aiLL June 9, 2023 - 1 - The Accelerating Pace of AI Innovation

Description

Chapters

Transcript

0:26 hello hello this is Kyle Shannon with
0:28 the AI learning lab
0:30 happy Friday everyone
0:33 the nuggets and the heat are kicking off
0:35 their game nuggets in the lead
0:39 they're up two games to one I know
0:41 that's not AI but you know it's on the
0:43 iPad next to me so I should talk about
0:45 it uh if you're new here my name is Kyle
0:47 Shannon I'm a long time entrepreneur
0:50 I was an innovator in the early days of
0:52 the web and I see a lot of parallels
0:54 between what's going on with AI right
0:56 now and what was going on back then and
0:58 so
0:59 this channel exists as a way for me to
1:01 just kind of process what the hell is
1:03 going on and teach people as much as I
1:05 can even though I'm learning like I'm
1:08 basically teaching what I'm learning I'm
1:11 teaching on the fly so um yeah so
1:15 uh feel free to ask any questions in the
1:18 uh what's it called comments ask
1:21 questions in the comments I'll do what I
1:23 can to answer my boys need a ride to
1:25 Boulder in about an hour so I'll
1:27 probably just bolt out of here
1:30 and uh what else
1:32 um
1:33 a lot of interesting stuff went on uh
1:35 this week Adobe
1:38 um Adobe announced Firefly for
1:41 Enterprise so they're basically taking
1:44 their generative AI their their
1:47 generative AI for images and making it
1:49 safe for corporations which I think is a
1:51 really big deal I feel that many people
1:53 are learning on the fly yeah I I think
1:56 that's the that's one of the parallels I
1:57 see to the early days of the web is that
2:00 when I when I first started learning
2:02 about it I'm like I don't know anything
2:03 about this and
2:05 I started this group called The World
2:06 Wide Web artist Consortium as a as a
2:09 Shameless attempt to surround myself
2:10 with people that knew more than I did
2:12 and I learned exactly that that everyone
2:14 was trying to figure it out and so
2:16 there's a very similar thing going on
2:18 with with generative AI right now
2:21 um yeah you're welcome for for
2:23 um me doing this I appreciate the I
2:25 appreciate the sentiment uh if you have
2:28 questions about uh AI pop them down
2:31 below if you're new here or you're new
2:33 to AI the reason I put these URLs here
2:35 these are kind of it's just like a
2:37 public service service announcement more
2:39 than anything I'm not attached to these
2:41 things in any way but
2:42 check that openai.com that's the
2:44 official chat GPT website so if you've
2:47 heard of chat GPT but you haven't played
2:48 with it do yourself a favor and just go
2:50 play with it and see what it can do have
2:52 it ask it to write you like this briefly
2:55 describe your pet or your spouse or a
2:58 kid
2:59 and have it write you a song or a poem
3:01 or a short story or something like that
3:03 and just
3:05 sit back and go wait it can do that
3:09 and then Bing is just like chat gbt but
3:12 it's on Microsoft it's you know it's
3:14 it's Microsoft's Bing it's gpt4 you can
3:17 use it for free and pose another one
3:18 where you can where you can do some chat
3:20 GPT stuff
3:22 I began my career over 35 years ago and
3:25 you're right the parallels are very
3:26 interesting I mean
3:28 some interesting things about what's
3:29 what's happening with
3:31 um AI compared to the early days of the
3:33 web
3:34 it took six years to get to 100 million
3:37 users of the web it took six weeks to
3:40 get to 100 million users of chat gbt
3:42 that's that's a stat that just blows my
3:45 mind and it was a piece of software that
3:46 was written in two weeks Sam Altman just
3:50 said to his engineering his engineering
3:52 team go build that thing that was in our
3:53 documentation the the in the
3:56 documentation of gpt3 were instructions
3:58 for GP for chat gbt and no one built it
4:01 so Sam Altman just told his Engineers to
4:03 go build it in two weeks
4:05 um unfortunately that isn't a hot link
4:08 no you actually have to type
4:10 chat.openai.com I don't know how to make
4:12 hot links if I can make hot links in
4:14 here teach me how I'll do it
4:16 um oh there's a lot of people in here uh
4:17 my name is Kyle Shannon this is the AI
4:19 learning lab I'm an entrepreneur for
4:22 many many years and uh I'm currently the
4:25 co-founder and CEO of a company called
4:26 storyvine we're an 11 year old video
4:29 technology a storytelling platform and
4:32 uh we do automated video production but
4:34 we don't use any AI for that but we are
4:36 using uh open ai's whisper for
4:39 transcriptions and now I'm creating a
4:41 bunch of content generation engines
4:43 based on the videos that are in our
4:45 system so that's pretty exciting and uh
4:49 I'm here to answer any questions I can
4:51 so if you've got any questions pop them
4:53 down oh
4:56 yeah show me how to do that Arthur that
4:58 would be awesome
4:59 all right isn't that how we all survived
5:02 by Consulting each other yeah that was
5:04 pretty much it those those are those
5:06 early days of the web were just amazing
5:08 uh and I I heard like it sounds like the
5:10 a similar things going on in San
5:12 Francisco right now I saw I saw a tweet
5:14 today that said there's there's like 44
5:18 meetings in San Francisco this week it's
5:20 like three meetings a day and that's not
5:22 even including the private ones
5:24 um I live in Denver it's not quite so
5:26 happen but I did start the AI salon so
5:29 if you go to my profile
5:31 um there's a link tree there and you you
5:33 can see a link to the salon so we do
5:35 both in-person meetings but also virtual
5:37 meetings we meet every other week
5:40 um and uh we have guests guest speakers
5:42 come in and things like that
5:43 so yeah
5:46 all right can I explain how generative
5:48 AI difference from classic AI
5:52 I don't know a ton about classic AI but
5:55 what I I ca so so here's what I can tell
5:58 you
5:59 there's
6:00 in 2017 Google puts out a paper called
6:03 attention is all you need and they
6:05 introduced this idea of a piece of
6:08 technology called the Transformer
6:10 and prior to that
6:13 um
6:14 machine learning you had to do it on
6:17 relatively small data sets and it was it
6:19 was very much the purview of you know
6:21 data scientists and researchers and
6:22 things like that so you would sort of
6:24 bring your data set I got a bunch of
6:26 medical data I got a bunch of census
6:28 data or whatever it might be and then
6:30 you train these these uh machine
6:32 learning models on your little data set
6:35 the Transformer comes along and
6:37 ostensibly what it does is it allows you
6:40 to train these machine learning models
6:41 on massive data sets
6:44 so the ter the the acronym GPT chat GPT
6:49 stands for generative pre-trained
6:51 Transformer so the Transformer the T in
6:54 GPT is Transformer so that's the
6:56 technology generative means it's
6:59 generating
7:00 generating content right it's generating
7:02 original content it's not copying and
7:04 pasting it's not stealing from other
7:06 people's writings it's actually
7:07 generating and then pre-trained is that
7:09 idea that it's been trained you don't
7:11 have to bring your own data they've they
7:13 brought the data for you open AI spent
7:16 years tagging and identifying data
7:19 sources so like all over the public
7:21 internet they found as many data sources
7:23 as they could Wikipedia and blogs and
7:26 websites and
7:27 you know research
7:29 things and universities as much data as
7:31 they could and they got it all you know
7:33 in in this trainable format and that's
7:35 why if you play with chat GPT you learn
7:39 really quickly it it's knowledge stops
7:42 uh in September of 2021. why is that
7:45 because it took a really long time to
7:47 train these things
7:48 so the big difference is that we as mere
7:51 mortals don't have to bring our own data
7:53 set it's been pre-trained and we you
7:56 know chat GPT turned it into a just a
7:58 simple little it's just as simple as a
8:00 Google you know chat box but when you
8:03 type in something simple you get back
8:05 something remarkable so so
8:08 the way I like to think of it is you're
8:10 effectively tapping into a massive
8:12 amount of the output of humanity
8:15 without having to learn it it knows how
8:17 to program it knows different languages
8:20 it knows
8:22 how to write it knows how to it's just
8:26 crazy so that's that's kind of the the
8:28 big difference is is the fact that you
8:30 don't have to bring your own data it's
8:31 not BYOD
8:33 [Laughter]
8:41 all right let's see
8:44 let's do this offline because I have a
8:47 teleconference to attend okay yeah
8:49 that's cool Arthur just just DM me in
8:51 the DMS and and we'll hook up and you
8:53 can teach me how to put hot links in my
8:54 in my lives I figured out how to do the
8:57 dual camera thing that's pretty
8:58 impressive people ask me about that all
9:00 the time let me get that out of my face
9:03 um
9:04 uh and welcome to contact you cool have
9:07 a good night and presentation thanks
9:09 again thank you
9:10 all right if you have questions about
9:12 generative AI pop them in the um uh in
9:16 the in the thing below tap the screen uh
9:19 if you if you you know to let people
9:21 know we're here or share this live
9:22 anything you want to to get to get
9:25 things out there
9:27 um I'm trying to think what else is is
9:29 worth talking about today
9:32 I saw a cool piece of technology that
9:34 tracks elements in a video but that's
9:36 kind of geeky and not all that
9:38 interesting
9:39 um
9:40 let's see oh well the the nuggets are
9:43 playing and Trump got indicted so so
9:46 there's that happening it's been a busy
9:48 day it's been a busy day at newswise but
9:49 that's not what this channel is about
9:50 there's plenty other channels for that
9:52 stuff
9:54 um if if you um if you have questions
9:57 about uh Chachi PT pop them down below I
10:00 can also show you
10:03 um if you're new to chat gbt I'll show
10:05 you a couple of uh you kind of have your
10:07 chat history down the left and then you
10:10 have this big dark window
10:12 this I guess the dark window represents
10:14 you know all of the possibility
10:17 um and then uh you can just type in uh
10:20 explain quantum
10:24 physics
10:25 as an m in M rap
10:30 Bang
10:32 then it will write you a wrap like it's
10:35 fast that's gpt3 gpt3 is much faster
10:38 than uh gpt4 but gpt4 is more capable
10:43 okay let's see metadata I don't know
10:47 what that's a question to
10:49 um what's the next big ai ai idea
10:53 you would do if you had lots of money
10:55 that's a good question
10:59 I think I would do I actually had an
11:03 idea for this today
11:04 I would do something around authenticity
11:06 I think the 2024
11:09 I think the 2024 election is going to be
11:12 if you thought 2016 was a show in
11:16 2020 was a show
11:19 um I think the 2024 election is going to
11:21 be
11:23 um the most and most sophisticated
11:25 misinformation and disinformation and
11:28 deep fakes we've ever seen it's going to
11:29 be absolutely insane
11:31 and I think there's going to be a
11:33 backlash to that in in 2025 2026 where
11:37 there's a real demand and there's a real
11:40 need there's a there's an actual need
11:42 for being able to validate the
11:44 authenticity of content so my company
11:46 storyvine we have people film videos of
11:50 themselves talking about whatever story
11:52 we designed we create a little story
11:53 framework our app guides them through
11:56 telling that story it goes up to the
11:57 cloud five minutes later you got a fully
11:59 edited video super cool but it's real
12:01 people telling real stories about their
12:03 lives
12:05 two years from now
12:07 there will be videos out there that
12:10 don't look any different from our you
12:12 know self-produced
12:14 um phone videos right so so how do you
12:17 authenticate you know is this human is
12:19 it is it synthesized is it a real story
12:21 but a synthesized person so I think I
12:23 would do I would do something around
12:25 around
12:26 um authenticity because I just think
12:28 it's a major problem major major problem
12:32 um who would you bet in the lead oh who
12:35 would I bet in to lead large language
12:38 models by 2025 that's not too far out
12:41 Google Microsoft or other
12:43 uh
12:45 I'm gonna just roll Microsoft and openai
12:48 together because they're ostensibly you
12:51 know they're rolling together although
12:52 open AI is free to compete with them but
12:54 I think I think they'll be pretty tight
12:56 at the hip at least by for 2025
13:00 um
13:01 Google is way behind
13:03 um they also but they have a lot of
13:04 money and they have a lot of power so I
13:07 wouldn't put and they have a lot of
13:08 smart people but they're really heavy on
13:10 engineering and not so heavy on the
13:12 human factor side and the the empathy
13:15 side
13:16 um I think open AI is much better at
13:18 that
13:20 um I think anthropic is going to be a
13:23 major player
13:25 um
13:26 meta while they're behind are doing some
13:28 interesting like they're open
13:30 sourcing a bunch of stuff and they're
13:31 putting stuff out there and I think one
13:33 of the advantages of open open sourcing
13:35 things in the AI space right now is that
13:38 because chat GPS like you know exploded
13:41 and and caught fire
13:44 um
13:45 there's so many developers in the space
13:47 I think by open sourcing your technology
13:50 you basically get free development
13:53 um which you can then just grab yourself
13:55 because it's open sourced right so
13:57 I would
13:59 I wouldn't put meta out but there I'd
14:02 say I'd say they're more behind than
14:04 Google
14:06 and Google's way behind Microsoft and
14:08 open AI so for for me right now open Ai
14:12 and and Microsoft are going to be in the
14:14 lead
14:15 I would say anthropic is probably going
14:18 to be there so you know I don't know if
14:20 you remembered a few weeks ago there
14:23 were four CEOs that were invited to the
14:24 White House to talk about AI it was the
14:27 CEO of Microsoft the CEO of openai Sam
14:30 Altman the CEO of Google and who was the
14:33 fourth the CEO of anthropic so anthropic
14:36 is a handful of X open AI employees that
14:41 started a competing company
14:43 um and they're really good they've got a
14:45 really interesting model so
14:47 that's my prediction I'm not a psychic
14:51 but I play one on the internet
14:53 make money
14:56 okay
14:58 oh and I'm also an idiot sometimes
15:00 that's just part of the King movies yes
15:02 dead people will be making so so
15:06 here's a really interesting thought
15:10 um all of the popular people today and
15:13 and you know in the past that we've got
15:15 videotape and audio tape of
15:18 um we're gonna be able to synthesize all
15:20 them like you there you know a month or
15:22 so back there was that song that came
15:23 out by Drake in the weekend that wasn't
15:25 actually Drake in the weekend
15:28 um and then a week after that
15:30 um uh Grimes Elon musk's ex-wife
15:33 um said hey steal them you know like
15:36 like synthesize me all you want just
15:37 give me half
15:39 um
15:40 so I think I think it's there's just
15:43 going to be a massive new industry this
15:44 actually might be something I'd go into
15:46 it's just there's going to be a massive
15:48 industry to to take the like license the
15:51 likenesses of famous people and you'll
15:54 just be able to put you want young Elvis
15:56 to Star in Star Wars okay it's something
15:59 you'll just go I want to watch Star Wars
16:01 where young Elvis is Luke Skywalker
16:03 it'll start playing and he'll be in it
16:06 and he'll shake his his uh his hunky ass
16:11 um so yeah and but so the thing that I
16:13 think is interesting is like like okay
16:16 that's gonna happen for all the famous
16:17 people now but but if we basically just
16:20 repackage and re-synthesize and all the
16:22 all of our favorite people in the past
16:24 like how do new people get popular so
16:26 I'm wondering if there's going to be
16:27 this like what's the battle between you
16:30 know how do I make it in the world if
16:32 I'm always competing against the Beatles
16:33 in their Prime you know I don't know I
16:36 don't know the answer to that
16:39 crazy it's crazy I loved you in Modern
16:42 Family yeah it's the second Modern
16:44 Family I joke at least that wasn't a
16:46 Rosie O'Donnell I got you look like
16:48 Rosie O'Donnell the other day and uh fat
16:50 Alec Baldwin
16:52 um if you have questions about uh
16:54 generative AI pop them down below what I
16:57 like about my trolls on on my lives is
16:59 that they're actually they're
17:00 semi-complementary you know I mean sure
17:03 he's chubby but he's funny on Modern
17:05 Family
17:06 mm-hmm
17:08 oh I don't have water jeez now dang it
17:11 [Laughter]
17:13 all right I'm not gonna I'm not gonna be
17:15 here for too long
17:17 um oh so let me let me show I was
17:20 showing you before about
17:21 um chat chat tvt you're gonna make money
17:23 I wanna chat you in my day
17:26 um
17:27 so if you pay for GPT plus which is the
17:31 little yellow flag there where did it go
17:33 okay it's back
17:35 um then you get access to gpt4 and you
17:39 can enable you turn on your beta
17:40 features and you can enable browse with
17:42 Bing and you can enable plugins so
17:45 browse with Bing is basically GPT
17:47 connected to the internet
17:49 and then plugins are these plugins
17:52 they're they're like little apps that
17:54 you
17:55 turn on in a session you can turn on up
17:58 to three of them at a time and what
18:00 you're essentially doing is creating a
18:02 little mini application so
18:05 um like there's this this one's video
18:07 insights there's web pilot and Link
18:09 reader those will read websites as well
18:12 just the normal
18:14 um the gpt4 connected to the internet
18:19 um but like this x papers one that'll go
18:21 out to the archive site where all the
18:23 white papers are and I can I can create
18:26 a list of actually let's go do that so
18:28 I'm going to turn on this one I'm going
18:30 to turn off wolf from wolf from Alpha
18:32 that's if you want to do math so I've
18:34 got
18:35 um what's it called X papers turned on
18:38 so I'm going to say
18:40 um
18:41 make me a table of white papers
18:47 about ML and AI that were released this
18:54 week and rank each of them
19:02 in order of
19:07 potential impact I don't know let's see
19:11 what it does so now we hit that
19:14 and it's going to figure out that I need
19:15 to go get some white papers and it
19:17 should fire off that plug-in so it fires
19:19 off the plug-in
19:20 and then we can we can open the plug-in
19:22 there's a little down you know little
19:24 carrot here we can drop it open so it
19:27 says it's querying machine learning or
19:30 artificial intelligence Max results 50.
19:33 start date end date so it automatically
19:35 so it's automatic so what the plugins
19:37 are doing is they're basically just
19:38 coding based on your input
19:42 it says apparently there are no papers
19:44 related to machine learning artificial
19:46 intelligence that were released
19:48 I don't buy that
19:51 um let's see
19:55 um was