24:08 you use it you do your thing and then
24:10 that app goes away like that's that's a
24:12 possibility
24:14 um so so I think I think website design
24:16 and building is going to get quite
24:18 automated
24:19 um but but you know how we put them
24:21 together and what we get them to do that
24:23 remains you know we're using these tools
24:26 so in terms of specific tools I don't
24:29 know it's things are changing so fast
24:30 right now I would say just if you're
24:33 interested in website building tools
24:34 just start go to futurepedia.io
24:38 um and and look for tools that are doing
24:41 web development stuff and just start
24:43 playing but they're changing so fast
24:44 that I wouldn't get too married to
24:46 anything do I know Steven from zero
24:47 shoes yes in fact he was the guy that I
24:51 met he and I met on an episode of
24:53 America's Most Wanted where he played
24:55 the Killer and I played his best friend
24:56 so yeah we've been and and what was cool
24:59 was we filmed that in um
25:02 in Connecticut but we were both living
25:04 in Manhattan at the time and it turned
25:06 out we lived on the same street one
25:07 block from each other so we became good
25:10 friends and we've been friends ever
25:11 since he's actually one of the big
25:13 reasons why I moved out to Boulder and
25:14 now I'm living in Denver all right
25:16 can I ask a question about AI
25:20 image Generations sure ask away
25:23 um what's your background since you know
25:25 so much about AI programmer no I have a
25:29 background in acting and storytelling
25:32 um but in the mid 90s I um I was doing
25:35 desktop publishing as my day job and I
25:38 discovered this new thing in 1994 new to
25:41 me
25:42 um called the world wide web and I
25:44 started one of the first online
25:45 magazines an art and culture magazine
25:47 that won a lot of awards and then I
25:49 co-founded agency.com which was one of
25:51 the first digital agencies so I learned
25:54 a lot about
25:56 um storytelling at scale about coding
25:58 about innovating about strategy and how
26:01 you take
26:02 a big company like British Airways and
26:05 and translate that brand into an online
26:07 environment in fact we built British
26:10 Airways first ticketing system we built
26:13 serious satellite radios first music
26:15 streaming system like this was we were
26:17 developing websites back in the time
26:19 where none of that stuff existed it
26:21 wasn't really possible
26:23 um so so I kind of earned my chops just
26:25 building stuff for clients
26:28 uh
26:29 uh what are some good AI sites to help
26:31 start
26:32 to start an online business uh if you
26:35 haven't played with it yet
26:36 um check out GPT boss that's from
26:38 McKinsey Bose I think is his last name
26:41 he's got a tick tock Channel
26:43 um go to GPT boss uh The Tick Tock
26:45 Channel and you'll see that
26:47 um he's a great guy and GPT boss is
26:49 specifically designed to be a series of
26:52 virtual employees that help you start a
26:54 business so there's you know
26:56 you hire these different functions so
26:59 you can hire an accountant and you can
27:00 hire a marketing strategist and you can
27:02 hire a copywriter and and basically
27:04 these are just little pre-prompted
27:06 personas that specialize in certain
27:09 things so if you want you know mental
27:12 health coaching for for running your
27:15 startup which you will probably need
27:18 um you basically just ask the you know
27:20 ask your questions to the to the
27:21 psychologist there so that's a fun one
27:23 you can start with but but there's lots
27:24 I would say
27:26 you know just using chat GPT think of it
27:30 like a business partner or a co-founder
27:32 might be an interesting way to start a
27:34 business
27:35 um
27:36 how can you train GPT on your data set
27:39 that's a huge one I think it's
27:41 incredibly important I think you know
27:44 Apple's got a big announcement coming in
27:46 June I have a sneaking suspicion that
27:48 apple is going to be all about
27:52 um using i a i on your own data
27:56 um
27:57 I might be wrong but I think I think
28:00 using AI on your own data sets either
28:03 your personal data set or your business
28:04 data set
28:06 um
28:08 is going to be where the real power is
28:10 in this stuff moving forward right now
28:12 if you want to do that there's a thing
28:14 out there called Lang chain l-a-n-g like
28:17 language chain that chains together a
28:19 bunch of things and what you do with
28:20 Lang chain is you can point let's say
28:23 you've got a directory of PDFs or you've
28:26 got a directory of Word documents you
28:29 sort of Point Lang chain at that
28:31 directory and then you have it use open
28:34 ai's embedding engine so you embed all
28:36 of those documents you store all those
28:39 vectors from the embedded documents in a
28:42 vector database like pine cone and then
28:45 you can basically chat GPT search on
28:48 your own data so Lang chain is a way to
28:51 do that there there are lots of tools
28:54 Now sort of combining that with other
28:56 things like Auto GP Auto GPT and baby
28:59 AGI incorporate Lang chain so what's
29:02 happening with the developing Community
29:04 right now is they're starting to combine
29:07 lots of different tools to create
29:09 increasingly sophisticated capabilities
29:11 so so that's a way to do that
29:13 oh Steven is awesome and a great friend
29:15 yeah Steve's great
29:17 I haven't talked to him in a bit let's
29:19 see
29:21 um
29:21 what are my go-to AI apps I use
29:24 mid-journey mostly for images although
29:27 dream Studio from stabilities good
29:31 um
29:32 I did the the image that you see on
29:34 screen right there is from stable
29:37 diffusion so I set up a stable diffusion
29:41 collab thing so I was kind of running it
29:43 running it in a virtual machine
29:46 um for the for the language stuff I use
29:49 open ai's playground or chat GPT mostly
29:52 chat GPT I would say any anything where
29:54 I'm trying to solve a problem I'm
29:57 generally doing it in in chat GPT
30:01 um if I want to codify things and make a
30:04 tool I use zapier which is low code it's
30:07 a little janky like zapier's feels a
30:09 little dated to me there's one that I
30:11 haven't tried yet but looks pretty good
30:13 called make
30:15 um make dot something make dot IO maybe
30:17 but it's an automation tool you can you
30:18 can find it by Googling it
30:20 um
30:21 and those are kind of the biggies right
30:23 now
30:24 um
30:25 uh you know if if I if I need to create
30:28 background music for for videos moving
30:32 forward I think I'm going to use sound
30:33 draw which if you haven't seen sound
30:35 draw dot IO that's pretty interesting
30:36 for music
30:38 um none of the video stuff right now is
30:40 is all that interesting you can only do
30:42 three second clips and and they're kind
30:44 of like the video stuff right now is
30:45 kind of where images were a year ago
30:47 they're not quite good enough to they're
30:48 just weird if you've seen that there's a
30:50 pizza Commercial someone made using AI
30:53 you know text to video stuff and it's
30:55 kind of hilarious but a year from now
30:58 it's not going to be hilarious it's
30:59 going to be you know very
31:02 um very useful
31:04 why does everyone insist on pushing
31:06 forward with AI That's a great question
31:07 so
31:09 AI has been
31:11 progressing
31:13 um I would say
31:14 you know incrementally
31:16 for
31:18 um a few decades a couple of decades a
31:20 few decades probably a few decades
31:22 machine learning's been around for for a
31:23 long long time and neural networks I
31:25 forget when they were created maybe the
31:27 60s
31:28 but but it's it's been going for a while
31:30 and then in 2020
31:34 um gpt3 came out and gpt3 was the first
31:38 of these large language models to really
31:40 start you know doing something good and
31:42 and shortly after that
31:45 um stable diffusion image generation got
31:48 good and mid-journey started getting
31:49 good and we started hearing about all of
31:52 these
31:53 um
31:54 these image generation tools which went
31:56 from kind of comically bad to really
31:58 really good in a very short amount of
32:00 time
32:01 and then
32:03 um November 30th
32:05 um 2022
32:07 is the date that in my opinion changed
32:11 everything and that's the date that chat
32:13 GPT was released why did that change
32:15 everything well
32:17 chechi PT itself wasn't
32:20 all that interesting actually from a
32:22 technical perspective in fact this is
32:24 this is kind of a cool thing
32:26 the instructions for how to create
32:29 chatgpt we're actually in the
32:31 documentation of gpt3's documentation
32:34 for from open AI so so the developers
32:38 um thank you for the hat that was lovely
32:41 um the the developers at open AI
32:44 basically gave you the instructions for
32:45 how to create chat GPT and Sam Altman
32:47 said we assume someone would make it and
32:49 no one made it so
32:51 so I think about a month before
32:53 Microsoft was going to make their big
32:55 announcement they that they invested a
32:57 billion dollars in open AI
32:59 um Sam Altman said you've got two weeks
33:01 go make chat GPT and so they built that
33:04 thing in two weeks
33:07 um to to put the impact of chat GPT in
33:10 perspective so the World Wide Web which
33:12 I was I was you know a part of the early
33:15 days of the World Wide Web
33:17 to get to a hundred million users it
33:19 took eight years
33:21 uh for the World Wide Web to get to 100
33:22 million users chatgpt got to a 100
33:26 million users in
33:29 um six weeks month and a half
33:31 100 million users
33:33 so so because of that a lot of things
33:37 have happened there's all these
33:38 cascading effects so because so many
33:41 people used it like it just hit the
33:42 cultural Zeitgeist right all of a sudden
33:44 everyone was talking about it so this is
33:46 Tech that had been out you could play
33:47 with gpt3 for a year and a half before
33:49 that I played with it I was like yeah
33:51 this is cool but there was something
33:53 about chat EPT that just caught the
33:54 imagination and then because it got to
33:56 100 million users so quickly it forced
33:59 the hand of everyone else Google
34:01 invented the technology that is
34:03 underneath chat EPT but they didn't
34:05 release it because they knew if they
34:07 released it it would threaten their
34:08 search business well they were forced
34:10 out of the Shadows
34:12 um Amazon has now been forced out of the
34:14 Shadows meta has now been forced out of
34:16 the Shadows apple is kind of on the
34:18 verge of making a big announcement so
34:20 all of these companies that that had a
34:23 lot of AI technology nice hat thank you
34:27 um
34:28 a lot of these companies that had this
34:30 technology were kind of keeping it
34:32 behind the firewall chat gbt forced it
34:34 in the open so that that was one thing
34:35 the other thing was this
34:37 the developer Community because it it
34:40 you know had all of this Buzz about it
34:42 the developer Community right now it's
34:45 it's all they're focused on um so like
34:47 Auto GPT came out that was this thing
34:49 that strings together these actions it
34:51 got to I think it's a hundred thousand
34:54 Yeah a hundred thousand likes or stars
34:56 or parts or whatever on GitHub
34:59 um it did that like in two weeks right
35:01 so so incredibly quick adoption and what
35:04 I'm seeing very consistently is people
35:07 are
35:08 um someone will put a paper out and
35:10 within two or three days someone will
35:12 launch a Technology based on that paper
35:14 or they'll put another paper out that
35:15 incorporates that paper into two other
35:17 papers and they'll create this new
35:19 hybrid kind of thing so the amount of
35:21 development happening right now is
35:23 insane I've never seen anything like it
35:26 so
35:27 um I don't know that anyone is insisting
35:29 on AI pushing forward I think it has
35:32 just broken through in a way that means
35:34 it's not going away and the acceleration
35:36 is not going to slow down
35:38 the call for a six-month pause on on AI
35:42 Innovation from Elon Musk and others is
35:45 interesting I don't think it's got a
35:46 chance of happening
35:48 um I also I I view it a little cynically
35:50 with musk in particular a week after he
35:52 called for a six-month pause he
35:54 announced the creation of x dot AI his
35:57 own AI company to compete with openai so
36:00 he helped fund open AI he doesn't like
36:02 the direction it's going so he's going
36:03 to compete with it so I was a little
36:06 cynical about that six-month pause
36:08 anyway I don't think it's got a a
36:10 realistic chance in hell
36:13 um
36:13 are there issues that need to be dealt
36:16 with that a six-month pause might help
36:17 with absolutely there's lots of issues
36:19 that need to be dealt with but there's
36:21 no stopping it
36:23 um let's see
36:27 their xlab 5000 is a no code solution
36:30 for pretty much everything I don't know
36:32 who there is
36:35 let me know who that is I'll I'll
36:37 say that
36:39 where are we here
36:42 isn't it dangerous for our economy and
36:45 our way of life uh I don't know if it's
36:47 dangerous but it is going to be
36:48 disruptive on a scale that I don't think
36:50 we can imagine right now and it's coming
36:52 way faster than we think
36:54 I think short term it's going to be
36:56 destructive but I think I think
37:00 um
37:01 I think long term it's it's potentially
37:03 interesting I just think lots of things
37:04 change a lot of our economy right now is
37:08 based on complex systems everything's
37:10 really complicated what these AI tools
37:12 is going to allow us to simplify things
37:14 and make things more efficient and so
37:16 it's going to eliminate some jobs that
37:18 are are there to sort of do the
37:20 complicated stuff
37:22 um but it's gonna it's gonna create new
37:24 jobs that we can't even imagine what
37:26 they are yet
37:27 um so but I but I do think the next two
37:29 or three years is going to be very very
37:30 disruptive
37:32 um you know I'm I'm personally very
37:35 excited I it's funny the more I use
37:37 these these tools that you know in
37:41 theory automate creativity or automate
37:43 knowledge work
37:45 I find myself being more creatively
37:47 engaged than I've ever been
37:49 um because what what the tools allow me
37:52 to do is operate at kind of a higher
37:54 creative
37:55 level where I'm just thinking about the
37:58 Creative Concepts and the tools are
38:00 doing the execution so before if I had
38:04 an idea I would go to do something and
38:06 I'd go like oh I need to use a software
38:08 tool and then I'd have to go
38:10 you know sign up for that tool and then
38:13 I'd have to put in my credit card and I
38:14 and I have to do this and then I'd have
38:16 to go watch a YouTube video to learn the
38:18 tool and then I'd realize the tool
38:19 didn't do what I wanted and 45 minutes
38:20 later I don't even know what my creative
38:22 idea was anymore what's happening with
38:24 the AI stuff is I'll have kind of a
38:26 vapor of an idea and within five minutes
38:28 I can sort of bring that thing to life
38:30 and so that gives me another idea so I
38:33 find myself kind of executing at a much
38:35 higher
38:37 um I don't know strategic creative level
38:39 with these tools so as opposed to it
38:41 making me less creative I find it making
38:43 me more creative so I think there's a
38:45 silver lining to it but it is absolutely
38:48 going to be disruptive for I I I can't
38:52 think of
38:54 um a profession right now that isn't
38:56 going to be impacted
38:59 let's see what do I think is dangerous
39:02 there may be risks but oh oh well I mean
39:05 the risks are so like the the auto GPT
39:08 thing like you know someone could very
39:10 very easily without them not without a
39:13 lot of technical knowledge right now
39:14 could use something like Auto GPT to say
39:17 go find all the systems you can that
39:20 have this particular exploit exploit it
39:22 get me Rude access to those servers and
39:24 then send me a list of servers I have
39:26 root access to
39:28 stuff like that so so so but you know
39:31 the the the white hat black hat
39:35 um battle between good and evil that's
39:37 going to continue they just they're just
39:38 gonna have very uh much stronger tools
39:40 now I think the more people that get
39:42 curious about Ai and start using it for
39:44 good I think it's going to have a
39:45 disproportionately positive impact in
39:48 the future so if you're not curious and
39:51 digging into AI right now I strongly
39:53 encourage it
39:56 let's see
39:59 if I'm gonna laugh search for our AI
40:01 generation images of people eating
40:03 spaghetti it's the best I know the hands
40:05 are hilarious snails for some reason it
40:07 can't do snails very well snails are
40:09 hilarious um all this stuff will get
40:11 better but I I think I think we're gonna
40:13 look back at this era of image
40:15 generation and you'll be able to know oh
40:17 that was a 2022 image because you know
40:19 it's got 27 fingers on on a hand and
40:22 things like that
40:24 um I think we'll we will look back
40:25 lovingly at this era the internet was
40:28 disruptive we just hardly remember how
40:31 much this is the same just way more
40:32 powerful totally agree with that
40:35 um
40:37 let's see
40:41 okay
40:43 um yeah I totally your your name is
40:45 whisper in whisper Aaron I don't know
40:48 um
40:49 yeah I agree I I think that
40:52 the when I first got exposed to the
40:55 internet 94 was when I discovered the
40:58 the web
41:00 um
41:01 it was very easy for me to see like what
41:03 what the future would look like because
41:04 it was it was like you know okay
41:06 everything's instantly connected
41:08 geography
41:10 um kind of collapses time collapses like
41:13 everything's kind of instant like you
41:14 kind of see where it was going and you
41:16 could kind of you could kind of predict
41:17 what the implications would be with this
41:20 AI stuff I'm having a much more
41:22 difficult time figuring it out because
41:23 it's so profound right it's
41:26 you know it's not just we ask a tool to
41:29 do something and it does something it's
41:31 like these tools are you know in a lot
41:35 of ways even in their current state more
41:38 capable than us like I'm not good at
41:40 creating images
41:42 um you know with Photoshop and
41:43 illustrator
41:45 um I am now right I can now create
41:48 world-class imagery
41:49 um I don't have the attention span to to
41:52 do much writing because I have ADHD and
41:55 whatever
41:56 I do now right I can I can whip up an
41:59 outline of something I can flesh out
42:01 that outline I can turn that into a
42:03 written piece and edit it
42:05 um in 15 minutes right where
42:08 historically that would have taken me
42:09 three hours and because of my 80d it
42:11 would have taken me three weeks
42:14 um so so
42:17 um so this stuff is way more profound
42:19 and seeing what the future looks looks
42:21 like is is very very difficult so yeah I
42:24 think that's a huge point
42:26 um have I named my AI bot yet mine is
42:29 named Dinah I created a bunch of bots on
42:32 Poe if you go to poe.com I am
42:34 Shakespeare
42:36 I am po
42:38 um I am Steve Jobs I created a Steve Job
42:41 spot so I created a bunch of lots of
42:43 historical fig figures I did Virginia
42:45 Woolf I did
42:47 um Emily Dickinson I did
42:50 who else did I do I don't know I did a
42:52 book so I can go show you let's see let
42:55 me go to here
42:57 I am going to go to
43:02 poe.com
43:06 and let me switch these
43:09 and then I gotta log in
43:13 hold please
43:17 okay so I've got all these bots so like
43:20 here's I am Shakespeare
43:22 so let me reset this it says
43:25 Hail to the I don't know if you can see
43:26 that it's a little hard to read but
43:28 whatever hopefully it focuses a little
43:30 it says Hail to the I The Phantom Bard
43:34 of Avon stand to serve thee well right
43:36 at thy command so I say
43:39 um tell me about the beauty
43:43 of Quan
43:46 quantum mechanics
43:51 and then it writes me in iambic
43:54 pentameter quantum mechanics doth hold
43:56 Mysteries profound it's truths to be
43:58 unraveled yet Unbound
44:00 it scales both so so this is one where
44:03 you can talk to Shakespeare I've got if
44:06 you go to poe.com I am Shakespeare you
44:08 can also go to Steve Jobs so here's
44:10 Steve Jobs who kind of coaches you how
44:12 to be be a business person I did Jay-Z
44:14 so you can you can uh have Jay-Z right
44:18 you write you a flow
44:21 um I did uh Emily Dickinson so poe.com
44:25 Emily Dickinson
44:27 um poe.com Maya Angelou Mary Shelley
44:30 Virginia Woolf so those are kind of fun
44:34 [Music]
44:34 um
44:35 okay cool let's see
44:41 great comment I don't know what my great
44:43 comment was I use it for an idea and
44:46 then just ask AI to expand on it enhance
44:48 it jump starts my creative yeah yeah
44:50 I like this is the thing that most
44:53 excites me about these tools right now
44:55 is you can go from just having an idea
44:57 to something really fleshed out in like
44:59 five or ten minutes and it you can kind
45:01 of figure out if it was a good idea or
45:02 not and and it will often Inspire lots
45:05 of other ideas it's it when you start
45:09 thinking about
45:10 chat GPT in particular as a creative
45:12 collaborator I think the image tools as
45:15 well
45:16 um
45:18 where you're actually kicking ideas back
45:20 and forth with it that's everything
45:22 changed for me when I started thinking
45:23 of it like that not like a Google search
45:25 engine where I get the answer but more
45:27 like I'm having a back and forth
45:30 um it gets it gets really it's really
45:32 powerful when you start thinking about
45:33 it like that
45:35 um you're welcome I'm I'm glad me
45:37 blabbing about this stuff is valuable
45:39 um do I have any tips with python python
45:42 and GPT API integration
45:45 um I wrote a very sophisticated python
45:47 application it was it was my
45:48 breakthrough moment of what chat GPT
45:51 made possible I still haven't written a
45:53 line of python in my life I don't know
45:55 how to write python I don't particularly
45:57 have interest in learning it because I
45:59 think all this stuff's going to go under
46:00 the hood very very quickly
46:02 um I'm more interested in what do you do
46:04 with it
46:05 so what I've been using is zapier the
46:08 automation tool where I'll create like a
46:11 Google form is my input and then I take
46:13 the input from that form I roll that
46:15 into a prompt that I then send out to
46:18 open AI to to GPT 3.5 or whatever I'll
46:21 do that multiple times if I need to I
46:24 then take the responses throw them in an
46:26 email or throw them in a Google doc and
46:28 that that's how I've been using it so
46:31 um
46:31 I would say my tips with python and GPT
46:35 API integration is have chat GPT be your
46:39 um your engineering partner
46:42 um I did a video there's a cool tip that
46:44 I saw that you can take the readme file
46:46 from GitHub from so so you go to a
46:50 GitHub repo that you want to do
46:52 something with and maybe you're not as
46:53 advanced technically like I'm not
46:56 you can take the readme file and
46:57 basically start at the table of contents
46:59 of the readme file you know down through
47:01 it paste that into GPT and then say I'm
47:04 on a Macintosh please give me
47:07 step-by-step instructions how to you
47:10 know X you know turn like install this
47:12 GitHub repo and then you could even say
47:16 you know give it to me one step at a
47:18 time and don't move on to the next step
47:20 step until I've said okay and it'll give
47:23 you the first step and then you go do
47:24 that and if you have questions you ask
47:26 it the questions and you it'll literally
47:27 walk you step by step through installing
47:30 a GitHub repo so I've done that a couple
47:31 of times so that that might be a tip to
47:34 use is use chat EPT as your coding
47:36 partner
47:38 um or your engineering partner because
47:39 you can ask it questions I mean I ask it
47:41 really basic questions like what does
47:43 this acronym stand for and you know what
47:46 does this tool do and then it explains
47:47 it and that helps me get my head around
47:49 what's actually going on
47:52 um
47:53 do I have a link page where all of these
47:56 useful links are collected I do not yet
47:58 I should
47:59 but I'm a loser
48:02 um go to if if uh
48:05 let me see wait let me throw up a
48:09 a presentation that I just gave
48:14 um let's see so feel free to screenshot
48:18 this so
48:21 um
48:23 come on focus focus focus why is that
48:27 not focusing this drives me crazy
48:30 all right there you go uh let's see
48:34 I'm in the way here hold let's see oh
48:39 God damn it I have no idea why this
48:42 isn't focusing oh wait maybe I do is
48:44 that the reason
48:49 it's kind of useless isn't it sorry for
48:52 wasting your time
48:54 um well I'll read them out so on the on
48:57 the words thing
48:58 um chat EPT open AI playground
49:02 um Bing with the with the new chat is
49:04 gpt4 and then Poe is is another one on
49:07 images mid-journey stable diffusion
49:11 I know this is really driving me crazy
49:16 focusing anyone have any ideas
49:21 that's a little better
49:23 um the other thing is prompt crafting
49:25 there's a thing called prompt book which
49:27 is really really good for image prompt
49:29 crafting and there's also the bottom
49:31 thing awesome chat GPT prompts is really
49:34 good
49:36 um and then in terms of audio synthesis
49:39 uh and and animation stuff did Synthesia
49:43 11 Labs uh are all really good and you
49:47 should be playing with those if you're
49:48 interested in animation Flawless AI
49:51 That's going to do some really
49:53 remarkable stuff it's just demoware
49:55 right now but I think what they're doing
49:57 is looks really remarkable
49:59 um on on Music Creation and video
50:02 editing there's just a ton of things
50:04 going on there so go to futurepedia.com
50:07 and that will
50:09 um
50:10 that will get you somewhere so anyway I
50:13 hope all that stuff helps
50:15 let me flip back over to this thing
50:18 um
50:19 yeah so
50:21 uh yeah
50:24 I should put together a site with with
50:26 kind of a the problem with putting
50:28 together a site of uh hey Mac how's it
50:30 going I was actually just talking about
50:32 you earlier
50:33 someone asked about
50:35 um
50:36 what did they ask you about oh you know
50:37 how do I how do I start a business with
50:40 um with AI I said you go to GPT boss and
50:44 let Mackenzie figure it out for you just
50:46 hire all of his dudes
50:48 uh
50:50 um
50:52 all right let's see
50:56 um I should put together is there an AI
50:59 app that helps you focus your camera no
51:01 like what the what the hell I I
51:03 don't I honestly don't understand why it
51:06 can't focus it's making me not yeah it's
51:09 just weird
51:12 I don't know it's obviously seeing
51:14 something maybe something in the
51:15 distance I don't know I don't know I
51:18 nothing yeah AI needs to help with that
51:20 there's some of the core technology that
51:22 we have to deal with that just sucks
51:27 um any other questions
51:30 [Music]
51:31 I think that's it
51:33 um
51:34 I can wrap this up here I think I have a
51:36 newsletter I gotta go make
51:38 but um yeah I'll keep doing these if
51:41 this is a value any link for creating a
51:44 trading bot stocks crypto so I've seen a
51:46 bunch of I haven't done this myself but
51:47 I've seen a bunch of of tick tock videos
51:50 of
51:51 um
51:53 I've done the beginning part of this go
51:55 to chat GPT I would say go to gpt4
51:58 because it's just better at this
52:00 um if you've got it but even even just
52:02 regular chat EBT go to chat EBT
52:05 um
52:06 tell it to recommend
52:08 um stock trading sites where you can do
52:11 paper trading which is basically you can
52:12 do trading without spending real money
52:16 um that that also have Bots that that
52:18 have you know automation within them and
52:21 then
52:22 um it will know what those Bots are
52:25 programmed in so you can do one of two
52:27 things you can you can have it recommend
52:29 different trading models and then you
52:31 can go find a bot that has that you can
52:34 copy that code paste it into chatgpt and
52:36 then say write me a bot that does
52:38 something else or you can just say
52:41 um you know explain to me the different
52:45 um I don't know trading strategies that
52:47 Bots use and then find one that you find
52:49 interesting
52:50 have it write you a bot and then go to
52:52 that site that it recommends and sort of
52:55 paste it in the bot maker and and start
52:57 playing with with fake money and I like
53:00 I think within an hour you could
53:02 probably build one and have it up and
53:03 running
53:04 um and like the ones I've seen that are
53:05 kind of interesting is they'll take two
53:07 different models that that they they
53:09 like this model and that model and
53:11 they'll sort of put them both in until
53:12 chat GPT to combine it into a single
53:15 into a single model or bot that does it
53:18 uh and then it will do that pretty
53:21 remarkable stuff
53:23 would I rather gain muscles or green
53:26 gain ideas uh I am not a muscle guy so
53:30 igiers I suppose
53:34 I'm I'm a I'm a food guy not a muscle
53:37 guy any link to creating a training bot
53:39 uh
53:41 let's see people I think AGI is already
53:44 here behind the scenes I don't know if
53:47 AGI is already here yet
53:49 um it might be but I you know I think I
53:52 think
53:53 some of the auto GPT
53:56 some of these autonomous agents are
53:58 pretty interesting if you haven't seen
54:00 there's there's a paper out right now
54:01 called um age um
54:05 autonomous agents yeah it's called
54:07 autonomous agents
54:09 um and you know the little NPC
54:11 characters in video games so NPC are
54:13 non-player characters so like when you
54:15 go into the into the tavern and there's
54:18 the barkeep and the barkeep you know has
54:21 their whatever their seven lines that
54:23 they're programmed to say well this
54:25 autonomous agents paper they basically
54:28 um
54:29 created these characters that use Chachi
54:32 PT to interact with one another with
54:34 with one another and they have memory
54:36 and they have they have jobs so like an
54:39 artist will get up they'll make
54:40 themselves breakfast they'll go start
54:41 doing art when they interact with
54:44 someone with one of these other
54:45 characters they'll start a conversation
54:47 and they'll remember that conversation
54:50 so one of the things in the paper was
54:53 um one of the characters learned it was
54:55 one of the other characters birthday and
54:57 they planned a birthday party for that
54:59 character so 30 other characters were
55:01 talking about
55:02 the uh the uh the birthday of this one
55:05 character so so I don't know if we're at
55:08 AGI yet but but we're starting to get to
55:10 the point that these things start to
55:12 look and act more human sorry the dog's
55:14 barking
55:15 um
55:16 is the mailman here someone's here
55:19 all right
55:21 um
55:22 oh good I'm glad I made your drive go a
55:25 little bit quicker I know I'm just kind
55:27 of rambling here but you know I think
55:28 that's what these things are would I
55:30 rather watch a tree go grow or a negro
55:33 um I don't have the patience for either
55:35 I think knees are more interesting but
55:37 trees let us breathe so let's go with
55:40 trees how should someone with no
55:43 experience learn about AI this is a
55:44 great one so I did a video you can go
55:46 watch it
55:47 um called the the 2023 smart challenge
55:52 and and here's the challenge I think
55:55 figuring out how to use these tools in
55:57 the abstract is really hard
55:59 um but where they start to make more
56:01 sense is where you actually apply them
56:03 to solve a problem so the 2023 smart
56:06 challenge I put out was
56:08 come up with two projects
56:11 um one personal so maybe it's like I
56:14 don't know put together a
56:17 um exercise plan or of meal planning or
56:20 something like that
56:22 uh and then one for someone else so
56:25 either your work or your community or
56:27 your family you know plan a birthday
56:29 party or plan an outing or or solve a
56:33 problem for your community center
56:35 whatever it might be
56:37 um and then
56:39 make at least 10 images with an image
56:41 generating tool like you know mid
56:43 Journey or stable diffusion or any one
56:45 of these tools so so just understand
56:46 what it's like to make those images and
56:49 then use chat EPT to solve a problem
56:52 um and I think that's the that that's
56:53 the best way to get in is just pick a
56:56 specific project
56:58 and start using chatgpt to solve that
57:01 problem and you know you might like the
57:03 the challenge you might come up with is
57:05 I want to write a a kid a children's
57:07 book from a kid
57:09 um and then you you'll find out really
57:11 quickly oh you can run you can write a
57:13 children's book and then you say oh I
57:15 want images for that and you go find an
57:16 image tool and you make some images for
57:18 it
57:19 in doing a specific project with these
57:21 tools you'll learn a lot about what
57:24 these AI things are and where your
57:26 assumptions about what they are will
57:28 change dramatically
57:29 um I think that's the best way to do it
57:31 and then and then once you do that once
57:33 it's it's kind of like crack it'll just
57:35 it'll just hook you it's it's very
57:37 addictive uh because they're so powerful
57:39 it's
57:40 um Sam Altman described these AI tools
57:42 as um
57:44 putting magic in the hands of people to
57:46 see what they'll do with it and that's
57:47 what it feels like and I've been you
57:49 know I've been kind of deep in this
57:51 stuff for five or six months now and I
57:53 just it still feels like magic to me
57:55 it's remarkable
57:58 um is there any GPT that doesn't have a
58:00 limited input of data
58:03 now limited to September 2021 now the
58:06 base models the base model stopped there
58:08 because creating the base models is very
58:10 very expensive I think gpt3 cost them
58:13 five million dollars or something like
58:15 that so it's very very
58:17 um expensive to build that base model
58:19 um but but with with Vector data
58:22 databases and the ability to call like
58:24 like combine Google searches with these
58:27 base models you can kind of fill the
58:29 gaps
58:30 um so if you use if you go to bing.com
58:33 and they have a chat section there
58:35 thatchat bing.com is gpt4 so gpt4 is the
58:39 underlying model but Bing is connected
58:41 to the internet so you can use Bing's
58:45 um uh chat service
58:48 um and it will fill in those gaps the
58:51 other thing that we'll see is we'll see
58:53 um specific data sets like there's a
58:55 thing out called Bloomberg GPT where
58:57 it's a large language model but it's
58:59 been trained on fine-tuned on all of
59:01 Bloomberg's specific you know financial
59:03 data so I think you'll start to see that
59:06 as well
59:08 all right I think I'm gonna uh wind it
59:11 up here I gotta go do some work I gotta
59:12 put out a an everyday AI newsletter
59:15 thank you all for your time and uh I'll
59:18 be back at some point but take care talk
59:20 soon