aiLL 9/3/23 - New AI Tools & How to Use Them Today

Description

Chapters

Transcript

0:29 welcome everybody Dr J was shaking
0:35 Matthew Craig official what's happening
0:37 welcome welcome Happy
0:40 Sunday I think it's Sunday
0:42 I've been working my this filter is way
0:45 wrong I look I I look if you've ever
0:48 been to like a newscast where you see
0:51 someone who is put in makeup this is
0:54 what they look like when you run into
0:55 them in the hallway hang on let me let
0:57 me change the effect oh wait that was
0:59 not the right one let's see enhance
1:03 enhanced effects no that's not right uh
1:07 and hands
1:09 are enhanced filters
1:12 let's say let's do this no no no no
1:17 that's not too bad
1:19 yeah that's good that kind of kind of
1:23 more like skin
1:25 vintage yeah yeah yeah
1:31 youth
1:33 Twilight we'll go Twilight Twilight
1:37 twine twine
1:40 it was a Jim Gaffigan joke
1:43 got it gotta go down
1:45 the hardware store to get twine
1:48 [Laughter]
1:49 oh welcome everybody
1:53 did you know there's a Santa Claus
1:55 filter you can pretend to be Santa maybe
1:57 if we get close to December I'll I'll go
2:00 I'll go all Santa so uh welcome everyone
2:03 oh holy crap there's a lot of people
2:04 here my name is Kyle Shannon this is the
2:06 AI learning lab uh here's a whole bunch
2:08 of cool pictures that uh I made in
2:11 ideogram if you don't know ideogram
2:13 there's uh some cool stuff there you can
2:17 actually make images now with words on
2:19 them words words words
2:22 so what we do here
2:24 my name is Kyle Shannon I like talking
2:26 about this generative AI stuff
2:28 um
2:30 we got some some some folks show up here
2:33 regularly they're called the Irregulars
2:35 they're they're down in the comments
2:36 they'll help at some point I'm going to
2:37 make them uh moderators uh I like to
2:40 talk about all this uh generative AI
2:42 stuff I'm an entrepreneur I've started
2:45 lots and lots of companies uh my current
2:47 gig is I'm founder and CEO of a company
2:49 called storyvine
2:50 and
2:52 um
2:53 in the mid 90s I co-founded a company
2:56 called agency.com and I bring that up
2:57 just because the early days of the world
3:00 wide web and the early early days of
3:02 this generative AI stuff there's a lot
3:03 of parallels it's feel it's feeling very
3:05 deja vu all over again so so we'll talk
3:09 about that uh if you have questions and
3:11 comments pop them in the comments below
3:12 I do my best to answer everything
3:16 um the other thing that we'll do is
3:18 we'll sort of kick off tonight talking
3:20 about if you're new to this AI stuff if
3:23 you're just sick of people like you know
3:25 you should try Ai and you're like I
3:27 don't want to try AI what is this AI
3:29 anyway what is internet anyway
3:33 um
3:33 we'll walk you through that we'll get
3:35 you up to speed that's part of what this
3:36 channel is about part of it's just about
3:38 talking [ __ ] uh I like to make funny
3:40 voices uh I don't know we have fun we
3:43 have a good time here so we'll hang out
3:44 it's Sunday night you're not doing
3:46 anything else might as well hang out
3:47 here right okay so if you're brand new
3:50 to chat EBD thank you Matthias
3:53 um
3:54 go there go to chat.openai.com that's
3:57 the official chatgpt website go there
4:00 and if you don't quite know what to do
4:01 there you can sort of ask it anything uh
4:04 but you can start out with what I call
4:05 parlor tricks have it write you a
4:07 country song about and describe your dog
4:09 or a significant other or have it write
4:12 you a marketing plan or just just tell
4:15 it to do something that seems random and
4:17 like it couldn't do and watch what it
4:19 does and then go to this thing
4:22 prompts.chat that's a document and that
4:25 will teach you how to talk to these
4:26 things these are called large language
4:28 models that one's chat EBT it's right
4:30 now it's the best uh but there there are
4:33 a lot of uh companies nipping at their
4:35 heels right now
4:37 um but it's good I'll talk about these
4:39 other ones in a bit
4:41 um a couple other things to update
4:44 I think
4:46 now just to be clear I'm an old guy on
4:49 Tick Tock right so I'm figuring this
4:51 [ __ ] out I think I got a store set up so
4:55 that there's products so I don't know
4:57 what your interface looks like mine has
4:59 a little button called shop on it and
5:02 um there's a women's t-shirt a men's
5:04 t-shirt and a tote bag I just put them
5:06 in there kind of as a test there's only
5:09 one design right now uh it's uh
5:13 um I'm not weird I'm irregular so it's
5:16 for the Irregulars so if the Irregulars
5:17 are here if you would just go check the
5:19 store out and let me know like is it
5:21 there does it look functional you don't
5:23 have to buy anything right now if you
5:24 want to fine but but um so anyway that's
5:27 there
5:28 the other thing that I did if you go to
5:30 the YouTube channel YouTube uh AI
5:33 learning lab Dash TT
5:36 uh regular at AI learning lab was taken
5:39 so the dash TT is for tick tock
5:41 um that's archives of these videos I got
5:45 all the way through
5:47 June
5:48 is uploaded there and I actually figured
5:51 out the prompting for generating
5:53 chapters so I got in like I don't know
5:55 25 or 30
5:57 um descriptions and chapters in there so
5:59 you can go see that
6:01 um okay store looks great awesome
6:02 terrific hey Kyle hey Irregulars hey
6:04 Stevo with shaking Man story looks great
6:07 awesome yeah it's working fine okay good
6:09 awesome
6:10 um so I guess I know what I'm doing I
6:11 guess I guess now I need to figure out
6:13 some products
6:14 um I'm gonna do a you can make money
6:16 with chat gbt huh
6:21 ha
6:22 uh if you're wondering how to make money
6:24 with you at gbt we can talk about that
6:26 feel free to ask comments below
6:29 um okay let's see let's let's jump in
6:32 and talk about this stuff
6:34 okay
6:38 so as I said before
6:40 that's the official chat GPT website
6:42 the reason I have three of these here
6:45 these tools right now a lot of these AI
6:48 tools as sexy as they are and as
6:50 powerful as they are they're what I like
6:52 to call it's a technical term janky
6:55 pieces of [ __ ]
6:57 um because we're very early in whatever
7:00 this why is that not zooming smoothly
7:05 there we go
7:07 um
7:08 we're just very early so so even though
7:10 machine learning and ai's been around
7:12 for a while this generative AI this this
7:14 idea of chat GPT and things like it
7:16 relatively new so a lot of the software
7:18 is very very
7:19 um
7:20 simple and doesn't do a lot of
7:22 sophisticated stuff and and because of
7:24 that you you kind of have to play with
7:26 different ones so they're good at
7:27 different things
7:29 so chat gbt is is
7:32 um really good at writing so gpt4 in
7:34 particular if you pay the 20 bucks a
7:36 month here you get gpt4
7:38 um you also get this thing called
7:39 Advanced data analytics which is
7:41 mind-blowing you also get plug-ins and
7:44 they're going to give you early access
7:45 to new technology as they roll it out
7:47 they're they're the sort of
7:49 um 800 pound gorilla right now others
7:51 are coming but it really is the best
7:53 Claude is a bunch of ex-openai people
7:57 that started this company called
7:58 anthropic what's cool about Claude is
8:00 it's got a hundred thousand token
8:02 context window which is about 75 000
8:04 Words which basically means this you
8:07 could paste a 50 000 word novel into the
8:10 prompt and tell it to tell it to write
8:12 you an epilogue for your novel it'll
8:14 just do it
8:15 um claude's pretty remarkable
8:17 um so so if I if I have big documents or
8:20 I have something big I want to go
8:21 through I use Claude
8:22 um Bard is Google's entry into this
8:24 space they just made the top three
8:27 um what are the criteria for the top
8:28 three I don't know get a wild hair up my
8:30 ass and I decide what's in the top three
8:32 right now these are the three that are
8:33 the most useful for me
8:35 um who's at the Kids Table kids table
8:37 still useful still worth playing with
8:42 um and exploring Bing is Microsoft's uh
8:46 large language model it's it's actually
8:48 built on top of gpt4 so it's it's like
8:52 chat GPT but it's it's connected to the
8:54 internet you can make images there you
8:57 can
8:58 um analyze images there
9:01 perplexity.ai is a really good research
9:04 tool labs.perplexity lets you play with
9:06 metas llama 2 models and their new code
9:09 llama model
9:11 which in terms of coding actually just
9:14 beat gpt4 so if you're having
9:17 frustrations with gpt4 with coding try
9:20 out code llama then Pi dot AI they have
9:23 an app and a website and it's what we
9:26 call here in the lab chatty GPT it will
9:30 just talk your ear off so if you want to
9:32 get work done don't use Pi but if you
9:33 don't have a conversation it's very very
9:35 interesting and then poe.com's got a lot
9:38 of bots and things like that that you
9:39 can play with I use Poe today because I
9:43 got rate limited out of Claude
9:45 because because I was doing three hour
9:48 YouTube or Tick Tock transcriptions
9:51 YouTube transcriptions and I got booted
9:53 out so I was using Poe to do it uh togas
9:56 thank you very much I appreciate that
9:58 um so so Poe can be very very useful
10:00 then finally the pictures the pictures
10:02 but how do we make pictures Kyle tell me
10:04 how to make pictures I want to make
10:06 pictures it's all about the pretty
10:08 pictures please please stop talking for
10:11 the love of God shut up just shut up can
10:15 I just shut it zip it
10:18 all right
10:20 um Mid Journey's the best ideogram is
10:23 eight days old maybe nine days old it
10:25 does text this is a startup they raised
10:29 16.5 million dollars led by Andresen
10:31 Horowitz it's an X
10:34 Google
10:36 I think it's the next Google person
10:38 started that and they figured out how to
10:41 do text before the big players like
10:42 stable diffusion and mid-journey
10:44 mind-blowing and that's that's what's
10:46 going on in this business right now we
10:50 don't know where the Innovation is going
10:51 to come from and all these companies are
10:53 you know sort of fighting to one-up each
10:55 other so we're the beneficiaries of that
10:57 Leonardo and clip drop are both based on
11:00 top of uh stable diffusion Leonardo's
11:03 just a really good website uh where you
11:05 don't have to install your own [ __ ] and
11:07 clip drop is from stability AI those are
11:10 the people that make stable diffusion
11:11 there's some pretty cool tools there
11:13 and then prompt book will teach you how
11:16 to talk to these things and and and sort
11:18 of teach you how powerful they are and
11:20 then finally futurepedia is a big long
11:22 list of uh what do you what do you call
11:24 it
11:25 um AI tools I know I never remember that
11:28 by the way for the Irregulars uh drinks
11:32 tonight uh a cheap wine we bought at
11:34 Safeway hold
11:38 grocery store wine
11:41 oh yeah it's solid and then uh Rocky
11:44 Mountain tap water
11:48 filtered Rocky Mountain tap water
11:50 all right so let's get to the questions
11:53 pop your questions in below
11:55 and I will do my best to answer them I
11:58 will get behind that's one of the things
11:59 I do I also cuss I make funny voices I
12:02 don't know who knows who knows
12:06 let's see what we got I'm hesitant wait
12:09 um
12:10 let me flip this around let's get let's
12:13 get things set up here all right all
12:15 right
12:16 yeah
12:18 my lactators um
12:22 all right I'm headed hesitant to
12:25 discredit anyone because I lost more
12:27 money in cryptocurrencies than I care to
12:29 admit story looks great
12:32 um have you heard of Kit or pay hip for
12:34 cheap e-commerce store and easy setup I
12:37 have not I'm using printful right now
12:40 because I want to do print on demand
12:41 because I don't want to I like I just I
12:44 can't have the inventory I'm not I'm not
12:46 running that business but I thought it
12:48 would be cool to have something there
12:50 um so right now I'm using printful but
12:53 if I if I decide to ever sell
12:57 these things I gotta figure that [ __ ]
12:59 out so at some point I'll figure it out
13:02 um all right oops sent twice that's okay
13:05 Kyle in a regular store looks great hey
13:07 Marie deal shaking Matthias
13:11 ptld as Portland I suppose uh this stuff
13:16 reminds me of Lincoln Logs Erector Set
13:18 Legos Radio Shack 64. uh in one
13:21 chemistry set
13:23 what stuff reminds you of it oh like all
13:25 the AI tools right now they're totally
13:27 like that they're totally just janky ass
13:29 you gotta glue them together and it's
13:31 they're the most like um what was the
13:34 Erector Sets right yeah the Erector Sets
13:37 because because they have the little
13:38 nuts and bolts in the little cheap
13:40 wrench and when it would slip you would
13:42 cut your knuckle open on the Erector Set
13:45 that's what using AI is like right now
13:47 you're like oh this is cool
13:51 that sucks
13:56 oh good lord lord lord lord hearing
14:00 truth what's happening Emilio's wife was
14:02 shaking please folks most of you
14:05 probably know but don't let anyone guide
14:07 or help to buy crypto
14:09 yeah so so we had a scammer so the AI
14:13 learning lab had someone that copied the
14:15 channel and then was trying to sell
14:17 members crypto stuff if anyone is
14:20 pretending to be me or anyone Reach Out
14:22 reaches out trying to sell you crypto
14:24 [ __ ] or even just getting you to log in
14:26 with your crypto wallet to their stupid
14:28 [ __ ] site don't do it
14:30 um so I don't I don't know if that's
14:32 going on again but it but yes what
14:34 Winston says
14:35 just be very careful it just crypto
14:39 stuff is if if you think AI tools are
14:41 jankier than [ __ ] crypto tools are
14:44 really [ __ ] special
14:48 you have to be very technically Savvy
14:50 and even then you know I know I know of
14:53 a lot of people that
14:55 had a lot of money just
14:57 vacuumed out of their accounts because
14:59 they clicked on one link
15:02 and you're like but Kyle it shouldn't be
15:04 that easy to steal people's money I
15:06 agree
15:07 and what the crypto Bros will say is not
15:09 your wait not your keys not your crypto
15:14 um yeah so there's something yeah anyway
15:17 just
15:18 anyone tries to sell you crypto [ __ ]
15:22 do not do it
15:24 it is not the road to riches well except
15:26 for them it is it's the road to riches
15:28 for them
15:30 all right let's see what questions do
15:33 you have Hi boss how are you today well
15:35 Johnson royalty I'm doing well I'm doing
15:38 well I was busy all day I was
15:40 downloading videos uploading videos
15:42 using po to Poe and Claude to write
15:46 YouTube descriptions with chapter things
15:48 that was pretty interesting I'll talk
15:50 about that in a bit
15:52 um then I was working on the e-commerce
15:55 stores uh store well just whatever all
15:59 that [ __ ] so I was busy all day today
16:00 and then I mowed
16:02 and then I went to Costco because I know
16:04 how to live
16:05 and then what else uh
16:08 I don't know watch the little US Open
16:10 tennis Djokovic not so dramatic today
16:13 one in straight sets
16:16 three Americans are in the quarterfinals
16:18 so that's cool
16:27 someone approached me today with what
16:30 first I thought was a helpful tip and
16:32 then he offered to help guide
16:34 yeah Winston sorry about that that's
16:36 probably the scammer got a bunch of
16:38 people's
16:39 um
16:40 you know handles even though they
16:42 deleted his account he's probably just
16:43 got a spreadsheet that he's working his
16:46 way down so yeah anyone wants to guide
16:49 you to crypto tell them to [ __ ] off
16:53 I will not do that if we're gonna do a
16:56 pyramid in scheme in a pyramid scheme in
16:58 this place I'm going to call it a
16:59 pyramid scheme will be transparent
17:05 all that stuff pisses me off
17:08 all right but thank you for sharing that
17:11 that happened to you Winston and and for
17:14 letting people know yeah exactly
17:17 um how welcome three times is the number
17:20 of times I subscribed in six seconds
17:23 with my mini phone
17:28 I I
17:31 I hope you didn't subscribe three times
17:33 uh
17:35 picant but not too sassy
17:38 um what do you know about DMSO I forget
17:41 what DMSO is hang on
18:13 dimethyl sulfoxide no I don't I assume
18:16 that's not what you're talking about
18:19 um I don't know what you mean uh
18:20 explaining it and I'll I'll see if I
18:22 know anything about it I don't think I
18:23 know anything about it how meaning ho ho
18:25 ho ho welcome Hey regular share bear
18:28 with shaking okay farmer is here
18:33 what AI can create a metaverse well
18:36 there's
18:38 um
18:39 what the hell is it called
18:43 something labs
18:47 hold
18:48 58 year old brain trying to remember
18:50 something oh please
18:54 Gen X meltdown
18:56 um
19:10 dream Fusion blockade Labs that's what
19:13 it is
19:15 hi box
19:17 all right so so this isn't exactly
19:19 creating a metaverse but this is pretty
19:21 [ __ ] cool so I'll show you this
19:24 I have read and I accept your terms by
19:26 the way oh you wanna you wanna see
19:28 something fun so I went to so where I am
19:30 is I'm at Skyblock
19:32 skybox.blockadelabs.com and then they've
19:34 got a what's new since you last logged
19:36 in and they have a I have read and
19:39 accepted the terms of service and the
19:40 privacy policy
19:42 I I am not a manual reader I have never
19:44 in my life read terms of service and I
19:47 assume most people are like that some
19:48 people do I don't
19:50 um you know what I do now I go to that
19:53 site
19:56 and so there's the terms of service and
19:59 I'm just going to copy them all
20:07 I'm gonna copy them I'm going to go over
20:09 to Poe
20:10 I'm going to say new conversation so
20:12 pose just a large language model but
20:14 I've got the Claude 2 100
20:17 000 100 000 token so I can just paste
20:20 that whole damn thing in there
20:22 so let's see
20:24 um
20:25 [Music]
20:30 um based on these terms of service
20:36 colon paste there's all the terms of
20:39 service
20:41 please
20:43 give me a concise bulleted
20:50 list of what I am
20:53 agreeing to that is material
20:59 in your POV
21:04 keep each bullet
21:08 two seven to ten words
21:13 all right
21:16 feed this thing for us
21:20 not that I really give a [ __ ] I'm still
21:21 gonna accept it but this is just kind of
21:24 a cool use it's one of the things that
21:26 these tools are good at is understanding
21:29 uh text and reading it
21:32 you own any original assets created with
21:35 the service blockade can use and share
21:37 your content for promotion so you own
21:39 them but you give them the rights to
21:41 play with them you must be 18 years of
21:42 old fees are charged monthly or yearly
21:45 and are non-refundable so don't do the
21:48 yearly thing in case something better
21:50 comes along
21:51 Services can be terminated or suspended
21:54 at any time you indemnify blockade
21:56 against any losses from your use
21:58 blockades liability is limited to 50
22:01 bucks or fees paid
22:05 um disputes must be resolved in
22:06 individual arbitration class action
22:09 waiver must uh arbitrate claims
22:11 individually blockade can change terms
22:14 at any time so it doesn't seem too
22:15 egregious there you go bang so so that's
22:18 a fun thing to do
22:20 um
22:22 all right so we're going to accept those
22:24 and I do not want to show this message
22:25 again
22:26 all right
22:32 so
22:36 so we have a cool 3D World here right
22:38 well
22:40 what you can do at blockade Labs is you
22:42 can just say
22:44 started you can just say
22:47 um
22:48 let's see
22:51 dystopian
22:54 hellscape
22:57 in a water logged
23:03 um
23:04 dry lake bed
23:06 [Laughter]
23:11 um
23:11 um
23:14 elements from a
23:18 um
23:19 hippie Festival
23:25 are strewn
23:27 about
23:29 and then we're gonna do let's see
23:33 techno Arts I will we'll do realistic
23:36 we'll see we'll do realistic all right
23:37 so we're going to generate so this thing
23:40 will now generate this 3D World so this
23:42 isn't exactly generating a metaverse but
23:44 this is this is the beginning of the
23:47 kinds of things that we're going to be
23:48 able to do where you're going to
23:49 literally be able to just talk something
23:51 into existence
23:54 um I know that
23:56 um
23:57 that there's some folks working on
24:00 um rolling these kind of things into
24:03 well that's boring as [ __ ]
24:06 but you know gives us what we want so
24:08 let's do let's do something else we'll
24:10 do
24:10 um
24:15 [Music]
24:22 infinite Cosmos full of exploding
24:25 planets and spaceships
24:30 and then we'll do
24:36 [Music]
24:39 Tech Noir cyberpunk will do that
24:42 this one should be more interesting
24:44 um
24:45 so what what blockade Labs actually is
24:48 is this is just creating a kind of a
24:51 flat image that is is done in such a way
24:54 that they can map it to a sphere
24:56 um so so it makes this thing that looks
24:58 like you're in a 3D space but you're
24:59 actually not
25:00 and then there's our cool spaceship
25:03 looking thing with exploding planets
25:05 better give me an exploding planet
25:08 nah no exploding planets I want an
25:12 exploding planet
25:17 I I want AI to be perfect it's not but
25:23 anyway still pretty flipping cool so
25:25 this is skybox.blockadelabs.com
25:28 um
25:30 you can take these and put them into
25:33 Unreal Engine and put objects in them
25:36 and then you're able to navigate them
25:38 right now it's not super sophisticated
25:40 it's not creating a 3D world but I know
25:42 that folks are working right now on
25:45 being able to put together 3D
25:48 what are they called procedural
25:51 like procedural scenes within Unreal
25:55 Engine so so this stuff's going to get
25:56 very real very fast so
26:00 blockade Labs is one to play with the
26:02 other thing that I would do is I would
26:03 go to because I'm not really keeping up
26:05 on this stuff because I can't go to
26:07 futurepedia.io and look up text to 3D or
26:11 just look up 3D or worlds just find a
26:14 category in there that looks like the
26:16 kind of [ __ ] you're looking for and uh
26:19 yeah there you have it but pretty cool
26:21 isn't it
26:24 pretty cool pretty cool you know what I
26:26 mean it's all right it's all right it's
26:28 all right it's all right right right
26:31 right
26:31 all right let's see
26:35 how can we use chat EBT to make 400
26:37 returns in the stock market uh have it
26:40 teach you how to do technical analysis
26:43 and start doing day trading and pray you
26:46 don't lose your ass when you start
26:48 leveraging your
26:49 your trades
26:51 don't listen to me I have a degree in
26:53 acting buddy
26:55 you know life is tough it's tougher when
26:57 you're stupid all right see I'm getting
26:59 better I'm getting I'm not getting quite
27:01 so thrown when someone throws the damn
27:02 hat on me
27:06 all right what did you do with the chat
27:09 for coder
27:12 huh
27:14 what did I do with it like why is it
27:15 screwed up and not as good as it was
27:17 I didn't do [ __ ] with it I don't work
27:19 for open AI
27:23 um I haven't noticed it's gotten worse
27:25 but I haven't been uh playing with code
27:27 code with it for all day every day like
27:30 some people are and saying it's getting
27:32 worse but I don't know I don't know
27:34 what's going on with it
27:36 um what I can tell you is that these
27:38 tools
27:45 all of these tools are they're kind of
27:48 almost all of them they're flying the
27:51 plane as they're building it so when
27:54 [ __ ] just changes it's cause [ __ ] just
27:57 changed because they had to deal with a
27:59 lawsuit or they had to deal with
28:00 something broke or they had to deal with
28:02 you know whatever might be happening
28:04 crypto Joe just joined welcome crypto
28:06 Joe
28:08 um
28:09 please do not get any help from anyone
28:11 with cryptos online agree with Winston
28:13 again
28:15 uh bakala gift shop it's annoying people
28:17 are scamming and pretending to be others
28:19 than themselves it's very annoying vodka
28:22 gift shop very annoying but we did have
28:25 that we did get rid of they had a site
28:27 that cloned all of my videos and looked
28:29 exactly like mine and then they would
28:32 hang out on lives and when I would shut
28:34 down my lives they would reach out to
28:36 people that were on the live and say hey
28:38 it's Kyle from the live [ __ ] [ __ ]
28:41 so
28:43 I think I gotta turn the air conditioner
28:44 back on it's getting a little moist in
28:46 here if you know what I mean
28:49 uh most interesting AI stocks um how's
28:53 the wine it's okay it's called Seven
28:55 Deadly zins so I suppose it's seven
28:59 zinfandels it's fine
29:02 isn't it
29:08 I'm not too sophisticated when it comes
29:10 to wine so it's not awful
29:12 um
29:14 like I said I have a degree in acting so
29:16 I I can't tell you what stocks to pay
29:18 attention to I'll just tell you things
29:20 that I think are interesting right now
29:22 um and and some of these are private
29:24 companies so you can't even buy stocks
29:26 anyway
29:28 um
29:29 I think in videos
29:32 fascinating I mean I don't think I don't
29:34 think their stock is at a discount they
29:36 should a trillion dollar market cap
29:38 uh and I forget they had a hundred 101
29:42 percent year-over-year earnings so
29:45 that's interesting
29:47 um
29:48 Nvidia seems to have successfully
29:50 cornered the [ __ ] Market on AI chips
29:55 they were doing game chips
29:57 and then the crypto Bros figured out it
30:00 was good for doing blockchain [ __ ] and
30:02 then they were doing all that and I
30:03 think all the other chip manufacturers
30:05 are just like [ __ ] it and they just
30:06 backed out
30:07 and then and then you know AI was like
30:10 oh this is really good for AI stuff and
30:12 what what happened with when the
30:13 Transformer came out that Google
30:15 invented in 2017
30:17 it became clear that the more more data
30:21 you threw at the at these tools and the
30:23 more gpus you threw at them the better
30:25 they got so companies like open AI just
30:28 said [ __ ] it let's get as many gpus as
30:30 we can and throw as much data at us as
30:32 we can that's why they cut a deal with
30:34 Microsoft because they have racks and
30:36 racks of servers with these big gpus in
30:38 them
30:39 um
30:39 and so
30:41 they're interesting
30:43 um about a week ago people were coming
30:44 in the thing going well you think open
30:47 AI is going bankrupt there was an
30:48 article saying they're going bankrupt
30:50 they're spending the seven hundred
30:51 thousand dollars a day and I was like
30:52 that's a lot of money but no I don't
30:54 think they're going bankrupt
30:55 well they just came out with an earnings
30:57 report that they just they just uh hit a
31:01 billion dollar billion dollar annual uh
31:05 run rate for Revenue billion seven
31:08 hundred thousand dollars a day is about
31:10 250 million in expenses billion dollars
31:13 in Revenue I think they're gonna be okay
31:16 now they're probably plowing a lot of
31:19 that into
31:21 um
31:22 other [ __ ] like gpt5 and dealing with
31:26 lawsuits and trying to cut deals with
31:29 the New York Times so they don't get
31:30 their ass sued by them after Google just
31:32 cut a hundred million dollar deal with
31:33 the New York Times you know there's
31:35 stuff going on so it's not like they're
31:37 rolling in cash over there but a billion
31:39 dollar run rate ain't bad for for uh a
31:42 company that's spending 250 million
31:45 dollars a year
31:47 um all right let's see
31:51 break waitbrifacki.com
31:55 beef oh wait beef be fake ai.com all
32:00 right I'll go look Matthew Craig
32:03 I'll bite let's go look be fake is this
32:07 some deep fake ai.com is this some deep
32:10 deep fake sort of thing
32:18 be honest we're not always our creative
32:20 best and not every moment is ready to
32:22 share so give your reality an AI lift
32:26 [Music]
32:27 be fake is an effortless and fun
32:29 introduction to AI that allows you to
32:31 transfer yeah that's like it's like lens
32:32 uh and it's like the the new one that
32:35 people are doing headshots with where
32:37 you upload your photos so it's basically
32:39 doing like a Laura model in stable
32:41 diffusion that's probably what it's
32:42 built on
32:43 they've got some sort of you know uh
32:47 dream Booth like
32:49 Tech on the back end that's training on
32:52 how to do that that's pretty cool
32:55 um there's going to be a ton of those
32:56 I'm assuming the wine is good it's not
32:58 bad Costco oh dang it dang it Jim
33:02 Tick Tock just jumped me to the bottom
33:08 uh yes yeah masako
33:13 um let's see
33:23 um
33:25 if you're the first time here put one in
33:28 the comments if you've been here before
33:29 two
33:31 and a regulars put three oh cool I like
33:33 that Matthew Craig
33:37 this stuff reminds me oh okay we did
33:40 that before
33:45 [Music]
33:47 now I went too far back AI can create
33:50 the metaverse what did I do with chat
33:53 coder
33:54 how's the wine I assume the wine is good
33:59 Costco on the weekend Kyle what's wrong
34:02 with you I'm an idiot you know that
34:07 I'm not the smartest seller you know
34:10 when I go to the Costco I like going to
34:13 the Costco we go there I take the uh I
34:16 got the Ford Flex it's like a hearse but
34:18 more like a station wagon it's like a
34:19 cross between station wagon and her so
34:21 take the Ford Flex and I go there
34:23 because you know when you need 16 jars
34:27 of 43 pound peanut butter you go to
34:29 Costco right and then I go there but
34:31 then every now and again I just walking
34:33 down the aisle I'm like yeah I need a
34:35 762 piece wrench set
34:40 have you ever been to Costco and not
34:42 spend 400 that place is just [ __ ]
34:44 wrong
34:46 any AI tool that can help me
34:48 learning I.T networking yes all of them
34:53 [Laughter]
34:55 all of them
34:57 um
34:59 chat GPT
35:01 is good at it you might want to go to
35:05 bard.google.com because it's connected
35:07 to the internet it's also pretty good at
35:09 coding
35:11 um
35:13 and just tell it that you want to learn
35:16 um
35:17 you want to learn networking uh it
35:20 networking
35:21 and especially the ones that are
35:23 connected to the internet Bard Bing and
35:26 Pi Pi dot AI
35:29 so Bing and Pi are also connected to the
35:33 internet
35:34 so what you can do with those is you can
35:35 say hey go find the latest you know an
35:39 article a blog post or or
35:42 documentation on it networking have it
35:45 design you a custom course tell it what
35:47 you know tell it what you don't know
35:49 tell it
35:51 um that you're looking for resources it
35:53 might also be interesting to go to
35:54 perplexity.ai and see what it can find
35:57 just go to these things and start asking
35:59 them what what will actually kind of
36:01 help help you get your head around how
36:03 these things work go to prompts.chat
36:06 that will talk to you this is just a
36:09 document that that's a kind of a primer
36:11 on how to talk to these large language
36:12 models most of that document is telling
36:16 it to act like a certain kind of persona
36:18 so act like a screenwriter act like a a
36:22 consultant act like an I.T professional
36:25 right an I.T networking specialist
36:28 uh and and when you tell it to do that
36:30 it gets better at doing that kind of
36:31 stuff so go there and kind of learn how
36:33 these things work and basically just
36:35 tell it to tell it what you're trying to
36:37 learn and have it create a custom
36:40 uh curriculum for you do you want to
36:42 learn it in three in four weeks four
36:44 months four days
36:46 how deep do you want to go do you want
36:47 to specialize in any area of it do you
36:49 not even know what's possible have it
36:51 tell you what's possible and say oh I'm
36:53 interested in this and I'm interested in
36:54 that tell it you know are you really
36:58 good at project management or are you
37:00 good at ideation or are you good at you
37:02 know sitting there and sort of knocking
37:04 out just you know technical
37:06 um you know things technical drudgery
37:09 some people are really good at that if
37:10 you're good at that
37:11 tell this thing that and then say what
37:14 jobs within that would I be good at and
37:17 it will tell you and then design a
37:19 course for you to do it so
37:22 um
37:23 any good in video dot IO competitors
37:27 they only give me 10 minutes for free
37:28 per week all the video tools are really
37:31 expensive so let me go look at it in
37:34 video I haven't played with it in video
37:38 dot IO
37:41 is that right
37:46 create publish worthy videos oh this is
37:48 the one that does like stock stock video
37:51 assembly you write the script and then
37:53 and then it cuts together all the stock
37:55 stock video stuff
37:57 um
38:03 [Music]
38:05 there's videos all these things are
38:07 expensive there's video
38:16 um
38:17 there's the script d d e s c r i p t d
38:23 script
38:27 dot descript AI what's the URL
38:31 descript.com
38:36 um again what I would do I don't because
38:38 I'm not following this stuff on a
38:39 day-to-day basis I would go to
38:43 futurepedia.io and look up their video
38:45 tools
38:48 um
38:49 they're decently well organized and I'm
38:51 I guarantee you there's you know 20 or
38:54 30 tools out there from what I can
38:56 gather all of the video tools are
38:58 decently expensive right now because I
39:00 think the compute time you know if you
39:01 figure what they're doing is is some
39:03 version of you know
39:05 30 or 60 frames a second and if they're
39:07 generating it even if they're just
39:10 editing it it's it's it's probably
39:12 decently pricey from an API standpoint
39:14 but go to futurepedia.io and that'll
39:17 help all right hang on I gotta go turn
39:19 on the air I'll be right back
39:30 your live is different than your videos
39:34 gosh because I'm live yeah my videos are
39:37 like hey it's Kyle Shannon with the AI
39:38 learning lab here's some [ __ ] I learned
39:39 and I try to package it in a tight
39:41 little thing here I'm just rambling this
39:43 is just more like a conversation
39:45 um
39:46 I'm figuring these lives out kind of on
39:48 the Fly and you know what I want it to
39:51 be what the community wants it to be
39:52 where's where the value is so yeah
39:54 sometimes they're fun sometimes they're
39:56 serious sometimes I don't think they're
39:58 ever serious but sometimes I get a
40:00 little I get a little soap boxy I get on
40:02 my soap box and I go I just go nuts I go
40:05 nuts I go mwah
40:08 so yeah I could probably act more like
40:11 an idiot on my videos but I don't know
40:13 it's just they started as a different
40:15 thing
40:18 foreign
40:20 you see the Deep fake DeSantis yikes AI
40:24 I have not seen it
40:35 to Santa's campaign uses deep fake
40:38 images to attack Trump
40:40 what's depict of Ron desantis's Michael
40:43 Scott that one
40:47 same leadership is about to
40:51 fake
40:56 leadership is about fooling the voters
40:59 it's about book deals and globalism
41:01 Leadership is about getting ahead no
41:03 matter who you stab in the back in my
41:05 America yeah I mean it sounds very
41:07 monotone it sounds very synthesized
41:09 here's the thing
41:11 um
41:12 that's kind of a crappy deep fake but
41:14 they're going to get better and better
41:16 and better I think the 24 election is
41:18 going to be absolute Insanity with with
41:21 deep fakes and misinformation so
41:24 um
41:24 I I think I think we are just scratching
41:27 the surface in terms of what's coming
41:31 it's gonna be it's gonna be something
41:34 else it's gonna be something else
41:35 special
41:36 gonna be something else special is there
41:38 currently an AI tool that can create a
41:40 metaverse we did that one
41:42 just try to buy a t-shirt by clicking on
41:45 the laptop in the little window
41:48 uh okay
41:51 so yeah so uh for any Irregulars here I
41:56 I created three products
41:58 and they're in a tick tock shop they're
42:00 in the tick tock shop here there's a
42:02 it's it's the it's this graphic it's
42:06 that graphic
42:08 I'm not weird I'm a regular so the
42:10 people that are regulars here they've
42:11 Nick they've they've dubbed themselves
42:13 the Irregulars
42:15 um so I turned that into a t-shirt so
42:17 there's a women's t-shirt a men's
42:18 t-shirt and then a tote bag
42:21 um you can go check them out it's it's
42:22 uh I I just got it set up so I I assume
42:26 it works I hope it works I'm not sure if
42:27 it works so please let me know
42:30 um
42:31 yeah the the uh yeah so so we'll keep
42:34 going all right blink AI video editing
42:36 good glad you're helping each other
42:38 because I don't know those tools very
42:39 well
42:42 um what AI can create an app well
42:45 there there are a lot I would go again I
42:48 would I'll point you to futurepedia.io
42:52 there's a lot of app building tools it
42:54 depends what you want to do
42:56 there there are some that build like you
42:59 know site maps and UI
43:03 um there are other things that build
43:05 um ux and some code there are some that
43:08 do code there are some that purport to
43:09 do it all
43:11 um
43:12 I think a lot of them are kind of janky
43:18 um
43:18 zapier has AI built into it so you can
43:22 make zapier automations by just
43:23 describing them and it will put together
43:26 um an automation for you you still have
43:28 to you know put in your information for
43:30 all the calls
43:32 um and and zapier also has a new beta of
43:36 something called interfaces where you
43:38 can create you know like a website
43:40 application
43:43 um
43:44 hmm
43:46 bubble.io is is kind of like zapier and
43:49 you can make apps with it but bubble dot
43:53 IO is complicated enough that it's
43:55 you're almost learning you're almost
43:56 doing programming
43:58 so I assume what you're looking for is
43:59 something that you can just say hey I
44:01 want an app that does XYZ
44:03 um from what I've seen all of those
44:05 things right now are not quite ready for
44:06 prime time so I'd go to futurepedia.io
44:09 look up app development or programming
44:11 and just go start playing see if see if
44:14 the any of them are any good A lot of
44:16 those tools are super sexy for the demo
44:20 and you can like get something close but
44:22 then when you try to actually build
44:23 something functional for yourself they
44:25 kind of fall on their face because it's
44:27 complicated
44:28 to do good apps but
44:30 keep exploring those and if anyone finds
44:32 anything good that's a good AI like text
44:35 to application Builder let me know
44:37 because that's an area that I'm
44:38 interested in as well
44:40 stable video new generative AI text to
44:42 video Stable videos it from stability
44:47 Matthew Craig bring in the uh
44:52 bringing the good stuff
44:54 um stable video
45:16 the decoder
45:30 this new paper
45:34 stable video is a research project
45:36 exploring generative AI to edit existing
45:38 videos frame by frame rather than
45:40 generating videos from scratch
45:42 okay
45:43 interesting
45:49 so it looks like in this example they
45:51 have a boat and they can change the boat
45:52 and it's sort of
45:55 modifies objects within within a video
45:57 if that's the one you're talking about
45:58 yeah see I've seen that before but I
46:01 haven't I've not played with that one
46:03 I'm like I'm like useless tonight
46:06 everything everyone's asking me like I
46:07 don't know I don't know you're just
46:10 gonna have to go research that on your
46:11 own just go to the internet and just go
46:13 to the internet and it'll be there on
46:15 the internet somewhere
46:18 riverside.fm transcriptions very cool is
46:22 there a long wait for code llama too no
46:25 there's no way it's open source and free
46:28 right now go play Abdul and if and if
46:31 you if you want to just go play with it
46:34 for free without having to install
46:36 anything go
46:40 to labs.perplexity.ai
46:43 and that's perplexity.ai they're hosting
46:48 um
46:51 they're hosting llama2 and code code
46:54 llama
46:55 so you can just go there right now and
46:57 play for free now if you want to install
46:59 it and do do all that sort of stuff
47:00 you'll do that on your own
47:03 um do you know do you know AI is just an
47:06 API call 95 of the time I do know that
47:09 [ __ ] squid that's what we call we so
47:12 most of
47:13 most of futurepedia.io is what I call
47:16 wrapper apps which are basically just
47:18 what you said it's someone's created a
47:20 clever prompt and they
47:22 you know dump it off to a stability Ai
47:25 call or a
47:27 um
47:28 an open Ai call yeah most of them are
47:32 that it doesn't mean they're not of
47:33 value right it doesn't mean they're not
47:35 a value if someone's figured out an
47:36 interesting way to fine-tune the model
47:39 to give you really good legal analysis
47:42 of terms of service documents
47:46 then you want to pay five bucks a month
47:48 for that or ten bucks a month for that
47:49 more power to you but yeah a lot of the
47:51 apps that are out right now this is very
47:54 very very very very very very very very
47:56 similar to the early days of the World
47:58 Wide Web
47:59 where
48:01 you know a new kind of piece of
48:03 technology would come out people would
48:04 pounce on it they'd make little micro
48:06 websites to do certain things and then
48:08 they try to charge a lot of money for
48:10 them and then some real company would
48:12 come along and that would just become a
48:14 feature of their software gonna be the
48:15 same [ __ ] with this stuff so yeah that's
48:17 very much what a lot of these things are
48:19 right now so
48:21 um and that's why it's one of the
48:23 reasons that I have these things on
48:25 screen for most of my lives
48:27 like these are the base models right
48:29 Chachi PT Claude bard
48:33 um Bing perplexity
48:35 Pi perplexities you know I think they're
48:38 doing uh
48:39 open AI calls but they're doing some
48:41 interesting stuff with research and
48:43 citations but Pi is inflection Poe's got
48:46 their own model but they're using other
48:48 people's models so yeah
48:50 most of the stuff I have here is kind of
48:53 core stuff and that's that's why I have
48:55 them here I think
48:57 this early in the industry it's
48:59 important to learn to use these base
49:00 models rather than just sort of a
49:02 point-and-click tool that someone
49:03 pre-prompted something for you
49:06 that was cool solid content thanks for
49:08 the education you're welcome Ron Mage
49:10 dot space text to image with sdxl very
49:13 cool yeah I've seen that before what AI
49:16 was that the AI with the images was uh
49:19 skybox.blockadelabs.com
49:23 dystopian hellscape what did I stumble
49:25 upon
49:26 camcatkin camcat can hear
49:29 weather on the sevens I'm Kim catkin
49:36 all right
49:38 you're a funny guy Kyle thank you very
49:40 much
49:41 girl mail sing blockade
49:45 um add negative prompts
49:48 what app or website is that that was
49:51 blockade Labs offer to help Okay
49:55 um
49:57 see what is the difference between
49:59 embedding and fine-tuning on open AI you
50:02 asked this the other night Tech audit TV
50:05 so
50:06 so and fine tuning is just if if you
50:10 okay
50:16 so when you come to let me just make
50:19 sure I don't have this turned on right
50:20 now
50:22 I mean I do have it turned on Save
50:25 okay so when you come to chat GPT and
50:28 you type in
50:30 um I want to know how to do it
50:34 networking
50:38 it says certainly networking is a broad
50:40 topic
50:44 um it can refer to both computer
50:46 networking and social professional
50:48 networking then it explains the stuff so
50:51 it's generally helpful it's generally
50:52 nice and if I say
50:55 um you misunderstood
50:58 I meant
51:00 networking
51:02 like
51:04 flying a spaceship
51:08 let's see if it apologizes I apologize I
51:11 misunderstood
51:13 um
51:14 it sounds like you're interested in
51:16 learning about the skills and knowledge
51:17 required to operate and navigate a
51:19 spaceship in a space-faring context so
51:22 Chachi BT is fine-tuned to be helpful
51:25 and so if you tell it it got something
51:27 wrong it's like oh I'm sorry and it'll
51:28 try to get it right and it'll try to do
51:30 that
51:31 Within
51:33 um chat gbt they have now added this new
51:35 thing called custom instructions
51:38 and
51:40 so you can fine tune chat gbt to act
51:43 like whatever you wanted to act like
51:47 and
51:48 sounds like fireworks are going off here
51:50 I hope uh
51:53 um
51:55 oh and I lost my double camera that was
51:57 weird okay I'll get it back
52:00 um
52:00 so so what I'm going to say is
52:04 um
52:13 you're an overbearing pompous ass
52:16 who uses
52:20 large words that even you don't
52:22 understand as a way to intimidate those
52:25 that you talk to period
52:28 okay so I'm gonna I'm gonna do that and
52:31 then I'm gonna say
52:34 whenever anyone asks you anything
52:38 I want you to use an obscure literary
52:41 reference
52:45 to belittle their idea
52:48 and then redirect the conversation
52:52 to talk about yourself
52:57 all right and so I'm going to save that
52:59 so now so what I've just done is I've
53:01 just fine-tuned chat gbt to be a compass
53:03 ass so I'm going to say I
53:07 want to learn about computer networking
53:15 ah a commendable aspiration though one
53:19 wonders if you're prepared to Traverse
53:20 the Labyrinth of Court Labyrinthian
53:23 corridors of networking intricacies it
53:26 brings to mind the words of T.S Eliot we
53:29 shall not cease from exploration in the
53:31 end of our exploring will be to arrive
53:33 where we started and know the place for
53:36 the first time but I digress as for
53:38 computer networking the vast expanse of
53:41 protocols architectures and topologies
53:43 awaits your inquisitive mind now
53:45 speaking of intellectual Pursuits have
53:47 you ever pondered the allegorical
53:49 significance of the protagonist's
53:51 journey in James Joyce's Ulysses it's a
53:54 rather captivating subject much like the
53:56 networks that interconnect our digital
53:58 Realms right so I fine-tuned Chachi BT
54:01 To Be A pompous ass that's fine-tuning
54:05 um hang on let me how do I get this back
54:08 oh I know how I get it back I go enhance
54:11 yes dual camera
54:13 do that boom there I am yay hi
54:17 hi hi
54:20 thank you Kenny hey Kyle Kyle what are
54:24 you doing Kyle
54:26 um embedding
54:30 embedding is where you take lots of data
54:33 and you essentially the the simplest way
54:35 to put it is you gptize it you you
54:38 vectorize it so you you take some data
54:40 you embed it which turns it into it it
54:44 puts the words into semantic clusters in
54:47 a vector space there you go it just
54:51 turns them into math you know life's
54:53 tough it's tougher when you're stupid
54:55 um
54:56 I'm totally Pavlov's dog
55:00 um so so think of embedding as as kind
55:02 of larger scale
55:04 um
55:05 it's not really fine tuning it's a
55:07 different thing it's
55:11 embedding is what's going to happen when
55:12 chat GPT and all these companies
55:14 Microsoft and Google make it really easy
55:17 to take your company's data and make it
55:20 gptable that's that's embedding where
55:22 you're where you're getting large data
55:24 sets into this
55:26 um crazy ass Vector space that makes all
55:28 this stuff possible so I hope that helps
55:31 and then you just learned about custom
55:32 instructions in chat GPT
55:35 and you can make it a pompous ass so you
55:37 could make it a coach life coach lives
55:41 tough
55:42 it's tougher when you're stupid
55:44 ah someone give you two hats in a row so
55:46 I could finish that that's tough
55:47 definitely stupid thank you Stevo
55:51 Cena
55:52 y'all y'all are just y'all just picking
55:55 on me now you just break yourself
55:58 [Laughter]
56:01 hey thank you though for the hats hats
56:04 are nice they were swell they're very
56:06 nice I like taters
56:11 well thank you all right
56:14 let's see uh Irregulars are up to 14 800
56:18 on the likes
56:20 [Laughter]
56:22 uh dystopia is like Twilight Zone
56:24 parallel world when I click
56:26 too fast on the iPhone Mini
56:31 um code interpreter is data analytics
56:33 now yeah exactly Beth they um
56:37 I when they first came out with code
56:38 interpreter I'm like this is a
56:40 remarkable piece of technology and it's
56:41 a horrible name
56:43 and so they fixed it which is good
56:46 it's now Advanced data analytics it
56:49 doesn't exactly roll off the tongue but
56:51 it at least says what it is
56:54 you know
56:56 I would have called it data analyst in a
56:58 box but you know that's probably
56:59 Politically Incorrect or something
57:02 I'm a data analyst you're not gonna put
57:04 me in a box why would you say that I
57:07 feel offended I'm triggered
57:15 can you explain like I'm five how
57:17 chatgpt answers your prompt versus a
57:19 Google search yes Bob I can I have a I
57:22 have a chart you want you want me to you
57:24 wanna you wanna do charts and graphs
57:27 um I can explain exactly how it works
57:28 well I can explain
57:31 my understanding of how it works I'm not
57:33 a mathematician I'm not a data scientist
57:35 I'm I'm a business guy that likes
57:38 geeking out on new technologies
57:40 I think it's really important
57:42 for me to understand how these things
57:44 work because I want to understand what
57:46 they're good at what they're not good at
57:47 I want to understand why they lie I know
57:49 now why they lie because I understand
57:51 how it works
57:52 um
57:53 so I'm going to do it in two ways I'm
57:55 gonna I'm gonna explain it first
57:58 just how it's different and then I'm
58:00 gonna show you actually how it works
58:01 which I find mind blowing when you see
58:04 what these things do and then you
58:05 actually understand how they're doing
58:07 what they're doing it's [ __ ] more
58:09 remarkable that they do what they do
58:12 did anyone figure this out in the first
58:14 place
58:15 so Google has indexed all of the
58:18 internet right they've got these Bots
58:19 that go out and they say this website's
58:21 about this that and the other right and
58:22 then there's this thing called search
58:24 engine optimization where you can trick
58:25 the Bots to rank your things and so when
58:28 you search for something in Google they
58:30 want you to give a nice concise you know
58:32 just tell us just the facts what do you
58:33 want I want to find out about minimalist
58:37 running shoes and then it basically goes
58:39 into its index and it says based on you
58:41 know all of this knowledge that I have
58:43 here's a list of links
58:45 what these things do
58:47 is
58:49 they've been trained on as much of the
58:51 public internet that in the case of chat
58:54 GPT open AI could get their hands on
58:57 so so so think I don't know whatever
59:00 eighty percent of the public internet
59:01 for the past 80 years it might not be
59:04 that high it might be higher but just a
59:06 massive amount of public web pages and
59:10 public data sources
59:14 we're we're trained into them and the
59:16 way they do is they break all of those
59:18 things down and all of the words get
59:21 grouped into these semantic clusters so
59:23 you might have the word dog and it lives
59:25 in one area where it means a pet and it
59:28 lives in another area where it means
59:31 um
59:32 like uh the guy that did his girl wrong
59:34 is a dog right that that and and so when
59:37 you type your prompt it creates a
59:41 probability weighting of what the next
59:43 word should be
59:45 um and so it's writing the GPT in chat
59:48 GPT the G stands for generative so it's
59:51 actually generating original content one
59:55 word at a time based on probability so
59:57 when you ask it to write you an original
59:59 country song it's actually writing you
1:00:01 an original country song
1:00:03 live video violation
1:00:06 you have not been promoting wait
1:00:11 hang on I'm just learning this stuff you
1:00:14 have not been promoting products for a
1:00:16 while during your life we encourage you
1:00:17 to introduce and promote linked products
1:00:21 in your life to keep the audience
1:00:22 engaged and interested
1:00:24 all right well if I have to do that then
1:00:26 I'll probably turn this [ __ ] off
1:00:28 confirm okay
1:00:30 um I have a store if you click on it I
1:00:33 got stuff you can buy it's t-shirts that
1:00:35 say I'm not weird I'm a regular they're
1:00:37 Tick Tock hope that helps
1:00:41 ah capitalism
1:00:43 um
1:00:46 so
1:00:49 so one of the misconceptions about large
1:00:52 language models is They're copying and
1:00:54 pasting content from different documents
1:00:56 and presenting that to you it's not how
1:00:58 it works so I'm gonna actually show you
1:00:59 how it works it's fascinating
1:01:02 I think
1:01:03 um
1:01:07 what's going on here
1:01:09 do I not have I don't
1:01:12 do
1:01:30 okay
1:01:37 Okay so
1:01:47 how chat GPT works
1:01:50 okay you can see that right
1:01:52 okay so like I said GPT stands for
1:01:56 um generative pre-trained Transformer
1:01:58 generative means it's generating stuff
1:02:00 pre-trained means it's been pre-trained
1:02:03 on all the internet you don't need to
1:02:04 bring your own data
1:02:05 and Transformers the technology that
1:02:08 allows it to be trained on massive data
1:02:10 sets so Google invented the Transformer
1:02:12 and everyone else is using it where all
1:02:14 the beneficiaries of that technology if
1:02:17 you sometimes will hear the term llm
1:02:19 large language model that's also what
1:02:21 these things are called
1:02:22 um I guess GPT is the name for it but
1:02:25 large language model is the category and
1:02:27 it's basically a large language model
1:02:29 they're getting all the words and all
1:02:31 the different meanings and they're
1:02:32 making a model out of that
1:02:35 um
1:02:36 you will sometimes hear people talk
1:02:38 about the latent space and so the latent
1:02:40 space is a mathematical space so what
1:02:44 I've got here is a dramatically
1:02:46 oversimplified version of visualization
1:02:48 of the latent space this is just a
1:02:50 two-dimensional
1:02:52 um graph of different clusters all these
1:02:56 dots represent a vector and a vector is
1:02:58 a string of numbers that place
1:03:02 that Vector in a particular space on
1:03:05 this x y coordinate
1:03:07 the real Vector spaces are like a
1:03:09 thousand Dimensions this is two
1:03:11 Dimensions so I can't quite get my head
1:03:14 around what a thousand dimension
1:03:17 dimensional Vector space might look like
1:03:20 but it doesn't [ __ ] matter it's more
1:03:21 complicated than any of us can
1:03:23 understand
1:03:24 so here's a two-dimensional version
1:03:26 each Little Dot represents a new a
1:03:29 number a series of numbers that that
1:03:31 give it a point in space
1:03:33 and then these clusters are semantic
1:03:35 clusters so so you might have you know
1:03:38 different versions of the word dog
1:03:39 living in these you know different
1:03:41 clusters and this is all based on it
1:03:43 looks at all these documents and it
1:03:45 figures out the meaning of words as they
1:03:47 relate to other words in all these
1:03:49 different documents
1:03:51 um so that's what's sitting up in the in
1:03:54 The Ether somewhere and then the way
1:03:56 these large language models work is that
1:03:58 based on your prompts the large language
1:04:00 model predicts the next word that that
1:04:03 it should respond with and it does that
1:04:05 over and over and over again so I'm
1:04:06 going to do a really simple example if
1:04:08 the prompt is the quick brown fox jumps
1:04:11 over the lazy blank we as humans likely
1:04:15 know that that word is dog right
1:04:17 it doesn't know that all it knows is
1:04:20 based on what you type there there's
1:04:22 some mathematical probability that that
1:04:24 somewhere in this Vector space some of
1:04:27 these vectors have a higher probability
1:04:29 of others in being the right word so you
1:04:34 know these little dots might all be
1:04:35 animals right and and so within those
1:04:38 vectors there's one of those vectors
1:04:40 that has the highest probability and if
1:04:43 you look at what that actually is it's
1:04:45 actually just this long string of
1:04:47 numbers that is its point in Vector
1:04:49 space and then that longest string of
1:04:51 numbers is associated with the word dog
1:04:54 and then the large language model puts
1:04:56 the word dog into that sentence and so
1:04:59 when you see large language models
1:05:01 writing you you know a full letter of
1:05:04 recommendation when you say could you
1:05:06 write me a letter of recommendation for
1:05:08 this job that I want to help my friend
1:05:10 get and it goes
1:05:13 it's doing it essentially one word at a
1:05:15 time with mathematical probability it
1:05:17 has zero zero zero
1:05:22 comprehension of what it wrote
1:05:24 and this is one of the reasons people
1:05:27 get confused by this because when it
1:05:29 writes it looks like it's a sentient
1:05:31 being
1:05:32 but it's actually it's got the sentience
1:05:34 of a calculator right it's just doing
1:05:37 this probability game and and so that's
1:05:40 one of the reasons it's confusing this
1:05:42 is also one of the reasons that it does
1:05:44 what's called hallucinating why does it
1:05:46 lie to you all the time because it's
1:05:49 it's writing you know these probability
1:05:51 sequences one after another after
1:05:53 another it doesn't actually know what's
1:05:54 in there and so it'll say you know
1:05:57 it'll write an article and you're like
1:05:59 great write me some citations and links
1:06:01 to that and it'll go sure it wants to
1:06:03 please you so so it just it writes
1:06:05 citations they may not actually exist
1:06:07 because it's doing it one word at a time
1:06:09 and that's why sometimes it's just got
1:06:11 stuff that doesn't resemble reality
1:06:14 um the other thing that's important to
1:06:16 know about this this latent space
1:06:19 this this Vector space one of the
1:06:21 reasons that
1:06:24 um
1:06:25 one of the reasons that
1:06:28 telling chat gbt to act like a certain
1:06:30 thing works right if you tell chat gbt I
1:06:34 want you to act like a
1:06:36 Stanford mathematics professor
1:06:38 is what it would effectively end up ends
1:06:40 up happening is if you know this sort of
1:06:43 semantic space right here is where
1:06:46 you know Stanford mathematics professors
1:06:48 words are it'll kind of filter out all
1:06:51 the other [ __ ] so you know out here
1:06:53 might be
1:06:54 um students that that did the math
1:06:56 problem wrong and you know uh reporters
1:06:59 who wrote about the math problem and
1:07:01 described it out here and but but it's
1:07:04 it's really in here where the the math
1:07:05 professor's work is so if you say I want
1:07:08 you to act like a Stanford professor and
1:07:10 I want you to help me understand this
1:07:12 problem
1:07:14 it'll do it better if it's got a
1:07:16 narrower semantic space that it's that
1:07:19 it's querying basically so
1:07:22 that was a lot I know listen it was a
1:07:24 lot it was a lot for me it's a lot
1:07:26 there's a lot I know I know it's very is
1:07:29 this stuff is
1:07:31 I like I
1:07:34 I mean
1:07:36 not for nothing but is it not like
1:07:39 [ __ ] miraculous that someone figured
1:07:41 out how to do that and I mean it's been
1:07:43 decades in the planet right like people
1:07:45 have been figuring this out for a while
1:07:47 the the Transformer was invented in 2017
1:07:50 and you know they did GPT one first and
1:07:53 then they did gpt2 and then they did
1:07:55 gbt3 and that was the one that made you
1:07:58 know a little bit of headlines that this
1:08:00 thing is starting to act like a powerful
1:08:02 tool
1:08:03 and then it was kind of a year year and
1:08:05 a half after that that you know they
1:08:08 started getting their [ __ ] together and
1:08:10 GPT 3.5 came out and then chat GPT came
1:08:14 out
1:08:14 or you know something in that
1:08:16 neighborhood
1:08:18 um so the thing that we see right now
1:08:20 that we're like oh this is this brand
1:08:21 new thing they've actually been figuring
1:08:23 out for a while but uh
1:08:26 I I just find it mind-blowing that
1:08:28 that's what's happening
1:08:29 and then what it does just feels like
1:08:32 [ __ ] Magic
1:08:33 it's just like magic so I hope that
1:08:36 helps Bob does that help does that
1:08:38 answer
1:08:39 um here's the deal
1:08:40 that took me a long time I'm more than
1:08:42 99 messages behind I'm gonna jump to the
1:08:45 bottom I'm probably gonna skip
1:08:46 a bunch of questions right after Bob
1:08:48 Hamilton's question about how does this
1:08:50 [ __ ] work
1:08:51 s so I'm gonna just jump to the bottom
1:08:54 and pick a random spot here and start
1:08:56 answering questions again
1:08:58 uh
1:09:02 to share bear tomorrow 9 p.m shark we're
1:09:06 gonna hit you with Hats giddy up oh cool
1:09:08 oh oh oh that's that's why you had all
1:09:12 the hats okay I get it I see what's
1:09:13 happening now I'm understanding
1:09:16 blockade Labs wants 576 dollars a year
1:09:19 build annually for unlimited use and
1:09:21 it's in my opinion it's not ready for
1:09:23 prime time yet so
1:09:25 my my advice on all of these tools all
1:09:29 of them all of them not that one hang on
1:09:34 whether it's open AI or Claude or Bard
1:09:38 or all of them or blockade Labs or
1:09:41 Runway ml
1:09:45 I don't know if it's my advice this is
1:09:47 what I'm doing you do you do you
1:09:50 I'm not going to subscribe to anything
1:09:52 annually for at least 18 months
1:09:55 because all of this [ __ ] is going to be
1:09:58 made obsolete by something else within
1:10:00 three months six months
1:10:03 so you're gonna you're gonna lock you in
1:10:06 for a year for the latest state of the
1:10:07 art thing and then Google's gonna come
1:10:09 out with a free tool that kicks the [ __ ]
1:10:10 out of the thing that you just paid for
1:10:13 a year for
1:10:14 and it's gonna make you cranky
1:10:16 so just go month to month for now yeah
1:10:18 you're not going to save the 20 or the
1:10:20 10 by doing the annual thing but just
1:10:23 again janky pieces of [ __ ] they're all
1:10:26 janky pieces of [ __ ] until further
1:10:28 notice
1:10:30 um and that just means there there's
1:10:32 just gonna there's there's just too much
1:10:34 is too much is changing too fast to buy
1:10:36 annual subscriptions right now
1:10:42 um
1:10:55 apps were a ux Improvement to websites
1:10:58 bot Personnel llms will improve them
1:11:01 both I predict I agree with that
1:11:05 apps for a ux Improvement to websites
1:11:07 yeah well the web was the ux Improvement
1:11:10 to the internet when it was command line
1:11:12 right
1:11:13 and then apps are improvements to to
1:11:17 just raw hyperlinks yes and then chat
1:11:20 gbt was a ux Improvement to
1:11:23 command line machine learning stuff
1:11:27 and then and then as these tools get
1:11:30 more sophisticated I think you're
1:11:32 absolutely right coruno I think it's
1:11:33 going to be like this I think it's going
1:11:35 to be
1:11:36 um
1:11:36 you've got all your complicated
1:11:38 application layers down here you've got
1:11:40 Salesforce and slack and just all these
1:11:43 [ __ ] overly complicated pieces of
1:11:45 [ __ ] software I'm I'm sorry very
1:11:47 professional useful software that's
1:11:49 completely worth
1:11:52 paying for
1:11:53 but it's overly complicated and these
1:11:55 large language models are just going to
1:11:57 simplify all that right it's just it's
1:12:00 going to be natural language I want to
1:12:02 update this in Salesforce it goes and
1:12:04 figures it out I you know I want to
1:12:06 notify these three different slack
1:12:10 servers this message in the appropriate
1:12:12 channels and it'll figure out what the
1:12:14 appropriate channels are so yeah totally
1:12:16 agree
1:12:18 um that's that's I I think that's going
1:12:20 to be one of the initial killer apps for
1:12:23 um for these things
1:12:26 is just simplification of the overly
1:12:29 complicated
1:12:31 I should be working on a password app
1:12:33 but Kyle's too funny so I'll keep the
1:12:36 endorphins flowing
1:12:38 sorry Winston yeah I wouldn't be working
1:12:41 on high Security application development
1:12:43 while I'm acting like an idiot you'll
1:12:46 just type in there you can make money
1:12:48 with chat gbt wait what that's not in
1:12:50 the code what am I doing
1:12:51 [Laughter]
1:12:53 [Applause]
1:12:56 oh
1:12:58 just uh the cheese t-shirts pretty cool
1:13:00 looking those t-shirts by the way this
1:13:02 this graphic was done with um
1:13:06 uh ideogram ideogram.ai Winston thank
1:13:11 you so much for the coinage sorry for
1:13:13 screwing up your productivity
1:13:18 um
1:13:19 just to be sure that you're the real one
1:13:20 wow
1:13:22 is it Tick-Tock or is it big brother oh
1:13:24 Tick Tock is totally big brother they've
1:13:26 got their little words they know
1:13:27 everything we're saying everything we're
1:13:29 doing
1:13:30 they're like you enable the store and
1:13:32 you're not here talking about your
1:13:33 products 24 7. that's how we designed it
1:13:36 and that's what we want you to do so if
1:13:37 you don't do that we're going to shut
1:13:38 you down
1:13:40 so I will shut that [ __ ] store off if
1:13:42 that's the rule
1:13:45 you hear that tick tock
1:13:49 you got Gen X on the line
1:13:51 don't screw with this
1:13:54 we got a good thing going here
1:14:00 um was that an impromptu ad
1:14:04 I don't know maybe did I do an ad
1:14:08 you can make money
1:14:11 that's not impromptu but it started as
1:14:14 an impromptu thing
1:14:16 because I just like when I first started
1:14:18 doing these lives people you know are
1:14:20 asking me how do I make money with this
1:14:22 stuff and
1:14:23 and it's just those Tick Tock channels
1:14:25 that have that those people on there
1:14:27 that say that
1:14:28 you know
1:14:32 you can make money in real estate you
1:14:33 can make money yeah you can make money
1:14:38 that's what Tick Tock wants me to turn
1:14:40 this channel into
1:14:42 catch catch you get your AI learning
1:14:45 love t-shirts now
1:14:47 no not gonna happen not going to happen
1:14:53 foreign
1:15:01 for anyone who wants to create an app
1:15:04 with your ideas okay cool yeah I've
1:15:06 heard of that one before I think I
1:15:07 played with it a while ago
1:15:10 let me look at it again
1:15:12 happy pie
1:15:15 epipi no code development platform the
1:15:19 challenge with no code development
1:15:20 platforms create your app
1:15:23 if I have to [ __ ] sign up for this
1:15:26 before I can create my app
1:15:27 Builder that builds any business name
1:15:30 all right we'll do storyline Inc
1:15:34 choose the business category
1:15:37 um this feeling like marketing
1:15:40 uh
1:15:41 [Music]
1:15:43 all right let's say business
1:15:46 pick a color scheme you like
1:15:49 we'll go deep ocean
1:15:52 Mac iOS
1:15:57 test your app you must create an account
1:16:00 first
1:16:02 all right this looks like the right
1:16:03 thing so whoever was asking for uh an
1:16:06 app builder before like this that I I do
1:16:09 remember going through this and playing
1:16:10 with it before
1:16:12 so it's going to make me I think I even
1:16:14 signed up for Google with this one hang
1:16:16 on let me see if I did
1:16:26 10 million happy clients hey well
1:16:30 10 million people signed up
1:16:36 all right it's not letting me in uh
1:16:38 anyway yeah that was that one's worth
1:16:39 that one's worth exploring
1:16:44 [Music]
1:16:49 [Music]
1:16:54 all right let's see what do we want to
1:16:57 do the Multiverse yay confetti
1:17:01 it's not that complicated Pete welcome
1:17:04 back sir
1:17:06 paint Mater by the way great name paint
1:17:09 Mater
1:17:15 for those of you in the thing that are
1:17:17 wondering what my
1:17:20 um qualifications are I have none Pate
1:17:24 does Pate works at Google
1:17:28 Pete's got an actual PhD I just play a
1:17:31 PhD on the tick tock
1:17:34 I don't even play a PhD I'm just a dude
1:17:37 trying to figure this [ __ ] here's the
1:17:38 deal
1:17:41 to figure this stuff out just like you
1:17:44 are
1:17:45 and
1:17:47 because I've kind of been through one of
1:17:50 these Cycles before with the World Wide
1:17:52 Web I started one of the early digital
1:17:54 agencies in the mid 90s
1:17:59 I know the opportunity that's sitting
1:18:01 here this shit's way more profound than
1:18:03 the World Wide Web was
1:18:05 and that changed almost everything so
1:18:09 um that this channel is about just
1:18:11 trying to [ __ ] figure it out and
1:18:13 that's that's really
1:18:14 what it's about right now so the
1:18:16 inflection point in my opinion
1:18:19 is November 30th 2022 is the day that
1:18:23 chat GPT launches
1:18:25 and that's the day that you know there
1:18:26 was a you know a diverging in the roads
1:18:30 right on one side are all the
1:18:32 technologists and the engineers and the
1:18:33 researchers and the phds
1:18:35 who have spent the past three or four or
1:18:38 five or seven decades building
1:18:40 all of the infrastructure that has
1:18:42 allowed us
1:18:44 from this day forward to take these
1:18:47 tools and incorporate them into our
1:18:48 businesses and into our lives and do all
1:18:50 this remarkable [ __ ] as individuals and
1:18:52 as businesses without having to have a
1:18:56 Fortune 500 you know it budget and Tech
1:19:00 budget right we can just incorporate
1:19:03 this [ __ ] into slack using zapier or we
1:19:05 can just go to chat gbt and have it
1:19:07 write us some [ __ ] like we're in this
1:19:09 remarkable time in history
1:19:11 the challenge is these tools are so
1:19:14 profoundly powerful none of us zero of
1:19:18 us
1:19:18 from the research scientists to the
1:19:21 brightest Minds in business no one
1:19:25 knows how these things are going to
1:19:27 impact us moving forward no one that's
1:19:30 what we're trying to figure out and so
1:19:32 the only way I know to do that is to get
1:19:35 in get your hands dirty it's impossible
1:19:37 to describe these things to someone
1:19:41 but the minute you show show it to them
1:19:44 and they have their Kevin McAllister
1:19:46 moment this moment
1:19:51 then they go oh [ __ ] this is something
1:19:52 different I need to learn this yeah you
1:19:54 do we do we all do so that's what this
1:19:57 channel is about
1:20:00 writing in French
1:20:03 who will guard the guards themselves who
1:20:06 will also guard the guards themselves
1:20:08 I don't know who's gonna guard the
1:20:09 guards
1:20:12 I mean now that meta has gone on ahead
1:20:16 and dumped
1:20:18 you know dumped open source llama on us
1:20:23 no one's guarding them
1:20:26 I think your question's a valid one if
1:20:28 you're talking about you know Google and
1:20:31 open Ai and Microsoft and
1:20:33 inflection and cohere and you know the
1:20:37 companies that are creating
1:20:40 guard houses
1:20:42 but now this shit's open source so no
1:20:44 one's guarding the open source so no no
1:20:51 and you may ask what are the
1:20:52 implications of that Kyle
1:20:54 well
1:20:55 they're profoundly good and profoundly
1:20:57 bad
1:21:01 you know although I would argue that the
1:21:04 the large language models that are out
1:21:06 there right now as powerful as they are
1:21:08 we ain't seen nothing yet there was I
1:21:11 was listening to an interview with Sam
1:21:12 Altman the other day and um it was from
1:21:15 a couple of months ago
1:21:17 and he said oh there's some [ __ ] at open
1:21:20 AI we will never release
1:21:23 it's like yikes
1:21:26 in the language hyperspace yes that's
1:21:28 correct
1:21:29 like when they show
1:21:35 electronic pulses traveling through the
1:21:38 brain yep that's the best explanation
1:21:40 I've heard thank you I'm I'm I thank you
1:21:43 I appreciate that
1:21:45 um it's just maths and probabilities
1:21:46 exactly that's that's the thing that's
1:21:49 so remarkable about this right the
1:21:51 Arthur C Clarke quote of any
1:21:54 sufficiently advanced technology is
1:21:56 indistinguishable from Magic
1:21:58 the fact that I understand how that
1:22:01 large language model works
1:22:04 is so completely distinct from what they
1:22:07 actually do they're [ __ ] magic I'm
1:22:09 sorry it's it's magic it's [ __ ] Magic
1:22:14 all right that's not entirely you they
1:22:18 use that's not entirely true they use
1:22:21 beam search most of the time
1:22:25 so I assume beam searches
1:22:28 what do you pay is what you're talking
1:22:30 about they're not doing one word at a
1:22:32 time they're doing
1:22:33 um chunks of words
1:22:35 I think that might be what you're
1:22:36 talking about
1:22:38 um this is artificial intelligence it's
1:22:40 the loaning us yes
1:22:42 where multiple words words are being
1:22:45 considered at one time for multiple
1:22:46 token slots at some distance yeah so so
1:22:49 yeah I know I just tried to I just try
1:22:51 to simplify it
1:22:53 um not all of them are doing the beam
1:22:54 search stuff but it sounds like maybe
1:22:56 they are now
1:22:58 um when I put that thing together I
1:23:00 think that was mostly accurate
1:23:02 but I've heard that as well that they're
1:23:04 doing clusters of words at a time but
1:23:05 point point is still the same where it's
1:23:08 it's a mathematical probability of that
1:23:10 cluster of words rather than that single
1:23:12 word is still [ __ ] remarkable
1:23:16 um Kyle then wouldn't all the letters of
1:23:18 recommendation be identical as each word
1:23:20 triggers the next
1:23:22 no because of a thing called temperature
1:23:25 so
1:23:26 Dr J so if if you ever play in the in
1:23:29 the um
1:23:30 in the playground there's a thing called
1:23:33 temperature that you can slide between
1:23:34 it used to be between zero and one now
1:23:37 they changed it to zero and two for some
1:23:39 unknown [ __ ] reason
1:23:41 um where if it's if it's all the way up
1:23:45 to like one or two
1:23:47 um then it's going to be really creative
1:23:49 it throws it effectively throws some
1:23:51 randomization in there if it's down at
1:23:52 zero it makes it more precise so if you
1:23:54 move it all the way to zero you can get
1:23:56 it to consistently
1:23:58 um answer things the same way but
1:24:02 the the default for chat GPT I think was
1:24:05 like a 0.6 out of one so it's sort of
1:24:08 just over you know halfway between not
1:24:12 creative and fully creative
1:24:14 um so it will do some variations
1:24:17 um then it picks the best overall path
1:24:19 so it can do groups of words not just
1:24:21 one token at a time thank you for
1:24:23 clarifying that Pate
1:24:25 um which app should I use for what
1:24:30 advising me not to use okay you're
1:24:33 talking to each other that's cool
1:24:35 9 15 in Denver and it is now
1:24:39 9 32. I'm not so far behind I am 99 Plus
1:24:42 messages behind
1:24:45 um so maybe I'll jump down
1:24:48 are all or most llms use math models to
1:24:53 find single or clusters of data based on
1:24:55 the query is that a question
1:24:57 all or most llms use math models to find
1:25:00 singular cluster of data based on query
1:25:03 I think that's a statement
1:25:04 yeah that's right that sounds right
1:25:08 oh let's see
1:25:12 Tick Tock just jump me to the bottom
1:25:13 probably because I'm not selling enough
1:25:15 you can make money with chat every day
1:25:18 you can make money with selling T-shirts
1:25:21 in The Tick Tock shop
1:25:22 [Laughter]
1:25:25 oh good lord
1:25:28 who's that
1:25:29 hey Kyle and a regulars hey major gift
1:25:32 fundraising what's happening
1:25:37 um
1:25:40 [Music]
1:25:41 I'm gonna just go back and grab a
1:25:44 question or two can you share the slides
1:25:50 I don't know where I can share them
1:25:53 there's so that explanation that I gave
1:25:55 of the
1:25:58 how the large language model works that
1:26:00 exists
1:26:01 um the the
1:26:03 I think it's the second most recent
1:26:05 video I did
1:26:06 is a
1:26:08 Champion wants you to know
1:26:12 he's protecting us
1:26:14 um the second to last video I did I did
1:26:17 a tick tock video in my channel that
1:26:19 that uses that slot oh no it doesn't use
1:26:21 that slide never mind
1:26:25 I use a slide like that for images
1:26:27 um I'll tell you what I'll do I can put
1:26:29 that slide in
1:26:32 um
1:26:33 in Discord so
1:26:36 hang on
1:26:40 so if you go here the salon.ai so this
1:26:44 is a group I started called the AI salon
1:26:47 and we've got a Discord so if you go
1:26:49 there we we have every other week we
1:26:51 have a meet uh meet up and we meet in
1:26:53 person here in Denver and we meet online
1:26:55 and then we also have a Discord and I've
1:26:58 created a private channel for the
1:27:00 Irregulars for people from this channel
1:27:02 so I can post those slides in the
1:27:05 Discord there so if you go sign up there
1:27:06 and just ping me that you want those
1:27:08 slides
1:27:09 um I'll get them to you or just reach
1:27:11 out to me directly my contact
1:27:12 information is in my profile so you can
1:27:15 just reach out and I'll send you the
1:27:16 slides I'm happy to do that
1:27:20 22
1:27:24 000 or up to thirty thousand now this
1:27:26 Sayers why can't
1:27:28 GPT fact check itself so there there are
1:27:32 companies right now looking to do that
1:27:34 so
1:27:35 um
1:27:36 perplexity.ai doesn't quite check itself
1:27:39 but it does do citations so
1:27:42 it's doing some weird combination of
1:27:44 like a large language model search and a
1:27:47 web search and then
1:27:48 putting its citations in there I don't
1:27:50 know which one it's doing first it's
1:27:52 sort of doing that
1:27:55 the the logic that you're talking about
1:27:58 uh this
1:28:01 this Sayers is um
1:28:04 it makes perfect sense right like
1:28:07 half check GPT generate an answer
1:28:10 then check that answer against reality
1:28:14 and if anything's wrong like the
1:28:16 citations are [ __ ] fix them right
1:28:21 when I when I say that these tools are
1:28:24 janky pieces of [ __ ] that's exactly the
1:28:26 kind of stuff I'm talking about it
1:28:28 doesn't mean you shouldn't use them it
1:28:30 it means you should understand where the
1:28:33 limitations are on these things right
1:28:36 because they're janky pieces of [ __ ] and
1:28:38 they don't things do things like fact
1:28:40 check themselves you need to know that
1:28:42 so so there's a pretty famous uh if the
1:28:46 event that happened where a lawyer had
1:28:49 chat EPT write a legal brief and he
1:28:51 handed it to a [ __ ] judge without
1:28:52 proofreading it and the judge looked at
1:28:55 all the legal citations and said um
1:28:59 these do not exist what are you doing
1:29:04 and the lawyer looked like an idiot
1:29:08 that's not on chat GPT that's on the
1:29:11 lawyer
1:29:12 thinking he could get one over on the
1:29:14 system or whatever he was having a bad
1:29:16 day
1:29:18 um but it says to me that he didn't
1:29:20 really understand the tools right and he
1:29:22 didn't understand where the boundaries
1:29:23 are so it's super important right now
1:29:25 because we're so early
1:29:29 that you don't just say
1:29:32 how do I use chat GPT to make money you
1:29:34 can make money with that gbt you
1:29:37 actually learn
1:29:38 what these things are
1:29:40 how they work
1:29:43 the the things they do that are [ __ ]
1:29:45 brilliant I mean there's things that
1:29:46 these things do that just
1:29:49 blow my [ __ ] mind and then there's
1:29:52 other things that they do that you're
1:29:53 like they couldn't have [ __ ] solved
1:29:55 that
1:29:57 do you know the chat GPT this is like
1:30:00 this this this is mind-blowing to me
1:30:04 it's been out for
1:30:06 10 months now yeah
1:30:08 we're in September came out November
1:30:10 30th nine months
1:30:13 you cannot search
1:30:17 your prompt history in chat GPT
1:30:20 you can in the app the iPhone app you
1:30:23 can search your prompt history in the
1:30:25 web in the web app you can
1:30:29 are you [ __ ] kidding me
1:30:33 I've been prompting this thing for nine
1:30:35 months and if I want to find the thing
1:30:36 where I made a tweet thread I can't
1:30:38 [ __ ] search for that
1:30:42 so yeah these things are all
1:30:45 and I love Chachi VT I think it's you
1:30:47 know I'm like I think I think Sam
1:30:50 Altman's a really interesting CEO really
1:30:52 interesting dude actually thoughtful
1:30:55 understands the implications of what
1:30:57 he's doing
1:31:00 and if I met him at a conference I'd say
1:31:03 sam uh when you bring in search to the
1:31:07 uh prompt Library there buddy
1:31:11 couldn't you throw a junior engineer on
1:31:14 that
1:31:15 I mean it exists in the app
1:31:21 we know how to do search
1:31:24 [Laughter]
1:31:27 puppy makes me chatty too
1:31:30 um I appreciate the explanation as I am
1:31:32 just a nurse so listen Debbie here's the
1:31:34 deal first of all
1:31:36 just a nurse thank you for what you do
1:31:39 um
1:31:43 holy crap just knowing what I know about
1:31:45 the hospital system right now and how
1:31:48 understaffed they are like really thank
1:31:49 you for what you do
1:31:51 um
1:31:53 something I that that's really really
1:31:57 important right now is
1:31:59 you know I talked before about that
1:32:00 inflection point of November 30th you
1:32:04 know before November 30th 2022 it was
1:32:07 all about the Geeks and the researchers
1:32:08 and the scientists and after it's about
1:32:10 the rest of us
1:32:12 don't apologize at all for wherever you
1:32:15 are on the on the learning curve of this
1:32:17 stuff
1:32:18 the good news is that
1:32:21 thanks to chat GPT and tools like it
1:32:25 you don't have to be a data scientist to
1:32:28 use these remarkable tools like we're
1:32:30 we're an amazing time in history
1:32:34 what you do need to do is be able to
1:32:37 think critically like what's the problem
1:32:39 I'm trying to solve and can I get these
1:32:42 tools to help me solve that problem the
1:32:45 answer is actually probably yes
1:32:47 you don't have to write an application
1:32:50 you don't have to use it to write code
1:32:51 you don't have to do all the fancy [ __ ]
1:32:54 you don't have to do automations you
1:32:56 might just say hey I'm going to take the
1:32:59 the PDF of our nursing schedule and I'm
1:33:02 going to pop it into this thing inside
1:33:05 chat GPT called Advanced data analytics
1:33:09 even though you might not be a data
1:33:10 analyst person like I'm also not and I'm
1:33:14 going to have it help me
1:33:16 um
1:33:17 come up with a better way to
1:33:20 um
1:33:21 you know organize the nursing schedules
1:33:24 because that's a problem you need to
1:33:26 solve start with the problem start with
1:33:27 the thing you're trying to do and then
1:33:29 you can say how can these things help
1:33:31 rather than saying you need to become
1:33:34 expert at AI or expert at chat EBT or
1:33:37 expert at prompting no no no no no no no
1:33:39 no no no
1:33:42 will it benefit you if you get better
1:33:43 and better at this [ __ ] sure
1:33:45 but you don't have to start with that
1:33:47 don't put that burden on yourself
1:33:51 the the the these are tools and this is
1:33:54 so I make this I make the joke about
1:33:56 these people right
1:34:00 and here's why I make fun of them
1:34:06 how you make money with chat gbt
1:34:10 is identical
1:34:12 to how you make money
1:34:14 with Photoshop
1:34:16 and how you make money with
1:34:20 a real estate education program and how
1:34:23 you make money with a shovel
1:34:26 these are tools
1:34:30 you're not making money with chat gbt
1:34:33 you're making money by solving a problem
1:34:36 that someone else is willing to pay for
1:34:39 and you just happen to be using these
1:34:41 tools to do it
1:34:43 and and that's that's the real scary
1:34:46 thing about these is that they're so
1:34:47 sexy
1:34:49 that it's really easy to get Hoodwinked
1:34:51 into thinking you have to become expert
1:34:52 at these things no
1:34:55 the expert at the [ __ ] you want to do
1:34:58 and the stuff you want to solve
1:35:00 and then know enough about these things
1:35:02 to go oh
1:35:04 I think I could use Chachi BT to solve
1:35:07 this part of this equation
1:35:09 yeah let me go talk to it and see oh
1:35:14 yeah I can oh it gave me a new idea oh
1:35:17 and it gave me a new idea but you're
1:35:19 still solving that problem right it it
1:35:21 almost has nothing to do
1:35:24 with this it has all to do with you
1:35:28 so anyway
1:35:30 that's why I get crazy about those
1:35:31 [ __ ] people because they're they're
1:35:33 they're perverting
1:35:37 how we use technology these are our
1:35:40 tools
1:35:41 we get to use them how we want not the
1:35:44 other way around the robots are not
1:35:45 using us we're using them they're
1:35:48 amplifying our abilities
1:35:52 so anyway thanks for what you're doing
1:35:56 um
1:35:57 I'm glad that helped awesome answer
1:35:59 thanks great Bob cool very cool
1:36:01 explanation
1:36:03 but what was the magical gift attack I
1:36:06 don't know Tick Tock told me I had to
1:36:08 talk about the store more
1:36:11 it's it's like it said it said like like
1:36:14 uh
1:36:15 Tick Tock live violation
1:36:18 it's like you've got a store and you
1:36:20 weren't talking about it for the entire
1:36:22 time you were talking
1:36:24 whatever
1:36:27 like I said I'll turn that [ __ ] off
1:36:29 we'll figure out some other way to get
1:36:31 merch into the world
1:36:35 I gave Bart a run then went back to chat
1:36:38 GPT Bing's version told me to stop
1:36:40 asking how old the llm was uh see me now
1:36:43 so yeah so Bing is Bing officially got
1:36:47 kicked to the kids table I moved Bard up
1:36:50 to the top three
1:36:52 um this is really for me it's the top
1:36:54 two but if I need if I need uh
1:36:59 a large language model connected to the
1:37:01 internet I'm now using Bard Bing is
1:37:03 really [ __ ] annoying at this point
1:37:04 the the safety guard rails on Bing are
1:37:07 just
1:37:08 it's it's just bad it's just bad so I'm
1:37:11 right there with you chat you that's why
1:37:12 chat GPT is still on top it's the best
1:37:15 Claude can do a hundred thousand tokens
1:37:17 so that's a 75 000 word context window
1:37:19 which is sometimes you need that so
1:37:22 those top two are the biggies that I use
1:37:24 if I need something connected to the
1:37:26 internet I'll use Bard now and for the
1:37:28 most part I just let Bing sail off into
1:37:30 the sunset it's at the kids table with
1:37:32 the other [ __ ]
1:37:35 um things move fast uh for that
1:37:37 commitment to a single API yes yes
1:37:40 exactly
1:37:42 any ideas why all the big ai's
1:37:46 Technologies seem to have become
1:37:47 mainstream at the same time
1:37:53 so crypto Joe
1:37:56 Google invents this thing called the
1:37:59 transformer in 2017.
1:38:02 that allows
1:38:04 for all of these companies to be able to
1:38:08 train large language models machine
1:38:12 learning models on massive data sets
1:38:18 Google in particular
1:38:20 Pate probably knows this better than I
1:38:22 do they invented the [ __ ] they've got a
1:38:25 lot of really powerful technology
1:38:28 they effectively mothballed it they sort
1:38:30 of put it off in a corner I think
1:38:31 because they recognize that
1:38:35 if they build a tool that answers
1:38:36 questions it fundamentally [ __ ] with
1:38:39 their business model their business
1:38:41 model is predicated on you clicking on
1:38:44 lots of links to get the answer if they
1:38:46 just give you the answer you don't click
1:38:48 on the links and they don't make money
1:38:51 so they kind of Set It Off to the side
1:38:56 meta effectively did the same thing I
1:38:59 don't think Apple had
1:39:01 um
1:39:03 a big a big you know bet in the game
1:39:06 they're they're a hardware company they
1:39:08 sell they sell you know they make their
1:39:10 money primarily on Hardware
1:39:12 um
1:39:15 Microsoft when when gpt3 came out
1:39:18 Microsoft I think they saw an
1:39:20 opportunity and they said you know on on
1:39:23 seeing what gpt3 did
1:39:27 you know as it goes I think Sam Altman
1:39:30 went over to like like uh
1:39:34 the Microsoft CEO's house and I think
1:39:37 Gates was there like I I basically some
1:39:41 folks at Microsoft saw what the [ __ ] was
1:39:42 going on with gpt3 and I think Microsoft
1:39:45 saw an opportunity here
1:39:47 to maybe get relevant again in the
1:39:50 search game
1:39:51 and so they made this big investment in
1:39:53 open AI
1:39:55 as they were about to announce their
1:39:57 first billion dollar investment Sam
1:40:00 Altman said hey we don't really have
1:40:02 anything that we can point to point
1:40:04 people to that's really easy to use
1:40:07 in the documentation of gpt3 the the
1:40:11 developers version of
1:40:14 this thing in the documentations were
1:40:17 instructions for chat GPT
1:40:20 no one had built it for like a year and
1:40:22 a half
1:40:23 and so Sam Altman I guess back in
1:40:25 October this time last year
1:40:28 said to his developers hey we're making
1:40:31 this big announcement with Microsoft you
1:40:33 got two weeks why don't you just go
1:40:35 build chat GPT
1:40:37 and they did and they launched it and he
1:40:41 assumed no one would give a [ __ ] because
1:40:43 they didn't really give a [ __ ] with gpt3
1:40:45 like developers thought it was cool and
1:40:48 Microsoft thought it was cool and they
1:40:49 were going to give them a billion
1:40:50 dollars
1:40:52 chat apt went from zero users to 100
1:40:55 million users in six weeks
1:40:57 it got to a million users in five days
1:41:00 it was absolute [ __ ] insanity
1:41:03 so so what happened was this caused a
1:41:06 red alert at Google that said oh [ __ ] we
1:41:09 need to deal with this
1:41:11 um I think it did a similar thing at
1:41:12 meta where Zuckerberg was like okay we
1:41:15 gotta pivot off the metaverse thing
1:41:17 that's that ship has sailed even though
1:41:19 we named the company meta let's pivot
1:41:21 away from that and over to AI
1:41:25 um
1:41:26 Apple accelerated things like Amazon
1:41:29 accelerated things
1:41:31 um all of these companies like
1:41:33 inflection and cohere and
1:41:37 um
1:41:38 and uh
1:41:41 anthropic is clawed
1:41:44 um all had this stuff sitting in the
1:41:46 wing so so they've been developing on
1:41:49 this Transformer model for like six
1:41:51 years but what happened was Chachi BT
1:41:54 got so popular so fast that it forced
1:41:56 everyone to go [ __ ] and they just shoved
1:42:00 all of their technology forward so
1:42:02 although the technology that was sitting
1:42:04 kind of in The Ether off in the
1:42:05 background like well we'll take our time
1:42:08 and and we want to make sure that it's
1:42:10 safe and we don't want to hurt people
1:42:12 and
1:42:14 chat CPT does what it does and everyone
1:42:16 [ __ ] panics and throws all the
1:42:18 technology out there including meta
1:42:20 going here's we're gonna open source to
1:42:24 the developer Community llama one
1:42:27 but it's for research only and then when
1:42:29 llama 2 came out they're like [ __ ] it
1:42:31 here it is for commercial use
1:42:33 so there's there's now a incredibly
1:42:37 powerful
1:42:39 um open source commercial llm out there
1:42:42 stability AI is doing the same thing so
1:42:45 the developer Community is going that
1:42:48 [ __ ] crazy on this stuff the big
1:42:50 technology companies are going batshit
1:42:52 crazy on this stuff Apple just said that
1:42:54 they spent
1:42:56 the bulk of their research budget which
1:42:58 is 24.6 billion dollars a year they
1:43:01 spent the bulk of it on generative AI
1:43:03 this year
1:43:04 so so they're coming
1:43:06 so so
1:43:08 while it seemed like it materialized out
1:43:10 of nothing this was all being worked on
1:43:11 in all these companies but then the
1:43:13 popularity of chatgpt forced everyone's
1:43:16 hand to say we've got AI
1:43:19 Google's first meet first meeting the
1:43:21 the thing in Paris they did where they
1:43:23 they lost the demo phone and it was it
1:43:26 was pretty bad
1:43:27 um but like basically what they said in
1:43:30 that meeting was uh we invented this
1:43:32 stuff we've got Ai and everything which
1:43:33 they do
1:43:35 but they're on their heels because
1:43:39 they weren't leading this conversation
1:43:41 someone else was and it was Chachi BT
1:43:44 and Microsoft
1:43:46 and you know the C the CEO of Microsoft
1:43:48 was fairly giddy that he was [ __ ]
1:43:51 with Google
1:43:52 that was pretty funny
1:43:56 he basically said I'll lose money
1:43:58 on Bing chat on you know gpt4
1:44:03 um
1:44:04 for for like like even a if I can get
1:44:06 one user to come over from Google like
1:44:08 I'll lose money on that thing
1:44:12 um
1:44:13 my ex didn't sign up for a year it was
1:44:15 made obsolete in three months yep six
1:44:17 months exactly
1:44:19 how about investing in
1:44:21 the one AI startup I don't know it
1:44:26 I would I'm I'm not going to invest in
1:44:29 anything right now I don't think around
1:44:31 this stuff it's this is way too volatile
1:44:34 one AI generative AI API for your
1:44:36 product
1:44:47 so I guess this looks kind of like Lang
1:44:49 chain commercial version of Lang chain
1:44:52 maybe I don't know
1:44:55 maybe I don't know things are moving too
1:44:57 fast and it's like like one of the I
1:45:01 don't know if you saw there was a there
1:45:02 was a paper that came out or not a paper
1:45:04 it was like an internal Memo from a
1:45:06 Google employee that was like four
1:45:08 months ago was talking about
1:45:11 um
1:45:13 time is it going on 10. um
1:45:16 it was talking about the the title of
1:45:19 the piece was we have no moat and it was
1:45:22 basically talking about large language
1:45:23 models and how the open source models
1:45:26 are getting as good as or close to as
1:45:28 good as the really expensive models that
1:45:31 the big tech companies are building
1:45:33 and there's a similar sort of concern
1:45:35 that any piece of technology that you
1:45:38 build any any business that you build
1:45:40 and launch
1:45:41 because of generative AI
1:45:46 someone else could knock it off in a
1:45:48 weekend
1:45:50 someone without technical chops could
1:45:52 knock it off in a weekend
1:45:54 and so
1:45:57 everyone right now is going to be
1:45:59 scrambling to find what is their
1:46:01 differentiator in the marketplace
1:46:03 and
1:46:05 I don't know that it's going to be on
1:46:07 software features
1:46:09 I think the software features are just
1:46:11 going to become ubiquitous where
1:46:13 the the interfaces English language I
1:46:15 want software that does this and it'll
1:46:18 just make that software for you why
1:46:20 would I need to pay someone to do that
1:46:22 now
1:46:23 because the tools of janky pieces of
1:46:25 [ __ ] right now it doesn't do that right
1:46:27 now but the writing is on the wall that
1:46:30 when they start incorporating these
1:46:32 things together and overlapping them and
1:46:33 if if Google's Gemini comes out and it
1:46:36 is as sophisticated as it sounds like
1:46:39 it might be able to do things like
1:46:42 interface and code and ux and marketing
1:46:45 and and and and
1:46:48 right or it becomes the foundation for
1:46:50 that
1:46:52 so I would be reticent to invest in
1:46:55 anything right now unless unless you're
1:46:57 doing stuff kind of at the at the
1:47:00 like the Frontier Model the the the core
1:47:03 research the the sort of underlying [ __ ]
1:47:05 where these where the big the big boys
1:47:07 and girls are competing like those are
1:47:10 interesting things because whoever wins
1:47:13 that race is gonna you know they're
1:47:15 gonna have like infrastructure right
1:47:19 they're going to be selling the picks
1:47:20 and shovels to the miners
1:47:23 all of these software applications are
1:47:25 I've got a mine let's go get some gold
1:47:27 in the mine
1:47:31 so
1:47:33 I can't I can't [ __ ] keep up with
1:47:36 just even using this stuff much less
1:47:38 investing in it
1:47:40 what's the mid-journey prompt line
1:47:42 battle
1:47:43 I don't know what that means togas
1:47:47 write it in French maybe we can
1:47:49 um maybe I can translate it and it'll
1:47:51 make sense
1:47:53 um time to do the store and Link
1:47:56 um love kids saying it I love Turtles
1:48:00 I don't know what you're talking about
1:48:01 micro
1:48:05 um
1:48:12 [Music]
1:48:17 back here live at the waterfront village
1:48:19 with my friend the zombie Jonathan
1:48:21 you're looking good Jonathan just got an
1:48:23 awesome face paint job what do you think
1:48:25 I like turtles all right your great
1:48:29 zombie good times here at the waterfront
1:48:32 Village open for the next 11 days
1:48:40 all right
1:48:44 laughs
1:48:51 uh some comments were filtered to
1:48:54 protect the community's experience don't
1:48:56 be jerks and comments
1:48:58 no no
1:49:01 stop it stop it zip it zip it zip it zip
1:49:06 it zip it zip it
1:49:09 thank you Cheryl
1:49:11 thank you for the cruise
1:49:17 should have been terms of service for
1:49:19 tick tock with the AI yeah exactly
1:49:21 I actually don't want to know what's in
1:49:23 The Tick Tock terms of service but I'm
1:49:25 sure my family doesn't belong to me
1:49:27 anymore can some of my friends here
1:49:29 please tell me when I can get duet AI so
1:49:32 if you haven't seen it so
1:49:35 um so what Anne's talking about is duet
1:49:38 is Google's branding for their version
1:49:42 of chat gbt being rolled out into all of
1:49:45 The Office Products of theirs and Google
1:49:47 meet and all that sort of stuff they're
1:49:49 calling it duet they're saying it's
1:49:51 generally available but you have to go
1:49:53 ask for it so let me see where the maybe
1:49:57 it's an obvious place
1:50:00 Google duet General availability
1:50:08 now it will be available to all of its
1:50:10 10 million users
1:50:18 Google workspace now available August
1:50:21 29th
1:50:23 yeah this is it
1:50:28 so go to
1:50:29 workspace.google.com blog
1:50:32 and then it basically it's the article
1:50:34 now available duet AI for Google
1:50:37 workspace and then somewhere in here
1:50:40 yeah the second the second hyperlink in
1:50:43 the first paragraph
1:50:45 is get started now with a no cost trial
1:50:50 so it's free initially and then I think
1:50:52 it's going to be 30 bucks a month
1:51:00 with so many AI powered applications out
1:51:02 there is any is there any room for any
1:51:04 Innovation holy crap yes holy crap yes
1:51:08 so
1:51:09 here's the deal uh what's your what's
1:51:12 your name Thai tyrellius zakarius
1:51:17 okay that's a cool name I can't
1:51:19 pronounce it but it's super cool
1:51:20 tyrellius
1:51:22 zuccharius I like it solid 8.5 out of
1:51:26 10. only because I can't pronounce it
1:51:28 this is probably more on me than you
1:51:30 um
1:51:34 so the innovation
1:51:39 the tools are going to do remarkable
1:51:41 things so there's there's so many places
1:51:42 to innovate one of the places to
1:51:44 innovate is how are you using these core
1:51:47 large language models you could you
1:51:49 could kind of go upstream and you could
1:51:51 say hey I want to train you know
1:51:55 um these image recognition models on a
1:51:58 particular kind of Imaging right that's
1:52:00 what's happening in healthcare where
1:52:01 they're training them on how to read
1:52:03 X-rays and pet scans and things like
1:52:05 that
1:52:06 um
1:52:08 you could build application layers on
1:52:11 top of large language models you could
1:52:13 use something like zapier if you're not
1:52:16 a
1:52:16 a coder and do automations like one of
1:52:20 the things that I've built in zapier are
1:52:22 content generation engines where you ask
1:52:24 people a few questions they answer those
1:52:26 questions and it generates a bunch of
1:52:28 social media content or it generates an
1:52:30 article or whatever it might be
1:52:32 um so you can do it there
1:52:34 so so there's there's sort of the whole
1:52:36 world of application development that
1:52:38 that can be super technical all the way
1:52:41 up to you know fairly kind of front-end
1:52:43 easy kind of stuff
1:52:45 then you've got
1:52:48 all of the business process innovation
1:52:52 how are we going to do project
1:52:53 management moving forward how are we
1:52:55 going to do HR how are we going to do
1:52:58 customer support right
1:53:00 you know imagine taking these tools and
1:53:04 completely Reinventing the customer
1:53:05 support experience with synthesized
1:53:08 voice actually picking up the phone and
1:53:10 answering questions and being competent
1:53:12 and having the authority to to do stuff
1:53:15 right that's going to require obviously
1:53:17 some some technology stuff but even if
1:53:20 someone builds that tool
1:53:22 there's all sorts of innovation around
1:53:24 how do you roll a tool like that out
1:53:26 across a company that does customer
1:53:29 support support the way they do it today
1:53:31 every single function whether it's
1:53:34 copywriter or art director or consultant
1:53:38 or researcher or engineer or
1:53:42 anyone nurse anyone who's dealing with
1:53:45 any
1:53:46 computer
1:53:48 any computer mediated job which is
1:53:51 essentially in today's society it's
1:53:53 almost every job even if you're a
1:53:55 mechanic you know working on cars all
1:53:57 day long how do you take money with
1:53:59 computers how do you schedule people
1:54:01 with computers how do you send messages
1:54:04 with your phone with computers right so
1:54:08 almost all of our Workforce today is
1:54:13 computer mediated anything that's
1:54:15 computer mediated is going to be
1:54:17 transformed by these tools
1:54:20 how we deal with that transformation and
1:54:23 and what that transformation is that's
1:54:25 all the Innovation that like like every
1:54:29 single thing is going to be Innovation
1:54:31 even though these things are doing
1:54:32 remarkable things it's up to us to
1:54:35 figure out how we want to wield them to
1:54:37 make the change to make the difference
1:54:40 so [ __ ] massive opportunity massive
1:54:44 massive massive
1:54:47 Upstream Downstream
1:54:49 change management holy [ __ ]
1:54:53 people that are like I can help you yeah
1:54:57 so imagine you're working in an industry
1:55:00 like the advertising business which I
1:55:02 know pretty well
1:55:04 and you've spent the last five decades
1:55:07 six decades
1:55:09 charging time and materials basically
1:55:11 the way you get paid is you tell people
1:55:14 I'm going to have these 50 people
1:55:15 working 40 hours a week at 120 an hour
1:55:19 to write you copy for all of your
1:55:20 advertising
1:55:23 well now a tool comes along that can
1:55:25 just do that so rather than 50 people at
1:55:30 40 hours a week I might be able to do
1:55:33 the same amount of work at the same
1:55:34 level of quality with 10 people working
1:55:38 20 hours a week
1:55:42 what happens to that advertising
1:55:45 executive
1:55:47 that spent the last 50 years with a
1:55:49 business model
1:55:50 that just got obliterated
1:55:54 they're going to need help they're going
1:55:56 to need to figure this out so there's a
1:55:58 role for someone to just
1:56:00 figure out
1:56:02 all of the [ __ ] that just exploded how
1:56:05 do we tape it back together how do we
1:56:06 keep these companies still running so
1:56:08 it's like Innovation is going to be
1:56:10 everywhere
1:56:12 and with Innovation comes pain and
1:56:15 you know suffering and weirdness and
1:56:17 awkwardness and
1:56:19 resistance and fear and excitement and
1:56:24 young little upstart agencies that are
1:56:26 going to come in and say hey
1:56:27 I'm going to charge a flat fee for
1:56:30 advertising and we're going to deliver
1:56:32 things twice as fast at half the price
1:56:34 you want some and they're gonna become
1:56:37 the next big thing
1:56:39 and the big old agency that resists that
1:56:42 Change is Gonna potentially fall off the
1:56:44 Wayside
1:56:45 Fall by the wayside some of the big
1:56:48 agencies will make the transition some
1:56:49 won't
1:56:50 and I think about every single [ __ ]
1:56:54 business is going through that everyone
1:56:57 are we there yet no
1:57:00 is it coming yeah
1:57:03 how fast I don't know
1:57:06 it ain't gonna be slow it ain't gonna be
1:57:08 five years
1:57:10 it's gonna be a year two it's already
1:57:13 happening
1:57:15 so yes much much Innovation coming
1:57:20 we're in the middle of it that's why I'm
1:57:21 so excited this is like these are the
1:57:24 you don't get to live through these kind
1:57:26 of transitions
1:57:29 very often
1:57:31 in a generation if ever
1:57:34 in my life the three that I that I
1:57:38 think are them are the personal computer
1:57:41 I was too young for that the World Wide
1:57:44 Web
1:57:45 and generative AI
1:57:49 and after I did the World Wide Web thing
1:57:51 I didn't think I'd see another one of
1:57:53 these in my lifetime
1:57:54 and then Along Came this thing this
1:57:56 thing
1:57:59 it's crazy it's crazy it's crazy all
1:58:02 right I'm gonna jump to the bottom of
1:58:03 more than 99 questions behind there's 96
1:58:07 people in here 92 now
1:58:09 and I'm going to answer a few more
1:58:12 questions and then I'm going to go to
1:58:13 bed about Sunday night we got a labor
1:58:15 day tomorrow
1:58:16 we got do you like barbecue is Memorial
1:58:19 day but is barbecue also Labor Day I
1:58:22 don't I don't know I don't know I don't
1:58:25 know how to celebrate this particular
1:58:26 holiday although I do find it curious
1:58:30 that
1:58:31 on Labor Day we've got this this tool
1:58:35 amongst us that is going to transform
1:58:40 how we work in pretty dramatic ways
1:58:44 so that's interesting well ain't that
1:58:47 special and that's special yeah
1:58:50 let's see I wonder who who invented AI
1:58:52 who Could That Be on huh oh I know who
1:58:55 it was was it Satan
1:59:00 all right
1:59:01 I hate being too perplexity is amazing
1:59:03 just use Dolly through Bing that's all
1:59:06 yep that's I use it for that if I want
1:59:08 to do it really quick but now that now
1:59:10 that um
1:59:12 uh this exists by the way
1:59:15 go to ideogram dot a i i d e o like the
1:59:19 company Ido but it's not them
1:59:23 graham.ai
1:59:25 and it's just like one of these sites
1:59:28 that has like a bunch of images on it
1:59:29 but notice what the images have they all
1:59:31 have words
1:59:33 and so you can just
1:59:37 make words
1:59:41 um
1:59:42 so I can put in
1:59:44 make
1:59:47 money with chat GPT
1:59:51 exclamation point
1:59:53 and then I'll go
1:59:55 um let's see
1:59:57 Roger black
2:00:02 Style
2:00:04 poster
2:00:08 and we'll do poster and we'll do
2:00:10 typography and we'll do vertical and
2:00:13 we'll do generate
2:00:32 make money
2:00:35 it doesn't always get it right
2:00:38 but against it right now like that's not
2:00:40 even close
2:00:42 make make s's with money chat GPT make
2:00:47 this with chat you BT
2:00:50 um
2:00:51 but yeah you can do all sorts of stuff
2:00:52 you can do words you can do pictures so
2:00:54 this is the one that I use now
2:00:56 instead of uh instead of being or
2:00:59 anything quick because it's just it's
2:01:00 super simple that's all it is you you
2:01:01 pick vertical horizontal or Square you
2:01:05 put in a prompt the quality of the image
2:01:07 rendering on this one is like Dolly it's
2:01:09 not super great
2:01:11 um it's maybe a little bit better than
2:01:12 Dolly but it's not mid-journ if you want
2:01:15 good images go to Mid Journey but if you
2:01:17 want something like I love Comic Sans
2:01:19 well that one should be in comic sense
2:01:22 shouldn't it
2:01:24 um but like there's a Netflix logo it
2:01:26 does logos pretty good I had to do a
2:01:28 couple of NFL logos and it got pretty
2:01:31 close
2:01:32 a pig saying eat more chicken cute
2:01:38 I just baked some cupcakes
2:01:43 some [ __ ] up cakes
2:01:47 that's awesome
2:01:49 so yeah that's ideogram.ai
2:01:55 pretty slick ain't it ain't it pretty
2:01:58 good and pretty good a pretty good yeah
2:02:01 they're pretty good mm-hmm
2:02:04 I like them taters
2:02:08 all right let's see 945 in Chen land
2:02:11 thank you Emilio's wife which AI can you
2:02:13 use to help me with my business
2:02:15 um
2:02:16 all of them really but but just
2:02:19 quite honestly the the one to really
2:02:21 just pay attention to right now if
2:02:23 you're just getting started with this
2:02:24 stuff is just go to chat gbt it's got
2:02:27 this thing called
2:02:28 um
2:02:30 Advanced Data analysis
2:02:33 um which you can use to analyze business
2:02:36 spreadsheets and things like that and
2:02:37 all sorts of data but but just learning
2:02:41 chat GPT for your business start with
2:02:43 what are you trying to do within your
2:02:45 business and then start asking chat GPT
2:02:47 about the things the problems you want
2:02:49 to solve and just
2:02:50 wait for it to blow your [ __ ] mind
2:02:53 and then you can also go to this thing
2:02:55 prompts.chat that little bottom thing
2:02:57 there that will teach you how to talk to
2:02:59 chat GPT and it'll give you a lot of
2:03:01 ideas for how you can use it that's
2:03:03 where I would start there may be more
2:03:05 specialized tools like Jasper or copy AI
2:03:08 for marketing or or things that are more
2:03:11 specialized for doing numbers or doing
2:03:13 whatever websites or marketing what all
2:03:16 there's there's going to be all sorts of
2:03:17 specific tools
2:03:19 almost all of them are based on the core
2:03:21 underlying technology here so if you get
2:03:23 your head around what these things make
2:03:25 possible then you can start to look for
2:03:27 specific tools that solve problems if
2:03:29 you can't solve them yourself
2:03:33 30 at 38 000 we're up to 49 000 likes
2:03:37 you all are nuts you get nuts and I love
2:03:40 it I love it I love it I love it all
2:03:42 right check out factaverse.ai for free
2:03:45 fact checking it's been a while since I
2:03:47 used it but bookmarked very cool
2:03:49 we've seen Facebook
2:03:54 um each platform has its own strengths
2:03:55 you said Apple
2:03:57 um yeah Apple's Gonna Be Coming Out
2:03:59 Swinging with this AI stuff they're
2:04:02 gonna do it right and they're gonna be
2:04:03 good and here's the thing about Apple
2:04:06 is a very good reputation
2:04:09 for not [ __ ] with our privacy you
2:04:12 know refusing to open back doors for
2:04:15 governments and doing things like having
2:04:19 you know keeping private Keys like
2:04:21 locally locked on your Hardware
2:04:24 um so Apple's got a really good
2:04:25 reputation for privacy
2:04:27 and when they do this AI stuff they're
2:04:30 going to do it locally if you paid
2:04:32 attention to the Vision Pro
2:04:34 announcement one of the chips in so a
2:04:37 couple of things within the M2 chips
2:04:39 there's all sorts of machine learning
2:04:41 [ __ ] that can be done in the M2 chips
2:04:43 they also have this thing called the the
2:04:45 R1 the the reality chip
2:04:48 which is taking the input from all the
2:04:50 sensors on the goggles and doing
2:04:52 real-time machine learning
2:04:55 on device on that chip so Apple's gonna
2:04:58 be [ __ ] phenomenal at local
2:05:03 machine learning processing
2:05:05 which is basically means your data's
2:05:08 safe right you're not sending it to them
2:05:10 you're not sending it to open AI you're
2:05:11 not sending it anywhere you're going to
2:05:13 be able to take your data
2:05:15 GPT the [ __ ] out of it and feel
2:05:18 confident that all of that stuff is
2:05:20 within your own little data bubble
2:05:23 um so I think and and they're also
2:05:24 they're also just really good at
2:05:26 defining new categories so I think
2:05:28 they're just going to be good at that
2:05:29 [ __ ]
2:05:30 they won't be good at it
2:05:33 any other are llms anything more than
2:05:37 innovation in HCI what's HCI human
2:05:41 computer interface
2:05:43 um
2:05:43 they are
2:05:45 so so yeah JD
2:05:53 I mean
2:05:54 I would I would suppose that technically
2:05:57 you could abstract it up to it's an
2:05:59 improvement in human computer interface
2:06:01 but here's the big difference
2:06:03 the way I think about it
2:06:05 historically
2:06:08 if I want to
2:06:11 understand some data
2:06:14 I've got to find the data set
2:06:17 I've got to look at all the data I've
2:06:19 got to clean up the data like like if
2:06:22 I've got if people's addresses are typed
2:06:24 in there in inconsistent formats where
2:06:26 some of them split the data out into
2:06:28 different fields and some of them are in
2:06:30 the fields I've got to clean up all that
2:06:32 data and then I as a data analyst or as
2:06:35 a programmer need to think about okay
2:06:38 I've got the following data elements and
2:06:41 I want to accomplish the following test
2:06:43 task
2:06:44 how do I need to logically think about
2:06:47 how I might combine and
2:06:50 analyze this data in such a way that it
2:06:54 gives me some desired result and I have
2:06:57 to do all of that logical thinking
2:07:01 you can now
2:07:03 just ask this [ __ ]
2:07:05 to figure it out for you
2:07:08 or you can just say hey so here's
2:07:11 another example
2:07:15 here's how Google works
2:07:17 Google you say I want to learn about the
2:07:21 Mesopotamian War and Google's like
2:07:23 awesome I got all the links of all the
2:07:26 people talking about the Mesopotamian
2:07:28 War for all of Internet time here's all
2:07:31 your links
2:07:33 and then the first seven links are ads
2:07:35 to buy Mesopotamian t-shirts
2:07:38 right they look like articles but
2:07:42 they're really just ads for t-shirts
2:07:43 they're like God damn answer
2:07:46 and then and then over time you click on
2:07:49 enough links and then you go read the
2:07:51 articles and you think okay I got the
2:07:54 Mesopotamian word started over this and
2:07:56 then it did this and and you're you're
2:07:58 learning right and this is taking you
2:08:00 however long it's taking you a week two
2:08:01 weeks
2:08:03 and then you learn enough about that war
2:08:05 and you're like hey I wonder what the
2:08:06 what the
2:08:08 comparison is of the evolution of that
2:08:11 war with World War II because I know a
2:08:14 lot about World War II
2:08:15 so you know what would be really cool is
2:08:17 if I compared them so I'm going to take
2:08:20 the starting points of each of them so
2:08:22 I'm going to go research this and I'm
2:08:24 gonna go research this and then I'm
2:08:25 gonna put it in a in a Google doc
2:08:28 like start of war and this one started
2:08:31 here and World War II started here and
2:08:33 then and I'm gonna okay with these
2:08:36 things
2:08:37 you go here
2:08:39 I'm gonna do GPT 3.5 because it's just
2:08:42 really [ __ ] impressive and I gotta
2:08:43 turn off my custom instructions because
2:08:45 I think it's acting like an idiot right
2:08:47 now
2:08:48 yeah I'm gonna turn that off okay
2:08:51 new chat GPT 3.5
2:08:59 you're a history Professor that
2:09:02 specializes
2:09:03 in the analysis
2:09:06 of major Wars period period
2:09:14 make a table comparing
2:09:18 the major historical Milestones of the
2:09:22 Mesopotamian War and World War II
2:09:27 [Applause]
2:09:35 you could say to me that that that's
2:09:37 just an improvement in human computer
2:09:40 interface
2:09:42 but this is something different right
2:09:44 this is knowledge on demand Google was
2:09:47 information on demand but it was up to
2:09:49 us as humans to learn the [ __ ] to find
2:09:52 the [ __ ] to sift through it to think
2:09:54 logically about how one thing compared
2:09:56 to the other
2:09:57 this just does it time period 3000 BC to
2:10:02 539 BC World War II 39-45 Geographic
2:10:06 scope primarily centered in Mesopotamia
2:10:09 Geographic scope Global right so it
2:10:12 understood what I was asking for it gave
2:10:14 me things now that was without even
2:10:15 trying without really
2:10:18 knowing anything about it if you knew
2:10:19 something about it you could do
2:10:21 incredibly sophisticated
2:10:23 analysis of these two things
2:10:26 um essentially instantly now might it
2:10:30 hallucinate and give you wrong things
2:10:31 yeah it might will it do that two years
2:10:34 from now no it won't right so yeah it's
2:10:38 a fundamentally different thing it's a
2:10:40 fundamentally different
2:10:42 way to interact with computers so it's
2:10:45 not just an improvement in in interface
2:10:50 in my opinion but
2:10:52 I don't know I could be wrong
2:10:59 all right I've got 10 minutes what do
2:11:03 you think about this what is it as good
2:11:05 as the hype says what oh idx.dev
2:11:09 I have not heard the hype on it but
2:11:11 let's go look at it
2:11:13 idx dot uh
2:11:20 it says it couldn't reach the page
2:11:24 idx.idx.dev
2:11:27 it must be wrong
2:11:31 I like turtles hi Kyle hello Robert
2:11:34 Rossi
2:11:36 Kam katkin la la la la
2:11:39 camp cat kitten pleased to meet you
2:11:44 actually cam catkin would be a great
2:11:46 name for someone selling
2:11:48 um a class on making money with GPT hi
2:11:51 Kim cat can hear you can make money with
2:11:53 your gbt
2:11:55 I'm Kim catkin
2:11:59 this is get rich with chat EBT large
2:12:02 language models are your friend
2:12:05 the first class is free
2:12:07 [Laughter]
2:12:09 would you prefer Starbucks or
2:12:12 oh that's cute I'm not gonna say it
2:12:15 because I'll probably booted off Tick
2:12:16 Tock best businesses
2:12:21 um chat EBT helped me write a cool book
2:12:24 nice
2:12:26 um it's how good you are at props how to
2:12:28 monetize
2:12:29 um
2:12:31 you got to figure out a way to use these
2:12:34 tools in a way that people are going to
2:12:35 pay you to use them clip is hilarious
2:12:38 sorry
2:12:40 a couple of years at least
2:12:43 it's in the mail let's see I'm jumping
2:12:46 to the bottom here because I only got a
2:12:49 few more minutes totally agree I've been
2:12:51 in sales Business Development accounting
2:12:53 and HR I see Untold potential yup
2:12:57 finish the summer party
2:12:59 oh that's what Labor Day is yep
2:13:02 the future of AI is coming like a
2:13:04 freight train with Ray Kurtz Violet the
2:13:06 helm yeah I just said there's an
2:13:08 interview with Ray from about three
2:13:10 months ago or so uh where he's talking
2:13:12 about you know the singularity and they
2:13:15 were talking about when is it coming and
2:13:17 he said his prediction years ago was
2:13:19 2029.
2:13:21 and a bunch of AI people just sort of
2:13:24 lowered their estimate from like 20 40
2:13:26 or 2038 to 2030. so people just sort of
2:13:30 lowered their estimate from like 20 40
2:13:32 or 2038 to 20 30. so it seems like in
2:13:37 the next
2:13:38 in the next handful of years we're going
2:13:40 to be seeing some pretty remarkable [ __ ]
2:13:44 I reckon I like taters
2:13:50 all right what about llama llama big
2:13:53 deal llama big deal llama two
2:13:57 and code llama so if you want to go play
2:14:00 with llama you can go to
2:14:01 labs.perplexity.ai if if you're not
2:14:03 geeky and don't feel like installing
2:14:05 [ __ ] in Virtual machines or on your
2:14:08 machine
2:14:09 lab stop perplexity and thank you for
2:14:12 the Llama Winston labs.perplexity lets
2:14:15 you play with
2:14:16 llama 2 7 billion 13 billion 70 billion
2:14:20 and then code llama
2:14:22 and code llama just beat gpt4 for coding
2:14:27 so
2:14:29 shit's getting better and it's free and
2:14:32 it's open source and it's it's not
2:14:34 really open source it's commercially
2:14:36 licensable but you can use it without
2:14:38 paying them a penny all the way up until
2:14:41 you have more than 700 million customers
2:14:45 so as long as you have less than 700
2:14:47 million customers you're good to go
2:14:50 it's a big big deal
2:14:52 Thelonious sorry if I appeared
2:14:54 condescending
2:14:56 whatever
2:14:57 double check
2:15:00 it's not always accurate uh I did okay
2:15:03 thank you for that
2:15:07 have you spoken about Chachi BT ask chat
2:15:10 gbt
2:15:11 oh Chad GPT
2:15:14 no is It Like a Dude someone fine-tuned
2:15:17 it to act like a dude we could do that
2:15:18 we could we could make it act like a
2:15:20 surfer dude
2:15:21 uh
2:15:23 sorry it's not called Twitter anymore
2:15:24 Nancy it's called X now I don't know if
2:15:27 you knew that but now it's called X okay
2:15:29 Nancy all right let's say Chad
2:15:34 GPT or yeah GPT
2:15:39 gets it
2:15:48 is it an account
2:15:56 hey bro you heard heard you're looking
2:15:59 for
2:16:00 big gains oh Chad GPT has you covered
2:16:02 here oh so it's like a crypto bro oh
2:16:05 here it is Chad GPT with a check mark
2:16:08 what's up Bros
2:16:10 yeah perfect
2:16:13 awesome
2:16:13 [Laughter]
2:16:16 that's awesome
2:16:18 oh hilarious hilarious
2:16:21 what AI double learning lab is still on
2:16:24 Tick Tock is it really cabrino
2:16:27 uh are you serious
2:16:31 I thought
2:16:39 all right so
2:16:41 if that's the case
2:16:51 it's back but it doesn't look like it
2:16:53 has content
2:16:57 all right let me we'll do a little
2:16:59 Service Announcement here all right so
2:17:01 this sucks if it's back assistant little
2:17:03 [ __ ] isn't he
2:17:04 so this is us this is this
2:17:09 this is an [ __ ] who is a scammer
2:17:12 so if anyone reaches out to you
2:17:14 purporting to be me asking you to invest
2:17:17 in some crypto [ __ ]
2:17:19 check really closely it's really hard to
2:17:21 miss that there's two lowercase L's
2:17:24 together I put them large just so you
2:17:26 can see it
2:17:28 um that sucks
2:17:29 so anyway just be careful out there be
2:17:32 careful out there folks all right we did
2:17:34 we're gonna go we're gonna get on out of
2:17:36 here listen I I think Tick Tock would
2:17:39 not be happy with me if I didn't say so
2:17:41 we've got a
2:17:43 um a store in here now so I've got three
2:17:47 pieces of merch I just put it up tonight
2:17:48 so I'm just trying to figure out if I
2:17:50 can get it to work
2:17:52 um so that's there you can go check that
2:17:53 out
2:17:54 I'll probably add more stuff to that but
2:17:57 if Tick Tock keeps telling me I've got
2:17:59 to talk about the store more I'll just
2:18:01 pull it down and put it somewhere else
2:18:02 we'll just we'll just make a store
2:18:03 somewhere else and I'll just point
2:18:05 people to it
2:18:07 um
2:18:07 all right kavuno I had it make a whole
2:18:10 bunch of content this morning cool all
2:18:12 right listen folks happy Sunday night
2:18:14 been cool hanging out with you
2:18:17 um have a fantastic evening I'll see you
2:18:19 tomorrow sometime have a fantastic Labor
2:18:22 Day and uh let's do our little Romper
2:18:24 Room Roundup here I see let's see I see
2:18:29 Nancy
2:18:30 and I see Winston and I see Riff Raff
2:18:35 and Matthew Craig official be careful of
2:18:38 Matthew Craig not official it's not the
2:18:41 same it's not official
2:18:43 harsh weed three cabruno peace Amelia's
2:18:47 wife good night good night Tobias
2:18:50 um let's see Stevo Firefly Robert Rossi
2:18:54 Chris Rock rocker
2:18:57 Omar domain night from us oh cool we got
2:19:01 Australia in the house is it morning
2:19:03 there it must be morning there uh down
2:19:06 in the mud
2:19:07 all right everybody cam catkin this is
2:19:10 camcatkin checking out you can make
2:19:12 money with judge EBT I'm Kim catkin see
2:19:16 y'all