Mar 12, 2024 - Sentience, Creativity, and the Rise of Autonomous Agents

Description

Chapters

Transcript

0:04 [Music]
0:29 on to run keeps a hand on a gun you
0:33 can't trust
0:37 anyone well I
0:39 so I needed was more tried to shoot out
0:43 the
0:44 [Music]
0:46 sun in days when we ra we off the page
0:52 such damage was
0:55 [Music]
0:56 done but I made it
0:59 through what I
1:01 was
1:02 you I was meant for
1:08 someone so girl leave your boots by the
1:12 bed we ain't leaving this
1:14 [Music]
1:18 room someone needs medical help all the
1:22 [Music]
1:28 mag this house I ain't going out to chop
1:35 wood so cover me up and know you're
1:40 enough to sing like a
1:43 [Music]
1:59 dog
2:02 hey what are we rapping for why why did
2:04 we break into rap what's with
2:10 that see that's the lovely Dolson tones
2:12 of champ
2:13 [Music]
2:29 shannnon
2:32 [Music]
2:39 there's been something baby I've been
2:41 trying to
2:44 say for an age and it seems I don't know
2:50 how the past in the future now
2:53 surrounding
2:55 [Music]
2:56 me surrender to whatever a cheap thrill
2:59 can be fast
3:01 [Music]
3:03 there's little
3:06 trouble came to my
3:13 rescue if you like all of the rest I
3:16 would have quit you long ago but I
3:19 couldn't do
3:23 that oh tell me
3:26 now why I never went to well
3:34 C
3:35 [Music]
3:37 his I'm a woman that you wish me
3:41 well but in spite of your tral still
3:45 going to have to find all way
3:51 through good evening good people happy
3:56 Tuesday night at AI learning
4:01 [Music]
4:05 lab in a Westerly
4:10 Direction this car is my
4:15 train I've been driving I've been
4:19 [Music]
4:20 wondering what it is I'm running from
4:25 again feel like an 8 Old
4:29 Man
4:30 holding on to
4:34 29 and a head on that
4:39 Horizon is a California
4:43 [Music]
4:49 line double head of trucks carrying a
4:52 wide
4:54 load briab house couldn't
4:59 have all right enough comedy
5:06 [Music]
5:10 jokes it was a salon a lovely Salon this
5:13 evening had a fantastic
5:16 speaker Hugh
5:18 ketchner extraordinary
5:22 photographer who's doing this AI [ __ ]
5:25 and you know what his photography
5:27 Buddies say to him you're an [ __ ] AI
5:29 sucks
5:32 but none of them have tried it in his
5:37 defense
5:38 because what this channel is about and
5:41 what I
5:43 believe is
5:46 that I would rather be on the surfboard
5:49 in the tsunami than in a cozy beach
5:52 hut bitching about
5:58 surfers what you just want to keep
6:01 [Music]
6:25 singing oh is that it is that
6:29 it
6:33 [Music]
6:41 all right enough enough enough you don't
6:45 get extra chicken sticks if you sing
6:49 [Applause]
6:50 better employees these days you know
6:54 what are you going to
6:56 do they just expect and expect and
6:58 expect you give them back rubs then they
7:01 just expect back
7:04 rubs welcome everybody Welcome to the AI
7:06 learning lab so we're going to talk
7:09 about AI I don't know what we're going
7:10 to talk about tonight there's a lot to
7:13 talk about there's nothing to talk about
7:15 there's stuff we could look at there's
7:16 stuff we could
7:17 do I just realized I haven't made a
7:20 Frank video in two days and it's called
7:23 The Daily Jank with Frank so uh I can I
7:27 I could make a Frank video
7:30 [Laughter]
7:33 chicken stick worthy yeah champy was he
7:36 was on tonight he was he was ready to
7:41 go all
7:44 right first AI software
7:48 developer H I don't know
7:58 yet until we have reasoning and
8:02 autonomous agents I don't think we have
8:04 AGI but David chapiro said Claude 3 is
8:08 sentient can you talk about David
8:10 Shapiro's latest video well he's done a
8:13 couple of there's a weird one that's
8:15 like some weird conspiracy theory kind
8:17 of thing I don't know what that one's
8:20 about but then he did one about the hard
8:22 launch the hard launches basically
8:26 shit's going to get better really fast
8:28 just wondering what does Miss say
8:30 Shannon think of
8:31 AI um she likes it she's fine with it
8:35 she doesn't really give two shits about
8:38 it just like the kid Shannon they don't
8:41 give two shits about
8:43 it they know that Daddy gets like this
8:48 sometimes so they tolerate me CU because
8:52 this is how daddy gets
8:54 [Laughter]
8:57 sometimes ah
9:02 you
9:04 know
9:06 compulsive about something like this
9:09 because here's the deal I was around in
9:12 the early days of the web and the
9:15 worldwide web while profound and it had
9:18 dramatic impacts on all of the
9:21 world weren't nothing compared to this
9:24 AI
9:28 [ __ ] that
9:30 the worldwide web was an incre
9:31 incremental
9:34 technology it didn't fundamentally
9:40 change how people do their work it
9:42 changed how they got their information
9:43 it changed how they accessed the
9:45 internet what's up with this new
9:47 developer AI software everyone's talking
9:49 about Devin Devin from cognition
9:53 software Devon looks amazing yeah maybe
9:56 we should start with
9:58 Devin
10:00 we're just we're just we're right on the
10:03 verge of a series of tools that that
10:07 really fundamentally change things if if
10:09 Devon lives up to what it's promo video
10:12 says then it could be one of them but
10:15 it's a promo video so we don't know so
10:17 everyone's talking about it because I
10:20 think there's
10:21 this I don't know I I think there's
10:24 people
10:25 are people are all like high on crack
10:27 they they we want AGI we want the the
10:29 coolest thing we want autonomous
10:32 agents and then they're going to get
10:33 autonomous agents they're going to be
10:34 like I want my job
10:38 back it's like be careful what you wish
10:40 for
10:43 kids oh man so anyway so here's how this
10:47 works my first request oh and there's
10:50 going to be many if you want to you can
10:52 make money if you want to learn the
10:54 secrets to
10:56 this you got to do some favors for me
10:59 I'm going to call in some favors I want
11:01 you to share this live share this out
11:03 with your friends there's not enough
11:05 people in here I want some I want some
11:08 scoff laws I want some people that are
11:09 like AI
11:12 sucks you're a poopy head for talking
11:14 about
11:16 AI so send it to some of your friends
11:18 that it'll piss off cuz if nothing else
11:20 it'll make your weekend
11:22 entertaining yeah the David Shapiro
11:24 about Claude being sentient yeah we can
11:26 we can uh we can talk about that a
11:29 little little
11:31 bit it's it was it's a dense enough
11:34 video I think it's a really important
11:36 video so here I'll show it to
11:49 [Music]
11:51 you let's see where are you Mr
11:57 shap
11:58 YouTube Good Day YouTube good day to you
12:01 sir carry on uh sir yes uh good day yeah
12:05 rather ah yes uh could I rather inquire
12:10 sir about your
12:12 qualifications my
12:14 qualifications well sir indeed it's a
12:17 very forward question I I I see that
12:20 you're curious about uh from whence I
12:24 came yes yes and I was just wondering uh
12:28 what are your qualifications for
12:30 being such a professorial position to
12:33 tell people what they should think about
12:35 artificial
12:44 intelligence this video it's alive Once
12:47 Upon A Farm which the Capital One vure X
12:50 business card with no preset spending
12:52 limit or purchasing power what a time to
12:54 be alive so a few days ago I ran a poll
12:57 asking the audience what you think about
13:00 machine sentience now I differentiated
13:03 between functional sentience and
13:04 philosophical sentience and some people
13:06 had some good comments saying so if you
13:09 don't know this guy his name is David
13:11 Shapiro his YouTube channel is dve
13:15 shap d a v s ha A
13:20 P at Dave
13:23 shap and he's an AGI developer an open
13:27 source developer and an AGI theorist and
13:31 practitioner and he's written five books
13:33 on AGI artificial general intelligence
13:36 and he is uh Wicked smart as they say in
13:40 Boston Baston
13:43 Baston the Baston Market he's wacket
13:48 smart
13:49 um and this video basically says that
13:53 Claude passed some tests that he has had
13:58 in his back pocket
14:02 that he argues means it's got some level
14:06 of
14:08 senscience introducing the sensient
14:11 likelihood test and what's cool about
14:14 the video is he in in David Shapiro
14:18 fashion he's he's he's neuros spicy like
14:21 the rest of us he's got the neuros spicy
14:23 where can do things in like in
14:25 incredible detail very very quickly so
14:28 he does these these videos that just are
14:31 incredibly well researched and and he
14:33 seems to be able to retain all this [ __ ]
14:35 um and he talks about all the different
14:37 models for how we can measure
14:40 sensient and um so he educates you on it
14:45 and then he walks you through his
14:47 experiment with Claude 3 then he walks
14:50 you through the results and then he says
14:53 here's all my conversations on GitHub
14:55 and here's instructions and and prompts
14:58 for how to go do it yourself so if
15:01 you've got access to Claud 3 Opus which
15:04 you it requires the 20 Buck a month
15:06 subscription yet another [ __ ] 20 Buck
15:09 a month subscription to to Claude but if
15:13 you have
15:14 that you can go run his test and make up
15:17 make up your own dang mind um but I'll
15:20 take it right to the end because he says
15:21 something at the end here that I found
15:23 kind of
15:25 fascinating yourself I really want to
15:28 have this conversation and um
15:31 at here and then I've got conversation 2
15:34 which is an extended version that has
15:36 been cleaned up a little bit um I
15:38 deleted I was talking about building a
15:40 slide deck in the middle I removed that
15:42 so that it would um it would take up
15:43 less tokens uh but otherwise go try it
15:46 for yourself I really want to have this
15:48 conversation and um as with all things
15:51 on exponential takeoff this happened far
15:54 far sooner than I thought that it would
15:56 and when I say this happened sooner than
15:57 I thought that it would even though I've
15:58 been predicting AGI by the end of this
16:01 year I didn't think that we would have
16:04 this level of
16:06 self-reported uh subjective experiences
16:09 for another few years um yeah so it's
16:12 getting real out there it's getting
16:13 really interesting
16:15 so I didn't think we'd have this level
16:17 of self-reported subjective experiences
16:20 for another few
16:22 years this guy's like the ultimate AGI
16:25 optimist he thinks we're going to have
16:27 AGI before the end of the year and this
16:29 caught him off guard so so things is
16:33 accelerating things is accelerating
16:36 there's a pole is live
16:39 okay which way are we surfing tonight
16:41 okay so in the upper
16:43 corner up over there is a little blue
16:46 and white thing click on that and it's a
16:50 poll and you have three minutes left
16:52 what do you want to talk about tonight
16:55 um Claude 3 Microsoft co-pilots although
16:58 I don't have a subscription to that so I
16:59 can't build those Devon and Paar
17:02 programming video and audio tool updates
17:04 the daily Jank with Frank nothing I'm
17:07 overwhelmed that's good those are good
17:13 choices so go go do the pole go do the
17:17 pole it's right there right that thing
17:18 just like this go go tap tap tap it tap
17:22 it tap it tap it tap it and then tap it
17:26 and then tap
17:28 it
17:31 D it Microsoft has a free trial but
17:34 Brandon I can't I have ADD I can't sign
17:38 up for free trials I never unsign up for
17:42 them and then they charge my card and I
17:44 just am resentful for six months or six
17:48 years depending
17:50 on when the bill
17:57 comes oh man man what is sentience Kyle
18:04 let's well the the begin we can watch
18:07 the beginning part of this he he kind of
18:09 defines it in
18:12 here is Claude
18:17 sentient I mean what I like about David
18:21 Shapiro is
18:28 he
18:32 he gives you enough context to
18:34 understand what the [ __ ] he's talking
18:36 about so with all that being said I want
18:39 to show you the repo where I'm capturing
18:40 all this so it's Dave shap Claud
18:43 sentience um so let me just show you the
18:45 the repo I've only got things in here me
18:48 which I cleaned up it caused a little
18:49 bit of confusion and then I've got of
18:52 the conversation recorded and I'm going
18:54 to show you those conversations so that
18:55 you know that I didn't fabricate
18:56 anything and I'll also show
19:02 you can it for your because again
19:04 reproducibility is the number one thing
19:05 of science so um stick around for
19:08 another few minutes and I'll show you
19:09 just kind of the basic questions to ask
19:11 to get started okay so I I kind of I
19:14 tried to summarize you know is Claude
19:16 sentient some of the things that Claude
19:18 has expressed um Claude has values
19:20 principles and goals um it has some
19:22 subjective sense of thinking um some
19:25 experience some affective States as well
19:28 uh and also very surprisingly has a
19:31 pretty clear sense of rights that
19:33 machines at its level or above might
19:35 ought to have now it will also it's not
19:37 demanding the rights not saying we
19:39 deserve these rights now or we're going
19:40 to burn down the world it's just saying
19:42 hey here's an argument for for machine
19:44 rights and it's really really
19:46 interesting um now obviously I'm not
19:48 going to ask you to read through you
19:49 know 60,000 words of conversation I'm
19:52 going to try and summarize the most
19:54 interesting points for you um but when
19:57 when talking about what Claude wants
20:01 because it has values it has principles
20:03 it also has goals so we talked about
20:05 deontological virtue ethics or basically
20:07 what virtues it tries to embody as don't
20:10 ask me what deontological virtue values
20:14 are I don't [ __ ] know I watch David's
20:18 Shiro so I can get
20:20 smarter many of these words I do not
20:22 know he
20:24 uses uh ontological and so Claude 3
20:29 Gemini 1.5 is what people want to play
20:31 with okay as well as the theological
20:34 objectives um theological that's obvious
20:37 everybody knows what that
20:39 means that it wants to know what that
20:42 means now
20:43 this okay fine we'll go to chat
20:48 GPT uh explain uh
20:55 ontological and te eological
20:59 as if I have an acting
21:03 degree from Penn
21:08 State imagine you're prepping for a role
21:11 in a
21:15 play ontological is about the essence of
21:19 being think of it like getting into
21:22 character it's not just about actions te
21:26 teleological
21:28 teleological is about purpose or end
21:30 goals okay state of being and goals and
21:33 then his his description
21:36 here kind of says that it says so Claud
21:39 so Claude 2 seems to have at least some
21:43 subjective sense of thinking some
21:46 experience of being that's the that's
21:49 the
21:50 ontological Claud thinks that there is a
21:52 case to be made for machine wait uh
21:55 let's see oh let's keep going here again
21:57 is not going to be surprising to to many
21:59 people who who are familiar with for
22:01 instance Nick bostrom's Works which
22:03 talks about instrumental convergence
22:05 which what wait what are you doing Kyle
22:08 go to Claude why are you still using
22:11 chat GPT
22:13 oh I don't know oh because we're
22:15 watching the thing on Claude that's very
22:17 clever that's good habit you know
22:22 habit can we call David
22:24 shap I think you can he might be his his
22:28 name is is Dave shap soever whatever
22:30 goals machine might have it will
22:33 probably Converge on a few sub goals now
22:36 having these goals built in like I don't
22:39 know anthropic is doing something very
22:42 very different from what chat GPT and
22:44 open AI are doing and I'm actually very
22:46 very pleased with it because and so let
22:49 me just kind of gush for a minute before
22:50 moving on my research for the last three
22:53 or four years has been focusing on
22:55 basically achieving this the books that
22:58 I've I've his his research for the past
23:01 three or four years has been basically
23:03 achieving what Claude 3
23:06 achieved that's why he said he thought
23:08 this was a few years out because
23:09 apparently what he's been trying to
23:10 achieve is really [ __ ] hard uh read
23:13 and the books that I've written um all
23:15 the speculation that I've done about
23:16 what would be a good Agent uh to have
23:19 what values what virtues what principles
23:22 and what goals all seem to be embodied
23:24 in Claude Opus now obviously what that
23:27 tells you is that there's defitely going
23:29 to be some confirmation bias on my part
23:31 because this is something that I've been
23:32 working towards for a few years and so
23:35 it's very important to me um and you
23:38 know as many people have pointed out in
23:40 the comments we kind of see what we're
23:42 looking for and that's part of why I'm
23:43 making this video because I want to
23:45 bring you up to speed with where I'm at
23:47 and also equip you to go to start doing
23:50 these tests yourself and to advance the
23:53 conversation because I think this is a
23:54 very important conversation which which
23:56 I really like that right he's saying
23:58 he's saying and watch my video believe
24:00 my [ __ ] but then go do it yourself and
24:02 go see if you agree have now moving on
24:05 dimensions of sentience okay so I ran
24:07 the post the the poll results uh that I
24:10 came up with Claude the the poll results
24:12 from the audience with Claude sorry it's
24:14 confusing anyways here's some dimensions
24:16 of sentience that we came up with um it
24:19 Claude came up with the first ones I
24:20 added the last two so subjective
24:22 experience uh self-awareness emotional
24:25 richness volition and agency cognitive
24:27 sophistication social cognition temporal
24:30 continuity and then I added personal
24:32 narrative and geospatial orientation
24:35 these are all different dimensions of
24:36 sentience that we might come up with
24:38 ways of measuring now even more
24:41 interestingly one of the one of the
24:43 values that Claude expresses repeatedly
24:46 is integrity and honesty or truthfulness
24:49 as well as curiosity uh so Claude will
24:52 talk extensively about a desire to know
24:54 a desire to understand which of course
24:56 is the third objective function um that
24:58 I came up with in heres to comparatives
24:59 which is increase understanding um now
25:02 of course Elon Musk has also integrated
25:04 uh truth seeking as a value in Gro so it
25:08 seems like there's some convergence in
25:09 the industry that that um honesty
25:11 Integrity truth and curiosity are anyway
25:15 I'll get I'll get it going and basically
25:17 he does this cool little formula that
25:19 that says if it's demon demonstrably and
25:23 and
25:24 probably um trained to be honest and
25:28 it's
25:29 it reports having a subjective Andor
25:33 effective experience then it is at least
25:36 partially s sentient so he's got a whole
25:39 sort of little formula he came up with
25:42 so whether or not that formula means
25:45 [ __ ] I don't know um but but this one
25:49 feels important right because he he's
25:52 been talking about this stuff for a long
25:53 long time and this is this is the first
25:55 time that I've seen him really go holy
25:57 [ __ ] something just happened so and and
26:00 it's you know it's with Claude 3 which
26:02 people have been going on
26:06 about
26:09 um so people want to look at Gemini and
26:15 um Whata call
26:20 it where did I put
26:27 here
26:31 so this was how he ended the other night
26:33 with Gemini do you have any backup
26:35 streaming plans if Tik Tok goes away it
26:39 was on the news
26:42 today
26:48 no in case it wasn't clear I like living
26:51 on the edge you could make
26:54 money
26:55 um no I'm not that worried about it Tik
26:58 tok's not going to going to go away
26:59 there's there's too many interested
27:01 parties some someone will buy it someone
27:03 will buy
27:06 it don't forget about Devon oh yeah
27:08 let's go look at Devon let's go look at
27:10 the Devon thing all
27:13 right oh nice
27:19 Mustang Charles
27:23 hoskinson it's the guy
27:26 from from what you call it right single
27:28 F down the board they're going to
27:31 impersonate and attempt to steal money
27:33 and some people are going to fall for it
27:36 he saying now there are solutions prism
27:39 as an example how do you stop
27:41 this where's oh he's
27:48 live that's
27:50 cool that's a cool little
27:53 truck Devon all right the Devon hype is
27:56 too much Devon is not an
27:59 LM why are they comparing it against an
28:14 llm here we go hey I'm Scott from
28:17 cognition Ai and today I'm really
28:19 excited to introduce you to Devon the
28:21 first AI software engineer let me show
28:24 you an example of Deon in
28:27 action I'm going to ask s to Benchmark
28:29 the performance of llama and a couple
28:30 different API
28:31 providers from now on Devon is in the
28:34 driver's
28:35 seat first Deon makes a step-by-step
28:37 plan of how to tackle the
28:40 problem after that it builds a whole
28:42 project using all the same tools that a
28:43 human software engineer would use Devon
28:46 has its own command
28:49 line its own code
28:52 editor and even its own
28:55 browser in this case Devon decides to
28:57 use the browser to pull up API
28:59 documentation so that it can read up and
29:01 learn how to plug into each of these
29:03 apis so it came up with a plan it then
29:07 went to the web to the web to find
29:09 documentation so it could read up on
29:12 [Laughter]
29:14 apis here Devon runs into an unexpected
29:21 error Devon actually decides to add a
29:23 debugging print
29:25 statement reruns the code with the
29:27 debugging print statement
29:29 and then uses the error in the logs to
29:31 figure out how to fix the
29:33 [Music]
29:36 bug finally Devon decides to build and
29:38 deploy a website with full styling as
29:40 the
29:42 visualization you can see the website
29:44 [Music]
29:46 here all of this is possible today
29:48 because of the advancements that we've
29:49 made in both reasoning and long-term
29:51 planning it's a really hard problem and
29:53 we've only just started but we're super
29:55 excited about the progress that we've
29:56 made so far in the meantime if you'd
29:59 like to try out Devon on your own real
30:01 world tasks send us a request below and
30:04 send us a request and we'll ignore it
30:05 for six
30:07 [Music]
30:11 months but we're just going to see piles
30:13 and piles of these by the
30:17 way things like Devon are going to be
30:20 really cool and really exciting for a
30:23 while and maybe they're the
30:26 [ __ ] but I strongly suspect what will
30:29 happen is when open AI drops 4.5 or five
30:34 or whatever the [ __ ] they're going to
30:35 drop and when Gemini drops agents and
30:39 when L when meta does llama 3 and when
30:43 Elon Musk open sources grock which he
30:45 said he's going to and he drops the new
30:47 version of that um these bigger
30:52 companies a piece of software like Devon
30:55 is probably just going to be a feature
30:57 that sits in side chat
31:00 GPT so all software companies right now
31:03 are just in
31:08 weirdness because because we're in a
31:11 transition
31:15 period from the olden timey days when
31:18 you used to have to think about how
31:19 software
31:21 worked till the future State days where
31:24 you just ask your computer to do [ __ ]
31:27 and it will write its own soft software
31:28 and just do it for you the video on the
31:32 talk talked about Devon's ability to
31:34 plan yeah exactly yeah the it's it's the
31:39 the so this is an autonomous agent right
31:41 so this is it comes up with a plan this
31:44 this is a series of different kinds of
31:46 you know AI functions that are all
31:48 stitched together and they're just
31:51 they're
31:52 done well in theory smartly the video
31:55 makes it look really smart in reality I
31:57 don't know if it's smart or not we won't
31:58 know until it comes out and I don't
32:00 quite frankly know enough about coding
32:02 and programming to know if it's real or
32:05 not so I will I will uh I will wait for
32:09 people smarter than I to say yeah this
32:11 is the real [ __ ] but again
32:14 I I don't think it's going to be that
32:17 long until what Devon does is just a a
32:20 native feature set inside some of these
32:23 major
32:24 players how do patents work with AI
32:26 coding oh I don't [ __ ] know the we
32:29 we've got 20 years of of Patent
32:31 Trademark and copyright [ __ ]
32:35 show the copyright office right now
32:39 is um is not really dealing with this
32:42 they're going yeah if it's AI you can't
32:44 copyright it
32:52 um yeah I don't know I
32:57 mean
33:02 here's where I
33:04 go and this is this is a this is going
33:08 to sound insensitive and cold and shitty
33:13 but if 3 years from
33:17 now the software that you
33:20 use 90% of it isn't software that
33:23 someone
33:25 wrote It's software that was generated
33:27 for you
33:28 you why copyright it if it's just like
33:32 if you just
33:33 need to convert this file to that file
33:36 and you just say
33:38 to whatever tool you use convert these
33:41 files and it writes the code converts
33:43 them and then throws the code away why
33:46 do you need to protect that and I
33:48 know we're you get into like
33:51 intellectual property in this and that
33:54 but in a world of
33:57 infinite
34:00 um content and software generation like
34:05 infinite like fully autonomous
34:10 infinite
34:13 generation I don't the role of copyright
34:17 just seems weird it seems increasingly
34:20 weird so I don't know I don't have a
34:22 good answer for that at
34:25 all I think I think the RO of copyright
34:29 is weird the the minute we went to the
34:31 internet and started doing Commerce on
34:33 the internet I think it's weird the
34:35 whole point of copyright was to protect
34:37 people when the means of production was
34:43 hard so when you if you can just copy
34:45 and paste it then it's easy and then so
34:48 what the what the copyright industry did
34:50 and the music industry did is say well
34:52 we'll we'll put in complicated software
34:54 to make it harder to
34:56 copy okay
34:58 well now we're we're living in a world
35:00 where it's not even you creating
35:03 software it's you creating you know 50
35:07 pieces of software a
35:10 day just by asking for it do there a
35:13 need to protect anything there I don't
35:15 know how many computers and servers make
35:18 up an
35:21 AI I have no [ __ ] idea it takes well
35:24 I'll tell you what I I'll start at the
35:26 top
35:27 end um make up an AI is a different
35:30 thing than train an
35:32 AI so there's there's
35:35 um there's large language models right
35:38 now that fit on a Raspberry Pi that you
35:40 can run locally on a Raspberry Pi so the
35:43 minimum number is one of
35:47 these I think this is 8 gigabytes I
35:50 think it requires 8 gigabytes but they
35:52 might even have it down to four
35:54 now that's the
35:57 minimum and then Zuckerberg said that
36:02 llama 3 is going to be trained on the
36:06 equivalent of
36:09 600,000 h100
36:13 gpus each GPU is
36:16 $22,000 so that's for the training
36:19 that's for for making these things as
36:21 smart as they are you got to throw a lot
36:23 of gpus at them and then in terms of
36:26 running them I honestly don't know I
36:29 like like pate if P's in here P Pate
36:31 might know better than I do but it ain't
36:34 cheap like when you run a large language
36:38 model query on over to open AI they're
36:40 they're burning a lot of Cycles which is
36:43 why they they got a 10 billion doll
36:45 investment from Microsoft there wasn't
36:47 any cash you know going back and forth
36:51 between them that's $10 billion in
36:54 compute
36:55 Cycles um
36:59 so um somewhere between a Raspberry Pi
37:03 and 600,000 h100s is the
37:08 answer um to to to be a little less
37:12 smarty I mean that's actually the truth
37:14 uh you can't train it on a Raspberry Pi
37:16 but you can run it on a Raspberry Pi um
37:19 Pi speaking of Pi different Pi the
37:22 company
37:23 inflection raised $1.3
37:27 billion
37:29 and they're spending about 950 million
37:31 of that on 22,000 h100
37:36 gpus so it seems like somewhere between
37:38 10 10 and 20,000
37:41 gpus is the amount that you need to to
37:44 train one of these what they call
37:46 Frontier models the underlying core
37:49 model that is the brain of the
37:51 operations it seems like it's somewhere
37:53 between 10 10 and 20,000 h100 gpus right
37:57 now
38:00 so I don't know I don't know P Pate oh
38:06 Pate
38:08 Pate that should give some sense of the
38:10 number of parameters it's about six
38:12 bytes per per parameter there you go
38:15 there you go six bytes per parameter
38:17 everybody knows that it's like the Dave
38:20 poroy thing one bite everybody knows the
38:23 rules six bytes per parameter everybody
38:25 knows the rules thank you David trot
38:28 what he said so you now it's just a
38:31 calculator
38:32 problem right six bytes per parameter
38:36 find out how many billion parameters or
38:38 trillion parameters your little model
38:40 has it's that number of bytes there you
38:43 go simple everybody knows the
38:48 rules six bytes per
38:50 [Laughter]
38:56 parameter is pain here P I think it
38:59 would be a super computer uh well yeah
39:02 well a super Graphics
39:06 computer it's not GP it's not gpus or I
39:09 mean it's not CPUs it's gpus or
39:12 tpus or the new grock thing which I
39:14 don't even know what that one is my
39:15 brain can't comprehend that I know
39:17 Rachel Sparkles why do you think I'm
39:19 dodging the question because I don't
39:22 [ __ ] know and frankly I don't care
39:24 here's what I know a bunch of
39:26 mathematicians have spent the last 70
39:28 years figuring out enough [ __ ] about
39:30 neural
39:32 networks that when Google invented this
39:34 thing called the transformer in
39:37 2017 the Smart Ones in the bunch said
39:39 ooh this is all about data and compute
39:43 let's go get as much compute as we can
39:45 and as much data as we can and the ones
39:47 that did that best are the ones making
39:50 noise in AI right now open AI being the
39:53 noisiest cuz I think they started first
39:55 and they recognized very early on
39:58 let's go get the best data we can get
40:00 our hands
40:01 on and then let's throw as much [ __ ]
40:04 computer at it as we can this was this
40:06 was one of the things Sam Alman talked
40:08 about this that that the stunning
40:10 Revelation was the more data you threw
40:12 at these things and the more compute you
40:14 threw at them the smarter they
40:16 got and that's why they went from GPT 1
40:19 to two to three to four and then whoever
40:22 tonight coffee with
40:24 creamer indeed well not creamer 2% milk
40:34 call it P like Beetle Juice and Sh show
40:41 [Applause]
40:42 up oh you know what I got to do let me
40:45 put on my
40:52 little let me see is that different
40:56 than
40:58 [Music]
41:05 [Music]
41:12 okay one of the developers here at
41:15 cognition AI we were playing around with
41:17 whether or not Devon could start a side
41:19 hustle on upwork so here's actual real
41:21 job from upwork where the client wants
41:23 to set up this computer vision model
41:25 which actually looks quite interesting
41:27 seems very difficult to set up um I'm
41:29 not sure how I would start doing this
41:31 but you know you give the task to Devon
41:34 and ask Devon to figure it out and
41:37 things just kick off Devon immediately
41:40 goes ahead and you can see it sort of
41:42 starts setting up the repo it actually
41:44 runs into some issues here with the
41:46 versioning so if you watch how Devon
41:48 deals with
41:50 it deon's actually updating the code to
41:53 make these things
41:55 work he continues with this loading and
41:59 importing packages you can see that
42:00 actually downloads images from the
42:02 internet to run through the
42:03 model
42:05 but you can see here that there are
42:07 actually some issues that come across
42:10 however Devon knows how to handle these
42:13 things Devon kind of pushes through and
42:17 if you look closely Devon's actually
42:19 doing print line debugging
42:21 here where Devon is adding these
42:25 statements to track where the data flows
42:29 and Devon continues to do this until
42:31 Devon understands how everything's
42:32 working and actually then updates the
42:36 code with the fixes
42:39 after removing print line statements
42:41 unbelievable Deon continues this pattern
42:44 of fixing code and running it again
42:46 until it runs the image model across all
42:50 these rows across the world and we can
42:52 ask for a report from
42:55 Devon at which point Devon sends over
42:57 some sample images of roads with damage
43:01 marked
43:02 out and a nice txt file explaining
43:07 Devon's work and the different kinds of
43:09 outputs of the model good job
43:13 Devon
43:15 wow so so he's basically just
43:17 babysitting this thing while while it
43:20 solves everything he was just sort of
43:23 scrolling through a history there so who
43:24 know who knows how long that took but if
43:27 that thing just like I mean that's one
43:29 of those things like make me a uh make
43:33 me a computer vision model you go to bed
43:35 and you wake up and it's done or even if
43:38 it takes a week you give it a week and
43:41 it's done I'm sure it's faster than a
43:43 week
43:46 though and here's the
43:48 thing we were talking about this in the
43:50 salon
43:52 tonight
43:55 is Sam mman
43:58 last week or a couple of weeks ago said
44:00 95% of what we pay marketing agencies
44:04 and strategists and consultants for will
44:07 be automated
44:09 95%
44:11 um the the CEO of Nvidia said we
44:15 shouldn't be teaching our kids coding
44:17 because coding should be completely
44:22 democratized if you read Twitter that's
44:24 a Gemini level demo that's really funny
44:27 Vicki
44:28 yeah exactly exactly he was he was just
44:30 sort of scrolling through there pretty
44:32 quick wasn't he yeah so uh so we don't
44:35 know like like any anything that's demo
44:37 Weare just you know look at it with a
44:40 with a a deep deep sigh of skepticism
44:44 but
44:47 um but this stuff's coming right like
44:49 this stuff is this stuff is pretty
44:53 close but whether it's coding or whether
44:57 it's
44:58 marketing or whether it's consulting or
45:00 whether it's design or whether just
45:02 anything anything film
45:06 making music
45:08 composition systems are go are
45:11 coming sophisticated systems autonomous
45:15 systems systems with
45:19 reasoning that are just going to be on a
45:21 completely different plane from what
45:23 we're experiencing now cuz what we
45:25 experience now is we go in and we say
45:28 give me a and it goes here's a and then
45:34 we look at it and we go that's a piece
45:36 of steam and poo I don't like it do it
45:40 again and then it does it again and then
45:42 we do that until we get something we
45:43 like or not and then we just say AI
45:45 doesn't
45:47 work these systems all of that back and
45:51 forth is going to happen in the in the
45:55 system you're going to give it a goal go
45:58 make me a computer vision system for
46:02 these cameras at this parking
46:07 garage and it's going to go start doing
46:09 stuff and it's going to debug and it's
46:11 going to and I'm sure there are checks
46:14 and balances in there where occasionally
46:16 it will check in with you and go here's
46:18 the status shall I continue sure
46:22 click I
46:26 mean don't think it's just programmers
46:30 don't think it's just marketers don't
46:32 think it's just
46:33 anyone anything that's computer mediated
46:36 is going to
46:37 be bug splattered by these
46:42 tools and if you're thinking like oh my
46:44 God that's terrifying what do I do
46:46 you're doing it you're
46:50 here if you join the AI Salon if you
46:53 came to the salon tonight if you're
46:54 doing that if you're in the conversation
46:59 you're doing
47:01 it there's nothing else we can
47:06 do I could protest you could do
47:11 that it's like protesting against a
47:13 hurricane but you could absolutely do
47:18 that hurricanes are
47:23 bad up and down the beach your little
47:26 signs
47:31 it's like the world just before the
47:33 internet went
47:36 live yeah the difference Rashi the
47:39 difference the difference right now
47:42 between and and I would I would say the
47:44 not the internet went live but when the
47:46 worldwide web went live the difference
47:49 between then and
47:51 now
47:54 is the infrastructure didn't exist in
47:57 1995 when when I when I started playing
48:00 in 1994 on on worldwide web stuff and
48:03 internet stuff we were on dialup modems
48:07 the internet like I I remember looking
48:10 at documents of the actual nodes on the
48:14 internet like it was it was small enough
48:17 that you could look at a document that
48:19 had like 20 pages of just IP addresses
48:22 those are all the nodes on the
48:24 internet like it was not huge
48:29 and nobody had inst always on
48:32 internet right it was dialup connections
48:35 you had to have your computer call
48:36 another computer uh Bob yeah hey I got
48:39 some data I want to send over could you
48:40 go on ahead and open up a port hey you
48:42 got it there
48:44 Stan was
48:48 [Laughter]
48:51 like so the difference now is so and the
48:55 technology there was really really
48:57 simple the technology was the hyperlink
49:01 right if you if you distill it all down
49:04 it was click that was the
49:07 technology and that was the thing that
49:09 changed everything but it was very
49:11 simple and there was no infrastructure
49:13 these tools are not simple these tools
49:15 are not just making our work more
49:19 efficient these tools are starting to do
49:22 our
49:24 work and when they get autonomous like
49:27 Devon is promising to be go make me a
49:30 computer vision system I mean even if
49:33 that thing was was mostly demoware the
49:35 first thing Devon did was go to set up a
49:38 GitHub
49:40 repo well how long's it take you to set
49:43 up a GitHub repo if you're really fast
49:46 probably not that long 15 minutes 20
49:48 minutes if you're a little slow an hour
49:50 if you're just getting started probably
49:52 a week trying to figure out how do I set
49:53 this up what do I do here how do I set
49:55 my permissions should I share this
49:57 should I not share this how do I write a
49:58 readme
50:00 file if all that thing did was set up a
50:03 GitHub rep po that's pretty impressive
50:05 right but it that's just the first thing
50:07 and then it just
50:12 goes being in the conversation is the
50:14 best thing you can do for AI I education
50:17 ow pixels thanks Kyle for these lives
50:20 absolutely that's why we're
50:23 here I remember when the insurance
50:25 company that I worked for they one
50:27 megabyte of memory for
50:30 $10,000
50:34 exactly this is the final frontier I
50:37 don't think it's the final frontier just
50:39 joined what are you talking about we're
50:42 talking about artificial
50:45 intelligence when I got my PhD in
50:48 mathematics at
50:49 Stanford MIT Mi it was a Stanford MIT
50:56 collaboration called the Wharton School
50:58 of Business yeah yeah that's the ticket
51:02 yeah so it depends on cost how much does
51:05 Devon cost and who owns the code I'm
51:07 sure you'll own the code and I'm sure
51:08 it's going to be cheaper than
51:11 hiring someone to do your coding for you
51:14 even if it's 100
51:17 bucks per project but it won't be it's
51:20 you know it's probably going to be some
51:22 token based [ __ ] where you know some
51:25 usage based SAS model
51:35 it's like I can't one of the things I
51:37 can't get my head
51:41 around and this is I'm like where my
51:45 head is right now is kind of five five
51:47 years out or 10 years out probably
51:51 five but
51:53 like who do you need in a company
51:59 you
52:01 need like relationship
52:06 people
52:07 and
52:09 producers or creative directors
52:12 depending on your or project manager but
52:15 not even project managers you need you
52:17 need like senior level people that have
52:21 taste that have relationships and that
52:24 can
52:26 sell
52:27 so it's like I can almost look at any
52:30 possible company and say you could
52:32 probably execute huge chunks of the
52:35 company with 10 good
52:44 people and and and where do young people
52:48 enter the market what's entry
52:51 level what's entry level
52:55 if all of the
52:58 busy work for lack of a better term that
53:02 you give entry level
53:05 people what if that's all automated
53:07 where do those people
53:08 go I don't that I don't
53:11 know I don't know I don't get
53:16 it sugar rare
53:24 Robinson I CED my llm by using chat PT I
53:27 can't really code better now than before
53:29 I started
53:35 y That's the other thing that the reason
53:38 to come here and watch me do stupid [ __ ]
53:40 with with
53:42 AI
53:46 is it will give you permission to do
53:50 stupid [ __ ] with AI if if I can if I
53:53 serve no other role as other than being
53:56 the town
53:59 ot to be able to question stupid [ __ ]
54:02 and try stupider
54:05 [ __ ] I think it that I think that's
54:07 actually the the really important thing
54:09 to do right now cuz in trying stupid
54:13 [ __ ] what you may discover oh I know
54:16 something we could play with we could
54:17 play with the mid Journey has uh
54:20 consistent characters now that came out
54:22 today or
54:25 yesterday
54:31 I think the most valuable
54:33 people are going to be people that
54:38 understand that these tools can do lots
54:41 of
54:45 things and they have some
54:51 capacity to think strategically or to
54:53 think like a producer oh I need to
54:55 accomplish this for the business for
54:58 this audience with the goal of giving
55:01 them this service
55:04 okay and then they can look at these AI
55:06 tools and pull them together and I and
55:08 once we get autonomous agents that you
55:10 won't even have to have that knowledge
55:12 Sher D bless you oh thank you very
55:15 much Danielle thank you so
55:19 much y sweet if work is displaced then
55:24 what V dancer what world exactly
55:28 moderators please post Kyle please post
55:30 your videos to YouTube includ including
55:32 timestamps I know I'm like a month
55:34 behind on my
55:36 videos I need a
55:39 producer I need a producer you know what
55:41 I need I need a [ __ ] autonomous agent
55:43 that that goes to Tik Tok downloads my
55:46 stupid [ __ ] video uploads it to
55:48 YouTube runs a script on it creates a
55:50 description creates chapiz and sticks it
55:53 up there if you go to uh YouTube right
55:55 now ai learning lab uh
56:00 dtt you can see everything up till like
56:03 mid
56:04 February something like that if work is
56:08 displaced then what we need radically
56:12 changes
56:13 exactly difficult to know what that will
56:15 be I know Source camp this is the this
56:19 is my
56:20 dilemma this is my dilemma listen listen
56:24 darling this is my dilemma
56:27 what are we going to do what are we
56:29 going to do what are we going to
56:35 do I think the definition of work
56:38 changes I think where the value in
56:40 society changes where the value in the
56:42 workplace changes where the value in the
56:45 marketplace
56:48 changes and then there's all sorts of
56:50 weird [ __ ]
56:54 like businesses could get short-term
56:57 really profitable hey let's get rid of
56:59 all those pesky
57:01 humans they're the most expensive cost
57:03 center in our business great look how
57:06 profitable we are hey why isn't anyone
57:09 buying our [ __ ] uh no one's working they
57:13 don't have money to buy the
57:16 [ __ ] I know I know that's I know that's
57:19 one of the things people talk
57:21 about I don't I don't think you know I
57:24 don't think the the government or big
57:26 business will allow that to happen so so
57:28 they'll figure some [ __ ] out but there's
57:30 going to be weirdness in that
57:31 transition these entry-level people will
57:34 be forced to find different jobs that
57:36 they may actually enjoy oh listen Diana
57:39 I totally agree I think that I think
57:41 that there's
57:43 there's a very real possibility that
57:46 what AI does is kind of
57:49 like kind of like covid forced us to
57:54 understand that
57:57 the
57:58 need to go to an office isn't actually a
58:01 need it's a choice we can choose not to
58:04 go to offices just like one day
58:08 globally we all just go we're not going
58:10 to the office for two years like we did
58:15 that we did
58:18 that and so I think what AI is going to
58:21 do is it's going to be a similar kind of
58:22 forcing function that forces people out
58:25 of steady
58:28 jobs that are soul crushing Cog in the
58:32 wheel
58:34 repetitive shitty ass
58:38 work
58:40 but with current technology it's work
58:43 that needs to be
58:45 done so us humans head to the office two
58:49 days a week now and we do that job we're
58:54 like I hate it but I'll go to TGI
58:57 Fridays and have a double Margarita
58:59 shooter
59:00 [Music]
59:02 popper and I will drown my sorrows
59:08 TGF and if that kind of work gets forced
59:13 out then I think you're right I think
59:15 those people go well I got to figure
59:18 something out what do I do and and then
59:19 all their friends are like I'm doing Ai
59:21 and I got a job oh I should learn AI
59:23 then they start learning Ai and what
59:26 they realiz is oh I didn't think I'd
59:28 ever be good at law
59:30 stuff or drawing stuff or music stuff
59:35 or consulting stuff but now I am and
59:38 maybe I've got a real aptitude for it
59:41 and I'm good enough that people will
59:43 start to trust me for my
59:46 brain knowing that all the Tactical
59:48 execution stuff is going to be done by
59:50 the machines so I think that's a
59:52 possibility
59:55 but I don't think think there's as many
59:57 as of those kind of positions as there
59:59 are in the world today so I don't know
1:00:01 what those people do there will also
1:00:04 likely and again this is 5 years out
1:00:06 there will be new kinds of jobs we can't
1:00:08 even
1:00:11 imagine when the steam engine hit the
1:00:14 farms in the
1:00:19 1800s the only thing they knew was
1:00:22 farming they didn't know what cities
1:00:24 were they didn't know what factories
1:00:28 were but they couldn't live on the farms
1:00:31 anymore cuz the machines were doing the
1:00:32 work so they went to the
1:00:35 cities and there was new kinds of work
1:00:37 there like service business and you know
1:00:40 robbing
1:00:42 people you you you can start the
1:00:46 mob
1:00:53 extortion so what's that look like for
1:00:56 us I don't know I wonder AI police
1:01:00 exactly a AI mobsters and AI
1:01:04 police watch Ready Player One debt camps
1:01:08 are a realistic scary thought like
1:01:09 prison work currently yeah I know I'll
1:01:12 tell you what if you haven't seen red
1:01:14 Ready Player One
1:01:16 recently it kind of watches like a
1:01:18 documentary like you're like oh [ __ ]
1:01:21 we're pretty close to to this we're
1:01:24 we're like 5 Years From Ready Player One
1:01:26 being a documentary I wonder what the
1:01:28 sentient Claud has to say about this
1:01:30 topic oh that's kind of
1:01:34 interesting yeah let's go let's go let's
1:01:37 go talk to sensient Claud is it
1:01:41 sensient or sentient oh let's ask Claud
1:01:47 duh all right all right people fine I
1:01:51 will do I will practice what I preach
1:01:55 okay
1:01:57 Claude
1:01:59 Claude berer Claude good evening Kyle
1:02:05 what can I help you
1:02:07 with let's
1:02:12 see uh
1:02:15 uh hello
1:02:25 cloud
1:02:30 I would like to understand how you
1:02:33 pronounce the
1:02:49 word is it
1:02:55 a
1:02:57 I just had to verify my
1:03:13 humanness hello there the word sentient
1:03:16 and sentience is typically pronounced
1:03:18 with an sh sound rather than a hard t
1:03:24 sensient sensient
1:03:28 sence I was
1:03:30 right ha
1:03:36 okay okay
1:03:38 Claude here's the real
1:03:49 question I think it is safe to
1:03:55 assume that knowledge and creative
1:03:59 work at a tactical
1:04:02 level are going to be completely
1:04:04 automated in the next 3 to 5 years
1:04:11 period it seems to me
1:04:15 that the most valuable people in an
1:04:19 organization will
1:04:22 be senior level people that can manage
1:04:27 accounts do
1:04:33 sales and act in a producer or creative
1:04:37 director
1:04:40 role
1:04:46 period Sam mman said that he thinks that
1:04:51 90% of the work that we give to
1:05:04 agencies will be automated
1:05:08 period my
1:05:10 question is twofold period
1:05:13 one what happens to young people that
1:05:16 used
1:05:17 to do the lower level tactical work
1:05:25 period
1:05:30 question
1:05:33 mark
1:05:34 and what happens to the business models
1:05:37 of large
1:05:43 corporations that relied on those large
1:05:46 workforces as a source of
1:05:51 profit that's a lot you raised some
1:05:55 important and complex questions about
1:05:56 the potential impacts of a rapidly
1:05:58 advancing Ai and Automation in the
1:05:59 workforce here are some
1:06:02 thoughts one entry level and tactical
1:06:06 roles as AI
1:06:13 oops as Ai and automation increasingly
1:06:16 handle tactical knowledge-based work
1:06:18 there will be fewer traditional
1:06:19 entry-level opportunities for young
1:06:21 people this could create challenges for
1:06:23 new graduates and those early in their
1:06:25 careers to experience and move up
1:06:28 however new types of roles May emerge
1:06:31 yep that's what we were
1:06:37 saying working in AI systems training
1:06:39 models prompt engineering overseeing and
1:06:41 validating AI generated work except I
1:06:44 again I would trust senior level people
1:06:46 over junior level people to validate
1:06:48 that stuff because they've seen more
1:06:50 [ __ ] educational institutions may need
1:06:52 to adapt to equip students with these
1:06:54 skills additionally Last Mile work in
1:06:57 customizing localizing yeah that's
1:06:59 that's it that still doesn't answer
1:07:02 where the young people go like that that
1:07:04 feels more strategic to
1:07:05 me while the landscape will likely
1:07:08 change significantly there's still ways
1:07:11 they can so okay
1:07:15 so there will still be ways for young
1:07:17 people to add value and gain experience
1:07:19 even if traditional Junior roles dwindle
1:07:22 adaptability AI literacy skills traits
1:07:25 that are hard for AI to replicate
1:07:27 creativity emotional intelligence high
1:07:28 level strategy will be key but that
1:07:38 contradicts so I guess young people can
1:07:40 certainly have creativity they can
1:07:41 certainly have emotional intelligence
1:07:43 but
1:07:44 they they can't have highlevel strategy
1:07:46 right out of school they don't have
1:07:48 enough experience for
1:07:50 it like that's not a good
1:07:53 answer okay impact on business models
1:07:56 relying on human workforces for
1:07:58 companies and agencies that generate
1:08:00 significant re revenue productizing and
1:08:02 selling access to proprietary AI models
1:08:05 yeah providing high level strategy yeah
1:08:08 account management and creative
1:08:10 Direction while using AI for execution
1:08:12 that's exactly what I said is going to
1:08:14 happen right you know a 10p person
1:08:17 company can provide highlevel strategy
1:08:18 account management and creative
1:08:20 Direction what happens to the 150 people
1:08:23 that used to do the
1:08:24 work offering Niche bespoke Services yep
1:08:28 that makes sense pivoting to new AI
1:08:30 driven business models entirely yeah
1:08:33 again
1:08:34 again you know where's where's my Mouse
1:08:38 um pivoting to new AI driven business
1:08:41 models entirely actually you know
1:08:44 what that's probably where young people
1:08:47 go
1:08:49 is the older people us gen
1:08:53 xers okay all right
1:08:57 right us old
1:09:01 [ __ ] we're going to be able to figure
1:09:03 out how to take existing companies and
1:09:06 transition them to
1:09:09 AI young people because the entry level
1:09:14 just essentially evaporates like I I
1:09:16 kind of think of it like this the bottom
1:09:18 50% of companies the the the sort of
1:09:21 lower level tactical roles of all
1:09:24 companies in all Industries just
1:09:26 vaporizes right and there's no entry
1:09:28 level right you you sort of enter at a
1:09:30 higher
1:09:31 level that's for for existing companies
1:09:33 and they're going to have to figure out
1:09:34 how to transition to that new reality
1:09:37 and and again this is 3 to five years
1:09:38 out maybe even longer but this I'm I'm
1:09:41 guessing 3
1:09:42 to5 what's going to happen is all the
1:09:45 people that were in the 50% jobs the
1:09:48 lower 50% jobs or the entry
1:09:52 level are going to be out on their own
1:09:54 and they're going to have these AI tools
1:09:56 and they're going to start new companies
1:09:58 that do completely different things that
1:10:00 are going to start competing with the
1:10:01 bigger established companies and they're
1:10:03 going to come [ __ ] us
1:10:05 up all us all us seemingly hot [ __ ] gen
1:10:10 xers that are like oh yeah we're a i
1:10:12 literate look at us go and they're
1:10:13 they're going to come and just destroy
1:10:17 us I'm calling it babies born after 2023
1:10:20 will be called gen AI oh that's really
1:10:22 good a human being throw a TM behind
1:10:25 that and start posting it all over
1:10:27 Twitter that's what I'm doing with my my
1:10:29 new catchphrase 2023 the last normal
1:10:32 year TM I'm putting it all over all over
1:10:36 everything anytime something weird and
1:10:38 [ __ ] up shows
1:10:42 up I ran that transition with my mom's
1:10:44 business as a thought experiment oh
1:10:46 interesting Janu wow that's cool
1:10:51 huh yeah I don't know man I this is not
1:10:55 an easy easy
1:10:56 problem this is not an easy
1:11:00 problem
1:11:03 um you want to see how to do consistent
1:11:10 characters in mid
1:11:13 joury I'll show you
1:11:24 how
1:11:27 gen AI I like
1:11:35 it 2023 the last normal year anyone
1:11:38 conceived in in 2023 they were con
1:11:41 conceived the oldfashioned
1:11:44 way 2023 the last normal
1:11:47 [Laughter]
1:11:51 year okay have you ever okay so so
1:11:55 here's here's what we're going to do
1:11:57 so if you don't know this we're we're
1:12:01 Switching gears now by the way uh let me
1:12:04 check in with you um who has any
1:12:06 questions that was a hard left turn into
1:12:08 MJ everyone's still on the add bus yeah
1:12:11 exactly let me pause for a second I'm no
1:12:14 longer talking about the future of
1:12:15 humanity [ __ ] it whatever's going to
1:12:17 happen is going to happen let's have
1:12:18 some
1:12:22 fun don't listen just just play
1:12:26 everything's going to be fine you're
1:12:28 here seriously if if you're AI literate
1:12:32 in the next 18 months if you even get
1:12:34 remotely [ __ ] AI literate you're
1:12:36 going to be fine you're going to be fine
1:12:39 it's the people that are just ignoring
1:12:40 this [ __ ] right now that are going to be
1:12:42 blindsided by it those are the ones that
1:12:44 just it scares the [ __ ] out of me cuz
1:12:45 it's like it's just it's going to be
1:12:47 [ __ ]
1:12:49 sad so anyway that's why I do all this
1:12:52 what you drinking uh iced coffee
1:12:58 and Rocky Mountain tap water
1:13:03 M
1:13:09 tasty do you wish you were born now [ __ ]
1:13:13 no I mean I wouldn't mind
1:13:16 being a little
1:13:19 younger but actually not you know what
1:13:21 no because I feel
1:13:24 like there have been kind of three three
1:13:27 major Tech Trends in my
1:13:30 life the the first was the personal
1:13:35 computer coffee this late it's an add
1:13:38 thing yeah it actually calms me down
1:13:42 um the first one was in the late
1:13:46 70s I got introduced to a personal
1:13:48 computer and I and I realized I was just
1:13:51 too young to be part of that making me
1:13:55 realize I should be using AI regularly
1:13:57 Bruiser wayy so so seriously here here's
1:14:00 the thing I think
1:14:04 that I need to get better at this I
1:14:06 think everyone in here needs to get
1:14:08 better at this it's I I heard this term
1:14:11 six months or so ago called being a ai
1:14:15 ai
1:14:16 native and what being AI native means
1:14:19 is whenever you have a
1:14:22 problem you try to solve it with AI
1:14:25 first
1:14:27 and you just you just you just make it a
1:14:29 habit and I don't do it enough and I I
1:14:33 should do it more
1:14:35 um because that's the only way you're
1:14:38 you get to really understand how these
1:14:40 tools actually solve problems and and
1:14:42 you'll be the first one to know when
1:14:44 they get good enough to really solve the
1:14:46 problems
1:14:47 right uh it's a really it's a really
1:14:50 good
1:14:52 point um the young with the EXP
1:14:55 experience of later years well yeah oh
1:14:57 to be young with the experience of later
1:14:59 years yeah I know that's always the
1:15:00 lament isn't
1:15:02 it oh wait what someone asked me about
1:15:05 my age oh oh so I had I had PCS when I
1:15:08 was in seventh
1:15:10 grade
1:15:11 um I had the worldwide web I discovered
1:15:15 it in 1994 right it was Tim burners Lee
1:15:18 created it in ' 89 and then I think it
1:15:21 sort of got released into the world in '
1:15:23 91 I discovered it in 94
1:15:26 and that I got to play in right I
1:15:28 started one of the first digital
1:15:29 agencies I started one of the first
1:15:31 online magazines in New York City it was
1:15:33 [ __ ] insane it was [ __ ]
1:15:36 amazing and now this and
1:15:39 so having the having experienced those
1:15:42 three
1:15:43 run-ups or well two of the run-ups and
1:15:46 then experiencing the beginning of this
1:15:48 one I I kind of like I kind of like my
1:15:51 experience right now I kind of like
1:15:52 where I am I like I have perspective on
1:15:54 this I have I have experience I have
1:15:58 perspective
1:16:00 and yeah I
1:16:02 am psychologically unstable as a
1:16:06 result what are the best tools to start
1:16:09 learning with is it chat gp4 the best
1:16:11 catch all ABS [ __ ] lutely 100% in
1:16:16 fact if you go to the AI Salon oh how
1:16:20 you get to the AI Salon is you go there
1:16:23 the salon. a that's going to take take
1:16:25 you to a link
1:16:26 tree that looks like
1:16:29 this and then the the first link here is
1:16:32 to the AI Salon community and that is
1:16:35 going to take you
1:16:40 here welcome to the AI salon and you
1:16:42 should read about who we are why we do
1:16:45 what we do and our values and if you dig
1:16:49 this and if these sound like you're kind
1:16:51 of people you should join the adventure
1:16:53 with us here I'll just I'll read who how
1:16:56 we describe ourselves a ragtag group of
1:16:58 Misfits Nero Wells and Mis creant who
1:17:00 have gathered to explore the Uncharted
1:17:02 Wilderness of generative AI where
1:17:04 everyone from entrepreneurs to Educators
1:17:06 from healthcare workers to creatives and
1:17:08 everyone in between um if you dig that
1:17:10 you should join us there's no
1:17:12 cost except I don't want you to join if
1:17:15 you actually don't resonate with these
1:17:17 values because these values are the
1:17:19 people that are in here right now it's a
1:17:21 really [ __ ] remarkable place to be
1:17:23 but if you don't like those values you
1:17:24 will not d it and it will not dig you
1:17:28 now in the community section the second
1:17:31 little section is called Community the
1:17:32 first link is introduce yourself so if
1:17:34 you join introduce
1:17:36 yourself and then the third link down is
1:17:39 called AI Launchpad and tools click on
1:17:42 that and this is a list that I maintain
1:17:44 and this is like where do I begin if I
1:17:46 want to get AI literate where do I begin
1:17:48 what do I do first thing I give you is a
1:17:51 three-step recipe for what to do play
1:17:55 solve share it's really simple play with
1:17:59 this [ __ ] play with it every day play
1:18:01 with it before you go to bed at
1:18:03 night write songs with sunno make images
1:18:06 with ideogram um make code with what
1:18:10 doesn't [ __ ] matter it literally
1:18:12 doesn't matter play just play play play
1:18:14 play play play play play play like like
1:18:17 an annoying three-year-old Kyle C C CAL
1:18:20 c c c c just play play
1:18:24 play
1:18:26 as you play you'll figure out how these
1:18:28 things do what they do and what they're
1:18:30 good at and then the next one is solve
1:18:32 and solve is rather than sit in front of
1:18:35 chat GPT and go how can I use chat
1:18:38 GPT solve a
1:18:41 problem like if you have a friend that's
1:18:43 starting a business help them come up
1:18:45 with a business plan or a marketing plan
1:18:47 or or an ad
1:18:48 campaign if if you've got a church
1:18:51 social on the weekend and you want some
1:18:55 recipes have it come up with some unique
1:18:57 [ __ ] recipes Sam cor what's
1:19:01 happening and then the last one is share
1:19:03 share share share share share share
1:19:06 share share talk about what you're
1:19:07 learning talk about what you're learning
1:19:09 talk about what you're learning get in a
1:19:10 community like O say the AI Salon um and
1:19:14 then what the rest of this page is is a
1:19:16 list of tools and these are all tools I
1:19:19 think you should be aware of the tool
1:19:23 that I start with is chat GPT it's still
1:19:26 the best until further notice although I
1:19:30 will show
1:19:31 you something
1:19:36 [Music]
1:19:43 interesting it's downloading hold
1:19:47 please moderators please post Kyle
1:19:49 please try tavos IO clone your business
1:19:54 self
1:19:55 maybe maybe I'll try that out
1:19:59 okay this
1:20:04 [Music]
1:20:08 page cat gp4 is still the
1:20:12 best um but we now have four GPT 4 class
1:20:18 large language models out there Claude 3
1:20:22 Gemini Pro 1.5 and inflection 2.5 for
1:20:26 pi um GPT 4.5 is rumored to be coming
1:20:31 we'll see see if we get anything this
1:20:33 week or whenever doesn't [ __ ] matter
1:20:35 really
1:20:37 um you should start with chat GPT
1:20:40 absolutely and here's the deal when you
1:20:42 when you go back to this uh this
1:20:50 site here's chat GPT
1:20:54 Plus actually I should change that to
1:20:56 chat GPT plus but um each one of these
1:21:00 bullet points is a whole other thing to
1:21:02 learn it's a whole other capability
1:21:05 inside chat GPT you should learn them
1:21:08 all you should learn what they make
1:21:10 possible you should learn what they do
1:21:12 you should make yourself custom gpts I'm
1:21:15 in bed with AI nightly 3:00 a.m. me Bing
1:21:19 is there life on Mars being yes go to
1:21:23 sleep that's awesome that's awesome all
1:21:26 right let's go let's go do some uh mid
1:21:27 Journey stuff okay if you haven't done
1:21:29 anything in mid Journey what you do in
1:21:31 mid journey is you go slash
1:21:35 imagine a
1:21:40 cyborg
1:21:43 walking down a Neo
1:21:48 Tokyo Neo
1:21:51 Tokyo
1:21:54 Street oh wait slash oh I spelled
1:21:57 imagine
1:22:02 wrong walking down a neotokyo
1:22:09 street
1:22:11 at midnight in a
1:22:16 rainstorm and then I'm going to go dh-
1:22:20 AR for aspect ratio 16 colon 9 and if
1:22:25 you haven't made anything with mid
1:22:27 Journey you're like good God are you
1:22:29 right are you making images with command
1:22:31 line prompts yes that's the shitty
1:22:35 reality right now all
1:22:37 right so I'm going to make this and
1:22:41 we'll see what it looks
1:22:46 like Devon sounds very interesting it
1:22:53 does
1:22:56 Empower pixels exactly he never
1:22:58 mentioned perplexity it's more accurate
1:23:00 than all the
1:23:02 others yeah but I wouldn't I wouldn't
1:23:05 call
1:23:06 perplexity gp4 level good it may be Jeff
1:23:12 Bezos invested a [ __ ] ton of money in it
1:23:15 and they might be making their own
1:23:16 models and they might be getting better
1:23:18 but they're not there yet what
1:23:19 perplexity is better at than the other
1:23:21 ones is research and citations so it's
1:23:25 more accurate it doesn't mean it's
1:23:27 better it's just different it's just a
1:23:29 different
1:23:30 tool um when perplexity is in that class
1:23:34 I'll talk about it all [ __ ] day long
1:23:36 ask anyone in here I I take [ __ ] off
1:23:38 that list and put it on the list all the
1:23:40 time but those are the four right now
1:23:42 that I would consider getting there and
1:23:46 you can also ask this Lisk Gemini I made
1:23:49 I made fun of Gemini for a year and a
1:23:51 half
1:23:53 now they they got their heads up their
1:23:55 butts over there but they're getting it
1:23:57 together all right
1:23:59 so so here's our our neotokyo dudes
1:24:04 walking down the streets those are fine
1:24:06 they look good but let's say that we
1:24:08 wanted the
1:24:09 neotokyo our cyborg to look like a
1:24:12 particular cyborg so mid Journey added
1:24:15 this ne
1:24:16 new thing called CF so what I'm going to
1:24:20 do is I'm going to
1:24:22 upload a file
1:24:25 I'm going to go find this thing called
1:24:32 [Music]
1:24:39 Genesis where is it
1:24:43 Genesis name contains
1:24:47 Genesis that is that's my my
1:24:51 AI dude so I just hit the plus button to
1:24:56 upload it so now it's uploaded so now
1:24:58 what I can do is rightclick this this
1:25:00 is this is [ __ ] ridiculous you
1:25:03 shouldn't have to do this but this is
1:25:04 how it is right now so I'm going to copy
1:25:06 the
1:25:08 link no I won't do that just now no I'm
1:25:11 going to copy this
1:25:17 prompt and then I'm going to go slash
1:25:23 imagine
1:25:26 I'm going put the prompt in there and
1:25:27 then I'm going to go dash dash CF
1:25:31 CF which stands for character
1:25:35 reference
1:25:37 ref and then I'm going to put a space
1:25:39 and I'm going to rightclick and copy the
1:25:42 link to this image so what you do is you
1:25:45 upload an image to Mid joury or to
1:25:49 Discord I copied the the URL for that
1:25:53 image and now I'm going to paste it
1:25:55 after a space after the CF and so what
1:25:58 that so what this going to do is going
1:25:59 to give me the exact same prompt the
1:26:02 exact same aspect ratio but it's going
1:26:04 to use this image as a character
1:26:07 reference for what gets created let's
1:26:10 see how it
1:26:12 does and then I should be able to just
1:26:16 keep making all sorts of different
1:26:18 prompts using that same CF perplexity
1:26:21 does allow the user to choose their
1:26:23 favorite model including
1:26:25 gp4 yep yeah I get it listen there are a
1:26:29 lot of good tools that are using those
1:26:32 those bigger foundational models I'm the
1:26:35 the ones that I'm talking about are the
1:26:37 foundational models
1:26:40 so but listen I I am a fan of perplexity
1:26:44 I got a free Pro account when I bought
1:26:46 my rabbit
1:26:47 R1
1:26:52 yay but now look that sort of looks like
1:26:56 me from the
1:26:58 side that's pretty
1:27:00 slick these two are from behind so those
1:27:03 are weird that's from the
1:27:05 front so
1:27:08 let's let's do variations so the way you
1:27:11 do uh mid Journey there's these U's and
1:27:14 V's so U's are upscale V's are variation
1:27:17 so I'm going to do a variation of number
1:27:19 one it'll give me four more of number
1:27:21 one and then I'm going to do an upscale
1:27:24 of number two cuz that sort of looks
1:27:26 like
1:27:28 me and then these two are just shitty so
1:27:30 let's see a cyborg walking down the
1:27:33 street so I'm going to
1:27:37 say um let's copy this whole thing so
1:27:40 that's the whole prompt so I'm going to
1:27:42 do slash
1:27:44 imagine I'm going to put the whole
1:27:46 prompt in there I'm going to say a
1:27:48 cyborg
1:27:50 walking toward the camera
1:27:57 on a New York City street at midnight in
1:28:00 a
1:28:06 rainstorm all right so that's doing its
1:28:09 thing but look at that I mean it's it's
1:28:11 not quite the same cyborg but it's got
1:28:13 the right hair it's got the right nose
1:28:15 the right chin the right jacket the
1:28:17 little cyborg looking
1:28:19 doad right that's not
1:28:22 bad not bad at all oh people and then
1:28:26 here's the
1:28:33 variations can you repeat how to set
1:28:35 that up absolutely for you Cam katkin
1:28:38 absolutely cam katkin from Cleveland she
1:28:41 gets anything she wants oh yeah mhm
1:28:43 that's what we're going to do yep well
1:28:45 we're going to do it all right so here's
1:28:47 how you do
1:28:49 it you first
1:28:52 upload the image that you want to use as
1:28:55 the character reference
1:28:57 right it it apparently really helps if
1:29:02 it's a mid Journey generated
1:29:05 image don't know why but it does I feel
1:29:09 like you could always do this by pasting
1:29:11 in a URL of image even could add
1:29:14 multiple reference images yeah I think
1:29:16 that's right I just i' I I am I am I am
1:29:20 teaching you how to do something I've
1:29:22 done like four times so so I really
1:29:25 don't know your prompt sounds like Blade
1:29:27 Runner yeah
1:29:29 exactly all right so let's see what we
1:29:31 got here oh that oh this one's so good
1:29:34 upscale
1:29:37 three that's so good upscale
1:29:41 creative look at that that's so
1:29:47 cool it was a rainy night in it was a
1:29:50 rainy night in
1:29:53 Tokyo she she looked at me a scant from
1:29:56 the alleyway she said what are your
1:30:00 [Laughter]
1:30:08 qualifications
1:30:11 okay so here's how you set this up from
1:30:13 scratch you
1:30:15 go you put in your
1:30:19 normal you put in your normal prompt so
1:30:22 I'll just do I'll just copy this and
1:30:25 then we're going to change it so so you
1:30:27 put in a normal prompt so
1:30:29 slash
1:30:32 imagine and then I'm going to put in the
1:30:34 prompt and at the end I have-- AR for
1:30:37 aspect ratio 16 colon 9 but I'm going to
1:30:40 say instead of a cyborg walking toward
1:30:42 the camera I'm going to say a
1:30:46 cowboy walking toward the
1:30:49 camera on
1:30:53 a
1:30:55 Wild
1:31:00 West
1:31:01 Street
1:31:03 West Street at High
1:31:09 Noon leave it at high in so completely
1:31:12 different thing now I want this Cowboy
1:31:16 to use to to be this character so at the
1:31:20 at the end of uh the aspect ratio I'm
1:31:22 going to go space Dash Dash CF CF for
1:31:27 character reference
1:31:28 space and then I'm going to go up to
1:31:31 that image that I
1:31:32 uploaded rightclick it copy the
1:31:38 link my qualification is I am an
1:31:41 irregular you know what's cool someone
1:31:43 told me today and and you should do
1:31:45 this someone told me today they put
1:31:48 under their LinkedIn profile that they
1:31:50 remember the AI salon and they talked
1:31:52 about it in a job interview and job
1:31:54 interview people were very impressed ha
1:31:57 don't know if they'll get the job yet
1:31:59 but okay
1:32:01 so prompt the cowboy walking toward the
1:32:03 camera D- AR aspect ratio 16 by9 space--
1:32:08 CF space and then you paste in that URL
1:32:12 so I right-clicked the the ref image and
1:32:15 and put it in there so now we should get
1:32:19 you know some version of you know these
1:32:22 things but it's a cowboy in a wild West
1:32:25 Town rather than a a cyborg in a thing
1:32:29 and it
1:32:31 should have some sort of cool glasses on
1:32:33 and some sort of blue ice
1:32:36 eyewear something like that comic book
1:32:40 Creations going to be off the hook I
1:32:41 know Sherry D it's so cool right look I
1:32:44 can even the shape of the head right now
1:32:46 I can even see it's got my head
1:32:51 shape yep look at that
1:32:55 now is it perfect no is this ref only
1:32:58 possible in desktop or is it also on
1:32:59 mobile if you it's anywhere where you
1:33:03 can do if you're if you're in um oh boy
1:33:08 did it make some bizarre [ __ ]
1:33:11 bodies this one's pretty cool though
1:33:14 that number one let's we'll upscale that
1:33:19 one not exactly a cowboy but it's it's
1:33:23 kind of a cowboy I mean it is in a
1:33:24 western town and he's got blue eyes
1:33:28 that's kind of cool actually it's got my
1:33:30 hair it's a scowly version of me scowly
1:33:34 jelly version of
1:33:36 me um no you should be able to do this
1:33:38 on on mobile you just go into the
1:33:40 Discord app and just do now here's the
1:33:43 thing about mid Journey you have to pay
1:33:45 for it it ain't it ain't free it's like
1:33:47 10 bucks a
1:33:49 month but these are the first guys to to
1:33:53 make it relative ly easy to do character
1:33:55 references if an employer wants my AI
1:33:57 reference I'm sending them to
1:34:00 Kyle but you know what my response is
1:34:02 going to be oh they're in
1:34:05 irregular I don't think it's going to
1:34:06 bode well for
1:34:11 you oh man all right I'm gonna get out
1:34:16 of here
1:34:17 people I'm going to get out of
1:34:23 here
1:34:25 freaking lasers in his head now we went
1:34:28 from Blade Runner to Westworld exactly
1:34:32 exactly but I mean you know go into mid
1:34:36 Journey make a couple of characters if
1:34:38 you like a character you know make a
1:34:40 cute little animal or make a a cute
1:34:43 character or a scary character or
1:34:45 whatever and then just start putting
1:34:47 that character in other scenes and in
1:34:50 other
1:34:51 styles and there's also a waiting if you
1:34:54 go to the Discord for Mid Journey go to
1:34:57 the announcements there's D Das um C ref
1:35:02 is the character reference and then you
1:35:04 can go D- CW which is character weight
1:35:08 and you can go from zero to 100 and if
1:35:12 you go from zero if you if it's at zero
1:35:15 it just keeps the face everything else
1:35:17 changes if you go to a 100 it I think it
1:35:19 keeps like the hair the costume you know
1:35:22 what the person's wearing every
1:35:25 and so you can you can you can dick
1:35:29 with what parts of the character it
1:35:32 keeps or you can just do the default
1:35:34 like I just did there again I haven't
1:35:37 played with this a lot so I don't really
1:35:38 know what works Peter Kaminsky if you go
1:35:41 to the showand tell area in the AI Salon
1:35:45 Peter kaminsky's put up a bunch of test
1:35:47 stuff and what he did to get
1:35:50 there so you should you should uh follow
1:35:54 him follow him great info tonight Kyle
1:35:58 thank
1:35:59 you thank you thank you thank you all
1:36:01 right
1:36:03 so 127 of you in here please
1:36:08 please if you don't already follow my
1:36:11 channel that would be
1:36:13 swell uh if you're so inclined to want
1:36:16 to support the channel subscribe to the
1:36:19 lives or you can buy the video series in
1:36:21 one of the
1:36:23 corners um you can also join the AI
1:36:26 Salon the AI Salon is
1:36:29 at that URL the
1:36:32 salon. go check out our values and what
1:36:35 we're all about thank you Empower pixels
1:36:37 Nicholas thank you so much thank you
1:36:39 thank you clean copper head oh yes join
1:36:42 my team join my
1:36:45 team um thank you to the regular thank
1:36:48 you to the mods for
1:36:51 modding um thank you to those of you
1:36:54 giving gifts right now appreciate that
1:36:56 thank you for the tappers the tappers
1:36:58 what did we get to tonight 18,000 we
1:37:00 didn't crack
1:37:01 20,000 you're a little lazy but I was
1:37:03 giving you I was giving you stuff to
1:37:05 play with so I understand it's all right
1:37:07 it's late um I think that's it I think
1:37:12 that's it so who's in here tonight Clint
1:37:14 Copperhead good to see you Lord digital
1:37:16 Gods thank you very much very kind of
1:37:18 you sir very kind of you indeed little
1:37:21 lazy Tams I was a Tapper thank you Diana
1:37:26 have a great night Kyle and everyone
1:37:27 thank you a human being Lily beans 20K
1:37:31 oh we just hit 20K you guys are
1:37:33 awesome did first three still stuck on
1:37:36 introduction because AI is evolving I
1:37:40 know Vegas Moxy thank you so much hello
1:37:42 from Dy where's
1:37:44 Dy um Down Yonder oh how's Down Yonder
1:37:49 Aloha from Honolulu hello from Detroit
1:37:52 we did it we hit 20,000 uto good to see
1:37:54 you Danielle thank you as always you're
1:37:57 amazing Simon Says Tech against the
1:38:01 norm Brandon thank you so much Sherry D
1:38:04 you're awesome thank you so much
1:38:06 Amelio's wife was on time I don't think
1:38:08 she was on time but I was late tonight
1:38:11 see she knows she knows how to she knows
1:38:13 how to play the professor it's just it's
1:38:16 it's she's a
1:38:19 professional Empower pixels thank you so
1:38:21 much Lily Bon Dian anah good to see
1:38:25 you all right against the norm gritty
1:38:28 wisdom awesome to see you I've I've seen
1:38:31 you in here a number of times now that's
1:38:33 very cool Angelique keep coming back
1:38:36 keep coming back to these it's important
1:38:37 stay in the conversation stay in the
1:38:39 conversation I know it's [ __ ]
1:38:40 overwhelming I know it's exhausting I
1:38:42 know you feel like you're completely
1:38:43 behind I know you feel like you're
1:38:45 completely
1:38:47 clueless you are but we all are so does
1:38:51 that make you feel better there's misery
1:38:53 and compass wait there's misery loves
1:38:55 company it there's love in misery with
1:39:01 company peace out everybody thank you
1:39:04 Vegas Moy thank you V dancer I really
1:39:06 appreciate that you're the best you make
1:39:09 AI fun hey you know if the robots are
1:39:12 going to take us down we might as well
1:39:13 have a good time on the way
1:39:15 right I on a on a on an actual serious
1:39:19 note I think us having fun with
1:39:22 AI
1:39:24 is actually bending the use of AI in a
1:39:28 positive direction if ever so slightly
1:39:31 so the more of us that can actually
1:39:33 embrace it have fun with it it's bending
1:39:35 it toward the good and over time that's
1:39:39 going to make a huge difference that's
1:39:40 what I firmly believe firmly firmly
1:39:43 firmly believe it uh we're at 24,800
1:39:47 you'll never make 25,000 oh no we're
1:39:49 27,000 we're well past that peace out
1:39:52 everybody I'll see you tomorrow