aiLL 8/02/23 - I'm Back! The Wild West of AI Explodes While I Was Gone

Description

Chapters

Transcript

0:01 foreign
0:26 I'm back
0:28 I was I was in the doghouse for seven
0:30 days
0:33 [Laughter]
0:36 I'm back [ __ ]
0:39 hey Robert what's happening David gold
0:42 thanks for the world I am back
0:46 andre44
0:48 Andre Andre yeah
0:50 Andrea Andrea
0:53 Andrea Andrea thank you I was afraid
0:57 you'd been abducted by aliens yeah
0:59 exactly I got hacked by AI
1:03 I'm out of jail
1:06 hey Troy thank you missed you guys too
1:08 welcome back thank you
1:12 I don't even know what to say I haven't
1:15 done this in so long I'm gonna have to
1:17 help me through this
1:18 welcome welcome everybody my name is
1:20 Kyle Shannon this is the AI learning lab
1:21 thanks for doing this stream much
1:23 appreciated you're welcome I'll try not
1:25 to get my ass kicked off this time
1:27 I was doing pretty good I had like a
1:29 two-month run going and then yeah I made
1:31 a stupid joke and no no I quoted someone
1:34 making a funny joke and but I think what
1:37 they said
1:38 hmm not good so anyway they booted me
1:41 whatever can you explain what happened
1:43 yeah I said the heart r word I I quoted
1:46 a comedian who said the hard word and I
1:48 think me quoting it was classified as
1:51 hate speech
1:54 whatever
1:55 doesn't matter we're back we're back
1:58 we're live uh if you're new here my name
2:01 is Kyle Shannon this is the AI learning
2:02 lab we learn about Labs here oh and AI
2:06 uh since I've gone away we've got room
2:09 temperature uh
2:11 superconductivity very controversial
2:14 might be faked some of the scientists
2:16 that did it were apparently part of some
2:18 faking that happened 20 years ago
2:21 oh
2:21 might not happen but you know
2:24 if we're going to transform the world
2:25 with AI we might as well have
2:27 superconductivity followed by free
2:29 energy and cures for cancer and all that
2:31 [ __ ] right Tobias was shaken told you
2:33 I'd be back later tonight I don't know
2:35 how late I can stay up it's a little
2:37 late already but I came back as soon as
2:39 I could it was 21.52
2:42 mountain time today I wasn't allowed to
2:45 come back before then
2:47 uh anyway uh let me get a sip of water
2:50 here and we'll get jump in
2:57 all right let's do this so
3:00 if you're new to AI
3:03 it's a generative AI if you're like
3:06 what is this hey ask stuff everyone's
3:08 talking about I don't know
3:13 these things here are places to get
3:16 started and you're like but Kyle which
3:18 one's best
3:19 the top one the top one is best in my
3:22 opinion just because
3:24 that's the official chat GPT website
3:26 that's gpt4
3:29 it's uh
3:30 it's got all sorts of cool [ __ ] on it
3:33 like
3:35 custom instructions and code interpreter
3:37 and plugins and it's just it's just the
3:39 best now
3:41 I don't think it's gonna be the best for
3:42 long quite honestly because there's a
3:44 lot of competitors coming so Claude over
3:47 there that's ex open AI employees that
3:50 started a competitor and that thing has
3:53 a hundred thousand token context limit
3:54 so you can put 75 000 words into
3:58 claude.ai that's a big deal uh Bing is
4:02 gpt4 connected to the internet it can
4:04 also generate images it can also read
4:06 images
4:08 um bard.google.com used to suck really
4:11 bad it's better now it can read images
4:14 labs.perplexity.ai that's
4:17 um perplexity has given us access to the
4:20 The Meta llama 2 models for free without
4:24 having to install [ __ ] and then Pi dot
4:27 AI is a real chatty
4:29 chatbot and poe.com lets you do Bots and
4:33 things like that so anyway that's where
4:35 to get started
4:36 um other things to to pay attention to
4:38 if you're super brand new are these
4:41 things so prompts.chat will teach you
4:43 about prompting and there are many
4:45 places you can learn about prompting hey
4:47 Silver Fox thank you for the rose that's
4:49 so sweet
4:51 um
4:52 promise.chat will teach you about it
4:54 it'll teach you this idea of Mr cgo yeah
4:58 baby Chief generative officer first in
5:01 the world
5:03 uh
5:05 what the [ __ ] is the chief generative
5:07 officer exactly uh
5:10 openart.ai prompt book that'll teach you
5:13 about prompting things like mid-journey
5:15 futurepedia.io that's just a pile of
5:17 tools like if you want to go geek out go
5:19 there just know that 90 of them are
5:21 pieces of [ __ ] and then you got
5:23 mid-journey Adobe Firefly and Leonardo
5:26 which is built on top of stable
5:27 diffusion those are kind of the three
5:28 biggies on the image front
5:30 um if you're a geek go play with all the
5:32 stable diffusion [ __ ] if you're not play
5:34 with leonardo.ai
5:36 then you got Runway ml which is videos
5:38 and blockade Labs which is just kind of
5:40 fun thank you for the donut I love me
5:42 some donuts
5:44 this is just like one of those uh uh NPC
5:47 channels where if if you give me toys I
5:51 react to them because I'm a trained
5:53 monkey all right if you have questions
5:56 about AI pop those in the comments below
5:58 uh if you haven't been here before and
6:00 it's very possible because I've been
6:02 gone for a week and uh
6:04 and uh what's the new 100K Nvidia
6:08 perfusion model that was just announced
6:11 I don't know let's go look
6:13 I saw something cool
6:15 I saw something cool from Nvidia today
6:17 that was not the the profusion model is
6:20 that the image thing Nvidia
6:23 it might be
6:25 and if it is it's super cool per Fusion
6:31 and we present perfusion it's lack
6:36 confusion only Persia yeah this is this
6:39 looks really cool okay
6:41 so
6:42 so what this is so one of the grand
6:45 frustrations of large language models is
6:50 um
6:52 of the image generation models is that
6:56 it's really hard to do predictable stuff
6:58 and what this perfusion model is from
7:01 Nvidia is like it's it's actually pretty
7:05 similar to
7:07 um Google announced that thing called
7:09 the hell is it called not snap something
7:15 I don't know Google had this image thing
7:16 where you could upload one image and it
7:18 would it would mimic that style this is
7:21 similar where you upload a character and
7:24 then you know you can
7:27 make different versions of that exact
7:29 character that's really hard to do right
7:32 now
7:34 um and so like you know you upload that
7:35 teapot and then you can make a gold
7:37 teapot or a you know Japanese rose
7:40 teapot in a whatever crocheted Garden or
7:44 teapot with the bear you know in the
7:47 lake so uh this looks really cool it's
7:50 not out yet I just saw this today
7:52 um this looks like a big deal this was
7:54 accepted at siggraph
7:57 um but yeah but anyway there's a white
7:59 paper on it and it's called key locked
8:01 ranked rank one editing for text to
8:04 image personalization it doesn't matter
8:06 nvidia's got some cool [ __ ] coming out
8:08 so when that comes out image generation
8:10 is going to get a lot better
8:12 uh
8:14 uh Amelia's wife we need to know have
8:16 you missed us of course I've missed you
8:18 who the hell else do I talk to
8:20 who well the dog of course you talk to
8:22 the dog you have to talk to the dog
8:25 um but yeah I missed you a lot and and I
8:28 appreciate a lot of people reached out
8:29 some were concerned where you been
8:31 everything okay yeah I was just in Tick
8:33 Tock jail just Tick-Tock jail puppies
8:37 puppies back he's happy he's laying here
8:40 right beside me he loves hanging out on
8:42 the lives
8:44 all right so I am gonna go
8:47 find uh the first question and let's get
8:50 started
8:52 [Music]
9:00 [Music]
9:01 all right
9:12 Robert Rossi got the number one gifter
9:15 badge wow Robert that's exciting that's
9:18 exciting I don't know what the hell that
9:19 means but that's a deal
9:22 hello Triumph truth what's happening
9:24 naughty boy I was a naughty boy I was
9:27 put in jail can I explain what happened
9:29 I said a bad word apparently classified
9:31 as hate speech if you've been on this
9:33 channel you know I'm not a hater I'm a
9:35 lover
9:36 I'm not a pushover
9:38 don't [ __ ] with me but you know I'm a
9:40 lover
9:44 all right but I am a little Saucy so if
9:47 you get a little saucy
9:48 there's a chance you're gonna you're
9:50 gonna get your you're gonna get your
9:52 little fingers slapped through the ruler
9:54 so that's what happened
9:59 let's get this Armageddon started
10:02 already oh yeah AI might be an evil
10:05 thing it might be
10:07 it is equally likely in my opinion that
10:09 it will do
10:11 uh dramatic good it could also do
10:14 dramatic bad the reality like always is
10:17 it's probably somewhere in between there
10:18 will be some dramatically good stuff and
10:21 some really shitty stuff and
10:24 this channel is about getting curious
10:27 enough to not be at
10:32 huh
10:33 not be at the mercy of it that's not
10:36 really it it's just to understand it
10:38 enough to not be blindsided by it that's
10:40 that's really the purpose of the channel
10:44 um get people to try it get you excited
10:45 about it if you try it and you hate it
10:47 then great at least you tried it but a
10:49 lot of people right now are afraid of it
10:50 without trying it or judging it without
10:52 trying it or making assumptions about
10:54 how it works or what it does and that
10:56 the robots are gonna [ __ ] come get us
10:58 no no we're still in charge these are
11:00 tools these are tools they're tools just
11:02 like photoshop's a tool right and people
11:05 freaked out when Photoshop came out and
11:08 these are tools just like the calculator
11:09 was a tool University professors when
11:12 calculators came out mathematics
11:14 professors were on campus with little
11:17 placards ban the calculator
11:21 these kids must learn longhand Edition
11:24 um
11:26 same [ __ ] different technology but this
11:29 technology is
11:31 kind of
11:33 you know in the neighborhood of
11:35 fire printing press steam engine this
11:40 you could maybe throw the internet and
11:43 the World Wide Web in there but I don't
11:44 think so I I think that's more of a
11:46 minor blip this is a major one this is
11:48 Industrial Revolution scale [ __ ]
11:51 and what's exciting is
11:53 we're at the beginning of it even though
11:56 machine learning and ai's been around
11:57 for a long long time this generative AI
12:00 stuff
12:02 um has been around since last November
12:03 so like nine months into this stuff
12:06 and the competition is heating up I only
12:09 put [ __ ] on here that I think is worth
12:10 checking out the fact that there's seven
12:12 things on there that I actually think
12:13 are worth your time enough to say it's
12:15 worth checking them all out is amazing
12:17 to me
12:19 um
12:19 I'd like to get that down to three
12:23 um but anyway yeah
12:25 interesting you Americans are obsessed
12:27 about drinking water well I live in
12:29 Denver and it's a high desert so I'm a
12:31 mile in the air and it's dry all the
12:34 time except for this summer when it
12:36 thunderstorms every afternoon but
12:38 drinking water in Denver is like a
12:42 full-time job it's if and if if you
12:44 don't do it
12:46 uh you wake up in the morning or I wake
12:48 up in the morning with just a raging
12:50 [ __ ] dehydration headache
12:53 and it was and it took me because
12:55 because I'm a little dim it took me like
12:57 a year and a half to put together like
13:00 if I wasn't
13:02 [ __ ] pounding water
13:05 um that uh that that was what was
13:07 causing the headaches I'm like why do I
13:09 have all these headaches that's what it
13:10 was and so I bought I went on Amazon I
13:12 got myself 32 ounce tumblers I just fill
13:15 one of those bad boys up before I go to
13:17 bed every night and try to drink it
13:19 before morning
13:22 it's crazy and I don't have to pee all
13:24 night because it's just my body needs it
13:26 interesting yeah with iced tea coffee
13:29 all the time yeah I like ice in it
13:31 uh welcome back so MP's MPC streaming
13:35 didn't work for you that's my other
13:37 Channel That's where I've been I'm
13:38 making fifteen hundred dollars a day
13:40 going what do they say
13:43 um thanks for the rose thanks for the
13:45 rose oh cowboy hat oh cowboy hat I could
13:48 do that if you want to turn this channel
13:50 into that just start gifting me [ __ ] and
13:53 then we don't we don't have to talk
13:56 substance at all I don't I don't think
13:58 I'd last
13:59 two days doing that what's this new 100K
14:02 Nvidia perfusion model that's what I
14:04 just talked about
14:05 uh can't take a screenshot due to
14:07 privacy I don't know what that means to
14:09 bias you can't or I can't
14:12 oh can't take a screenshot of that over
14:14 there that might just be a local setting
14:16 on your on your phone
14:20 Kyle We need oh Tick Tock just jumped me
14:23 to the bottom oh the pet peeve so in the
14:26 seven days they kicked me off they
14:27 couldn't have fixed the scroll to the
14:29 bottom bug no no tick tock
14:32 you missed champ too he's right down
14:34 there aren't you Champion you good boy
14:38 a good boy
14:40 he's a good boy
14:42 all right
14:45 welcome back to the AI party
14:50 are free ways to clone voices without
14:52 paying the monthly subscription well
14:56 yes but so so there's gonna be plenty of
15:00 them right now the best one is 11 labs
15:02 and you need to pay money to clone your
15:03 voice there
15:05 um there's probably another one I don't
15:07 know about 11 Labs is the biggie but in
15:10 iOS 17 which I'm running right now
15:12 inaccessibility
15:15 if you go to accessibility and speech
15:18 there's a thing called personal voice
15:20 and within the IOS operating system you
15:24 can clone your voice
15:26 um
15:27 they're sneaking it into under
15:29 accessibility because there's some cool
15:30 things you can do with that from an
15:32 accessibility standpoint but
15:35 I think that's the foundation for some
15:36 other stuff now I'm on an iPhone 11.
15:39 this is a shitty old
15:41 four-year-old phone or however old it is
15:44 and uh
15:45 I don't have access to that particular
15:47 feature so I can't tell you if it's any
15:49 good but I'm pretty sure how it works is
15:51 they give you 50 phrases to say so
15:53 they're basically trying to get you to
15:54 say all the what are they called
15:55 phenomes phonemes
15:58 the the parts of words they get you to
16:01 say enough words that you say all the
16:02 parts of words so they have how you say
16:04 all your words
16:06 um
16:07 and you record all that [ __ ] and then it
16:09 makes a voice and then you can send
16:10 messages with it and things like that so
16:13 um it's it's coming it's it's gonna come
16:15 like like these tools are going to start
16:18 to be incorporated and all the other
16:19 tools we use things like synthesizing
16:22 your voice synthesizing your face all
16:24 that's going to come
16:26 um but I don't know other than the iOS
16:29 beta right now I don't know of anything
16:30 you can do it for free but there are
16:32 probably some out there
16:35 in fact there are probably some open
16:36 source ones so but I don't know
16:39 I don't know how to go to sleep without
16:41 Kyle thank you see okay now that's a
16:44 that's a solid that's a solid AI
16:46 learning lab troll where is it a
16:49 compliment is it a put down like like he
16:52 puts us to sleep huh you see you see
16:54 this is solid I like it it's good it's
16:57 comedy it's comedy comedy
17:00 all right who's a comedian that says
17:02 that Mark Norman
17:05 all right too much twerking we knew this
17:08 would happen I know I know I tried I
17:10 resisted the twerking for months and you
17:13 all insisted and I said fine I'll twerk
17:16 and then what happens I Twerk and by the
17:18 time I turn around to the camera my live
17:20 has been shut down
17:23 oh it wasn't twerking I was just trying
17:25 to get my hair to do something normal
17:28 yeah failed all right I tried Gen 2
17:31 Runway not impressed at all
17:35 um try if you haven't tried image to
17:39 video It's better
17:41 it's better than so uh so what's your
17:44 name Matthias Windsor yeah the the text
17:47 to video for Gen 2 is wonkier than [ __ ]
17:51 now if you if you sort of look a year
17:55 before images got good they were
17:57 laughably bad I think the same thing is
17:59 going to happen with video but
18:01 in Gen 2 right now
18:03 definitely on their website and I think
18:06 they've added it to their uh their phone
18:08 app yeah they have
18:10 you can upload an image and turn that
18:13 image into
18:15 um a video and in fact I'll show you one
18:17 that I did
18:19 of a cloud photograph I took it's pretty
18:21 [ __ ] good dude
18:24 Matthias you know deed dude
18:32 let's go check it out
18:34 why is he talking like that that's just
18:36 what he does that's just what he does
18:37 he's you know he's okay he's for one of
18:40 them computer Smart Guys he's all right
18:43 you know he's all right
18:46 he's he's a little
18:47 he's a little bit of a dum-dum
18:52 and uh
19:02 barberheimer that was a pretty cool clip
19:04 oh there was there was me with Barbie I
19:06 thought that was that was good for the
19:08 uh for the Barbie movie
19:10 I I put that I was her uh her uh rebound
19:15 relationship after Ken
19:16 Gotta Laugh or two
19:18 ah here we go check this out
19:22 so
19:24 um
19:26 so that photograph
19:28 is a panorama I took with my own camera
19:32 and did some editing in Lightroom so
19:34 that's
19:35 if you want to see my photography you
19:37 can go to Kyle shannon.com
19:39 that's a bunch of my photography so
19:40 anyway so I started with that I uploaded
19:43 that to Gen 2 with no prompt and it
19:46 turned it into this
19:51 I mean
19:52 come on
19:54 now it does lose a lot of detail right
19:57 so it's like it's definitely
20:00 it's definitely early technology and
20:02 it's only four seconds
20:04 but I mean if you cut that into you know
20:07 other footage like
20:10 as as part of an edit that's that's
20:12 totally believable
20:16 totally believable
20:18 so you know for what it's worth for what
20:22 it's worth you know you know you put it
20:24 out there you stick an image up boom
20:26 makes a little movie not gonna win any
20:30 Oscars but you know
20:41 did you miss my singing I'm a singer
20:45 I'm not as good as singers I am a
20:48 hairdresser what the [ __ ] is happening
20:49 up there tonight
20:51 at least that's I'm sure you've missed
20:53 the hair
20:56 all right I think the sound effects
20:59 really helped I do thank you back again
21:03 Griswold
21:04 Mueller Bueller eleven Labs cost about
21:07 20 a month any free Alternatives no
21:10 that's what I was just saying the uh the
21:12 one in apple it's in you can if you
21:15 download the iOS 17 beta it's in there
21:17 if you've got a more modern phone than I
21:19 have
21:20 but I just have this janky ass crack
21:23 screen piece of [ __ ]
21:29 are you stuck for New York City uh how
21:32 much will AI be on the agenda well
21:36 a lot
21:37 so we're meeting with one of our Pharma
21:39 clients and we're talking about three
21:42 pilot projects they want to do that are
21:44 all AI related and then I'm potentially
21:48 knock on wood fingers crossed say little
21:52 prayers for me uh I think we're gonna
21:54 get a chance to meet with the CMO of a
21:57 major Pharma company and tell her about
22:00 what we do and what we're going to do
22:02 with AI and why she should be super
22:04 excited to work with us
22:07 and I'll go man
22:10 technology here you want to see it it's
22:12 pretty cool
22:14 she'll be like Ah that's awesome dude
22:15 but yeah
22:17 can I buy you lunch
22:19 she'll be like sure dude
22:21 pretty sure she's from California you
22:23 know got to learn to talk to your
22:24 audience
22:25 no she's not she's from New York
22:28 uh Skynet it's pretty pictures
22:32 that's true that was actually that's
22:35 that's a that was a good example of
22:37 skagnet painting pretty pictures I tried
22:39 to flip my car in the air fast and
22:41 furious like the only thing it did I had
22:44 it it added some smoke oh yeah
22:46 yeah so
22:49 so like the images the video stuff like
22:52 like video stuff's way more complicated
22:54 because you know what out of whatever
22:56 they call it continuity or
22:58 or it's not it's not continuity or
23:02 consistency it's
23:04 coherence
23:05 whatever
23:07 it doing [ __ ] that looks good
23:10 um it's really hard with images right
23:12 you see hands that look like this those
23:14 have gotten better now most people have
23:16 five-ish fingers and you know sometimes
23:19 you'll get hands that are this small or
23:20 fingers that are like this long but
23:22 there's normally five of them now so
23:24 that's that's getting better but with
23:26 video it's you know
23:27 30 frames a second more complicated
23:31 um so I think it I think it'll get
23:33 better and and honestly the the with the
23:37 image stuff
23:39 my experience has been about one in ten
23:41 that I do are good the cloud one
23:43 probably because there's there's so much
23:46 reference video of slow motion or time
23:50 lapse clouds panning across the sky like
23:52 that one probably has a pretty clear
23:53 reference that it could find
23:57 um and some of the other stuff just just
24:00 there's probably not enough in the
24:01 training set or something it'll get
24:03 better
24:04 uh what time is it it's 905 here not a
24:08 time check and I'm timing to see how far
24:09 behind you are it is 10 17 right now
24:14 so I am uh sorry math 12 minutes behind
24:18 uh but I assume you're an hour
24:21 I think I assume you're West Coast
24:23 Thelonious so 15 minutes or 12 minutes
24:26 behind and and if you're new here
24:28 welcome my name is Kyle Shannon we talk
24:30 about AI stuff here and answer questions
24:32 and we have fun we have fun we fun
24:36 sweaty balls you know
24:38 um
24:39 and uh
24:42 I try to read all the comments so if you
24:44 have questions or thoughts about AI pop
24:46 them in the comments I try to read them
24:48 all I do get behind uh people that are
24:51 the regulars here they will defend me
24:53 they will tell you I do answer my
24:54 questions if you come in here all truly
24:57 like like yeah I don't even think he's
24:59 live He's not what's he reading those
25:02 aren't questions that I wrote why isn't
25:04 answering my questions I've asked them
25:06 17 times
25:07 if you're one of those people man is
25:10 this gonna be frustrating oh it's gonna
25:12 suck
25:14 oh
25:16 do I have to sit here and watch the
25:18 whole thing to wait for him to answer my
25:19 stupid [ __ ] question yes
25:21 yes you do
25:23 I've been off for seven days so I'm a
25:25 little wound up
25:27 all right
25:30 what's the process of writing a scripted
25:32 wait of AI writing scripted or doing a
25:36 video
25:37 is that anime person
25:40 doing lip sync
25:43 I don't quite know what the question is
25:48 well so
25:50 I'll go I'll go show you how we can make
25:53 one of those it's pretty fun
25:56 um it ain't exactly Hollywood
25:59 quality yet
26:01 it's a little all of these Technologies
26:04 right now are a little as we call in the
26:07 business janky
26:08 so let me show you how we do some
26:10 scripted to animated voice stuff
26:15 all right
26:16 so let's see
26:18 go over here
26:22 and where will we go first you know
26:23 where we'll go first we'll go to the
26:25 classic
26:28 we're going to chat GPT you can make
26:31 money with chat gbt
26:33 all right so here in chat GPT and I'll
26:36 do GPT 3.5 that's the free one so this
26:39 just this is just your vanilla flavored
26:41 chat GPT and I'm gonna say
26:43 um
26:45 foreign
26:48 how
26:52 to write good
26:55 dialogue for a funny
26:58 Holly
27:01 would movie
27:04 and you're like but Kyle why would you
27:06 ask Chachi PT if it's good at that
27:09 because one of the things that happens
27:11 with good old Chachi BT is the more
27:13 context you give it the better it gets
27:15 so if you just say write me dialogue
27:17 it'll write you shitty dialogue if you
27:20 ask it something like you know what's
27:21 good dialogue for movies or do you
27:24 understand that like it said
27:26 what does it say here
27:28 um understand your characters timing is
27:30 everything use word play and puns use
27:34 exaggerated give me a bunch of examples
27:36 so I'll say
27:37 okay great
27:39 and here I'm gonna do this with voice
27:41 instead of typing because it's annoying
27:43 to all of you I know
27:45 hmm
27:46 [Music]
27:49 write me a funny opening line
27:54 of a simpleton who's memorized all of
27:57 the books in all of the libraries of all
28:00 of the world
28:05 in a world where wait in a world where
28:08 brains were bites and wisdom was Wi-Fi
28:10 I'm the all-knowing simpleton who
28:12 downloads the entire universe into my
28:14 Noggin but uh where do I plug in the
28:17 toaster
28:18 that's actually not bad
28:20 that's not bad all right there's our
28:23 line
28:24 pretty funny
28:26 now it's not going to be delivered with
28:27 comedy because it's going to be a
28:28 [ __ ] robot saying it but okay
28:32 so you go you if if you can write the
28:36 words you can write your own words but
28:37 we wrote those words and now there's two
28:40 things I'll show you one's 11 labs
28:44 and at 11 Labs I have
28:48 uh trained it to
28:52 be able to talk like me
28:54 sort of you know in in whatever the
28:57 state of art talking like me sounds like
29:00 and let's see Rachel KS I'll do chaos
29:04 chaos bright so I recorded I recorded
29:07 two different versions of me I recorded
29:09 one where I'm talking like a serious
29:11 person like this kind of like a
29:12 newscaster and then I I recorded one
29:14 where I was I was like I'm in a Pixar
29:16 movie isn't this exciting so we'll use
29:18 that one we'll use the peppy one because
29:20 it's comedy
29:21 and then I'm gonna pop that text in
29:24 there that I just wrote and we'll say
29:26 generate
29:30 in a world where brains were bites and
29:33 wisdom was Wi-Fi I'm the all-knowing
29:36 simpleton who's downloaded the entire
29:38 universe into my Noggin but uh where do
29:41 I plug in the toaster
29:43 not bad not bad so that's that so that's
29:45 eleven labs.io
29:47 is the website that one does cost you
29:50 money if you want to clone your own
29:51 voice but
29:53 you can also go to
29:55 dash ID and there's there's many of
29:59 these there's Synthesia there's a bunch
30:00 of these these do cost money I think
30:03 these things are ridiculously overpriced
30:06 but They Don't Really require much
30:10 knowledge
30:12 so yeah
30:15 no credit login sign up here we go
30:26 personalize your experience okay so
30:29 here's a bunch of videos that I've made
30:31 so like this is a
30:34 a South Asian woman that I made for a
30:36 work project the streets of Mumbai are
30:39 adorned with vibrant colors bustling
30:41 with excitement so the way it works if
30:43 you say I want to create a video you can
30:46 use one of their avatars or you can use
30:50 um one of yours so
30:52 this dude is like a Pixar looking dude
30:55 and this I I sort of did this dude to
30:59 look yeah to look like that bright voice
31:02 that I just recorded
31:04 so in this case I'll just pop it over
31:07 here oh I can upload oh you know what I
31:10 can do I can actually upload my voice
31:12 from 11 Labs so let's grab this
31:16 I'm going to download this
31:20 and then we'll go over to did
31:24 and then red so so what you can do here
31:26 is you can pick from
31:30 um all sorts of different uh languages
31:34 and
31:36 um accents
31:37 so so in English let's see where's
31:40 English
31:41 in English you know you can do
31:43 Netherlands Australia Canada Hong Kong
31:46 India so that that that South Asian one
31:49 I just did was English comma India or
31:52 you know parentheses
31:53 um so so you can just pick a voice and
31:55 and generate it based on that or you can
31:57 upload your own audio so I'm gonna
32:00 upload the thing that I just created in
32:03 11 labs
32:09 and then I'm gonna say generate video so
32:12 it's going to take that thing I made in
32:13 11 labs and it's going to put it on that
32:15 voice and it's going to animate that
32:16 picture to look
32:18 pretty darn swell
32:23 and so it's off doing that here's
32:25 another dude I did
32:28 hey there it's Kyle Shannon with the AI
32:30 learning lab and today we're going to
32:32 transfer that one's bad here's one I did
32:34 I think with my voice hello Tick Tock
32:37 live no how are you that was not my
32:39 voice one of these has my voice in it
32:42 this one
32:47 And So It Begins
32:49 there was a time until very recently
32:52 where you had to live life to experience
32:54 it
32:55 that time that was pretty good
32:57 all right so here's the one we just did
32:59 I'll hit the little play button
33:02 in a world where brains were bites and
33:04 wisdom was Wi-Fi I'm the all-knowing
33:08 simpleton who's downloaded the entire
33:10 universe into my Noggin but uh where do
33:13 I plug in the toaster
33:14 [Laughter]
33:16 comedy comedy
33:22 so anyway that's how you do it
33:25 did that help did that make things worse
33:28 or better did it make things worse or
33:30 better I don't know he seemed to talk
33:32 for a long long time did he say anything
33:34 I don't know
33:36 Industrial Revolution 3.0 here we come
33:39 it's coming it is coming it's absolutely
33:42 on the way
33:43 it's absolutely on the way without the
33:45 industry all the industry will be
33:47 replaced with robots but anyway yes the
33:49 industrial Rebel it's the it's the
33:51 knowledge it's the it's the artif it's
33:53 the robot thanks for the what is that
33:54 it's a boom box is that a boom box in in
33:57 my day back in the olden time of days in
33:59 the in the 80s and the 90s we had a
34:01 Boombox you'd walk down the street with
34:03 a big thing on your shoulders speakers
34:05 blasting in your head
34:07 and you play annoying music and annoy
34:09 the people around you
34:13 now we just get on Tick Tock and talk
34:15 into our phone
34:17 I I can annoy you in many different ways
34:20 I don't need a Boombox I got a
34:23 teeny little funk all right
34:30 do you have an affinity
34:32 for any reptiles
34:34 ah
34:36 I think they're impressive I just saw a
34:39 tick tock there's like a snake sort of
34:41 dangling out of the ceiling and then the
34:43 entire [ __ ] ceiling collapses and it
34:46 was a massive like the tail was just
34:48 like this teeny part that is massive
34:51 like 40-foot snake
34:54 um
34:56 I like lizards I learned how to catch
34:58 lizards in the uh in the desert with a
35:01 little piece of fish in line with a with
35:02 a loop on it and you just dangle it in
35:05 front of them they kind of freeze like
35:06 that's interesting and then you just
35:08 slide it over their head and go a little
35:10 slip thing and you catch them
35:13 so desert lizards are pretty cool
35:17 they can theorize and ask scenarios
35:19 based on existing scientific research
35:21 it's amazing wait what was that for you
35:24 can theorize
35:27 I don't know what that was about who
35:29 said that
35:30 glocal Global girl local oh as in
35:34 globally local that's a that's a
35:37 marketing term from like the late 90s
35:41 glocal
35:43 that's really funny are you in marketing
35:46 you got to be in marketing
35:49 uh I don't know what you're talking
35:51 about though it sounds cool
35:53 um what you're in Denver too I build AI
35:56 That's cool you should come to okay so
35:58 if you're in Denver even if you're not
35:59 one of the other things that I do
36:03 is called the AI salon so if you go to
36:05 the salon.ai that website it's it's a
36:09 link tree and we have a Meetup we just
36:11 had our meeting last night
36:13 we meet every two weeks in person in
36:15 Denver and we do it we have an after
36:16 party afterwards so we go hang out and
36:19 then uh we bring on speakers we do this
36:22 thing called AI confessionals where
36:24 people just talk about here's something
36:26 I did that blew my mind
36:28 do something or here's
36:31 these videos and they sucked right
36:34 um and we also have a Discord server so
36:36 there's a the invite to the Discord
36:38 server is there as well and uh then we
36:41 have a LinkedIn group so it's a cool
36:43 group so yeah if you're in Denver
36:45 sign up for the salon
36:47 so long we're fancy
36:51 it's French [ __ ] was that that was a
36:53 Dave Chappelle thing wasn't it
36:56 um
36:57 that's silly it's amazingly
36:59 smart to drink water yeah it's like you
37:01 know all of our cells run on it
37:04 and when you don't drink enough of it
37:06 bad things happen so flush that [ __ ] out
37:12 and I don't drink enough on these lives
37:13 so by generally I'll do these for like
37:15 two hours my ass gets numb and my voice
37:17 just gets like oh I can't talk
37:21 people in the live are like drink some
37:23 water buddy time to go to bed
37:31 my fans to take care of
37:33 them
37:35 all right and if ultimately wait it will
37:38 ultimately be the end of human dominance
37:40 on Earth rabbits in the Headlights
37:44 um
37:45 maybe maybe not
37:48 maybe maybe not
37:50 you know it it might be the end of
37:52 humans
37:55 exploiting humans to the extent that
37:57 they do now to be prosperous
38:00 which would allow us to be
38:02 more empathetic and more self-expressed
38:05 and less [ __ ]
38:06 is a possibility
38:09 um but I suppose there's that
38:10 possibility
38:12 uh yeah
38:17 I think what what we all individually do
38:20 right now like in the next year is is
38:24 more important than I think any of us
38:26 realize
38:27 each individual you know if I think
38:30 about just one person that I've talked
38:32 to that has
38:35 you know was skeptical of this or afraid
38:37 of it or whatever and they kind of had
38:39 that Epiphany they had their their uh
38:42 Uh Kevin McAllister moment there
38:45 their home alone moment
38:48 um
38:49 and and now like like I just know so
38:52 many people who are like their lives are
38:54 transformed in a positive way
38:57 um or or in a
39:00 completely in flux kind of way right
39:02 like they're excited but they're
39:04 terrified because things are changing
39:06 but at least they're in it and they're
39:08 they're active and so I think the more
39:10 of us that get curious about it and get
39:15 exploratory with it
39:17 and if and not evangelize it's not the
39:19 right word just share what we're
39:21 learning good or bad
39:24 um I think it's going to have a dish
39:26 disproportionately positive impact on
39:29 the trajectory of of how we use these
39:32 tools as Society so
39:34 um I I don't at all believe that we are
39:37 at the mercy of the [ __ ] tools that
39:39 we built
39:41 yet
39:43 we still have a chance
39:45 we still have a fighting chance we gotta
39:48 fight for it we gotta fight for the
39:50 right to not party but more like uh
39:54 responsibly use artificial intelligence
39:56 yes yes yes yes yes yes yes yes yes yes
39:58 yes yes yes
40:04 okay
40:06 sorry Tick Tock jump into the bottom
40:08 again I'm scrolling up to figure out
40:09 where I was and the girls ding questions
40:27 I can screenshot as well okay you're
40:29 helping one another I like that you're
40:31 helping whoever it was that couldn't
40:32 screenshot troubleshoot think it was
40:35 local settings
40:37 all right AI learning lab insert caption
40:40 is half
40:41 off screen oh that's because look
40:45 it's just I just have a phone on a stick
40:48 there you go is that better sorry
40:54 oh he's helpful but he doesn't seem seem
40:57 bright he's
40:59 he's slow on the uptake
41:02 all right
41:03 that's Kylie's special like that thank
41:05 you that's incredible
41:09 um
41:10 all right Andrea's helping felonious
41:12 it's good it's good
41:14 Tech Support Tech Support is right
41:16 around the corner here on the AI
41:17 learning lab
41:22 all right Robert Rossi killing it in the
41:25 number one gifter badge Arena sent
41:27 Donuts thanks for the donuts what phone
41:30 do you have I have an iPhone 11 Pro
41:33 Max I don't know the sort of big one the
41:36 four inch one
41:40 is gonna
41:42 oh how is AI going to help non-profit
41:45 organizations holy crap
41:47 let me count the ways what are
41:50 non-profit organizations
41:53 perpetually faced with
41:57 not enough resources
42:01 because not only do they have to go beg
42:04 for money all the time
42:07 they don't pay well oh David gold I'm
42:10 getting many worlds thank you for the
42:12 worlds
42:13 fight a World on a String I I good Lord
42:17 thank you I don't I need to come up with
42:20 the what do I do for a world
42:24 it's the world it's the world
42:27 no I'm not gonna do it I promise I can't
42:30 I can't resist it but thank you so much
42:32 for all of that
42:34 goodness goodness
42:36 um
42:37 so so how AI is going to help non-profit
42:40 organizations is lots and lots of ways
42:43 it can it can make you thank you very
42:45 very much for that that does very nice
42:48 um
42:49 it can make you much more productive
42:53 personally right so it can help you
42:55 write emails it could help you write
42:57 personalized emails to donors it could
42:59 help you write personalized stories
43:01 about you know impact that you've had
43:04 much more quickly it can you know
43:06 generate lots of organic content
43:10 um it can help you write grants it can
43:12 help you read you know Grant requests or
43:15 whatever their Grant offers or whatever
43:18 or if they're not rfps rfps in in the
43:21 business World
43:23 um
43:24 it can help with program development it
43:26 can help with
43:28 um
43:29 you know creating training materials
43:31 like it it
43:34 it's in a lot in a lot of ways
43:37 you know I think about
43:39 AIS for non-profit is is it feels like
43:42 one of the greatest areas of impact
43:44 because they're so under resourced
43:47 um
43:48 and and as the as the open source large
43:52 language models get increasingly
43:54 optimized where you're going to be able
43:55 to install like a GPT 3.5 like
44:02 um large language model natively on your
44:04 phone without having to connect to the
44:06 internet think about that anywhere where
44:08 you've got third world countries or
44:10 rural
44:11 um you know thanks for the donut you
44:14 know rural uh communities that don't
44:16 have access to the internet you're going
44:18 to be able to have these tools there
44:20 locally and and so you'll be able to
44:22 build all sorts of amazing tools
44:25 um
44:27 the way I would think about it is this
44:30 start to get literate if you're not
44:32 literate the reason I have these on
44:33 screen for my entire lives here is like
44:36 these are things to just go play with go
44:38 start getting used to playing with them
44:39 like Pi dot AI is one that you can talk
44:42 to and it talks back to you so there
44:44 there might be things with accessibility
44:46 that could be really powerful with that
44:48 one it's very very conversational so it
44:50 might be good for kids in some way
44:53 um
44:53 and and just start getting used to these
44:56 things and and the way to approach them
44:57 is rather than think about how can I use
44:59 AI for non-profits think about your
45:02 non-profit
45:03 and say what's a problem that I'm trying
45:06 to solve and then start to figure out
45:08 how can I use these tools to solve that
45:11 problem and it it totally it changed my
45:14 relationship with these tools where if
45:16 if you just sit down with
45:20 um like chat GPT
45:23 you know it's this little bar
45:27 and it's like it's kind of like a Google
45:28 search it's like well what do I do with
45:30 it
45:31 and it because it does so much like that
45:34 is such a deep well to dive in if you
45:37 just sit in front of it and go what can
45:38 I do with this it's almost impossible
45:40 but if you don't start with the
45:42 technology start with your problem
45:44 start with you know what's one problem
45:47 that you've got within your non-profit
45:48 that if you can solve it would make the
45:51 biggest difference start with that one
45:53 and say okay
45:56 how like what do I need to solve this
45:58 and and just go start interacting with
46:00 these tools and and and start having a
46:02 conversation with these tools almost
46:04 like you just hired an intern
46:06 who's
46:08 um an idiot savant right very very
46:10 simple you've got to tell them
46:12 everything but they know everything it's
46:15 almost as if they've memorized the
46:17 entire internet
46:19 that's what it's kind of like working
46:21 with these things you need to give them
46:23 really good direction because they're
46:24 dumb as a stone
46:26 but they know everything and if you give
46:29 them good instructions they're your best
46:31 friend they're your best assistant so
46:32 start with a problem and try to solve it
46:34 that should help but I like staggeringly
46:37 powerful impact on non-profits I think
46:40 for ones that embrace it like if if I'm
46:43 in a non-profit and I'm not like leaning
46:45 heavy into this world idiot
46:49 do you know how to
46:53 wait
46:54 ubicate a property on Google Maps with
46:57 only a couple of picks of the property I
46:59 don't know what ubicate means
47:01 do you know how to lubricate is that a
47:04 typo
47:05 well hell now subtitles right across
47:08 your eyes on tick tock's Revenge
47:13 oh you should be able to turn off
47:15 subtitles too
47:18 there is not is-17 at least not on
47:20 Apple's website ah yeah iOS 17 right now
47:24 is in beta so you have to go
47:28 into your settings
47:30 go into your settings and search for
47:32 beta and I think you have to sign up for
47:34 beta access
47:36 so it's probably in like software
47:38 updates and then you turn on beta access
47:40 and if you turn that on iOS 17 is now a
47:43 public beta I just got
47:45 version two but I still don't have the
47:47 stupid personal voice thing so
47:52 I hope that helps I hope that helps it's
47:56 6 a.m here here in Majorca wow welcome
48:00 welcome Matthias I uh hope you're gonna
48:04 join a fine coffee or tea uh I don't
48:06 know what you what what what's the
48:08 beverage of choice at six in the morning
48:09 in Majorca 12 minutes not bad it's early
48:12 your mountain time I am Mountain Time
48:14 Kyle answers questions I do I do darn it
48:18 I don't always answer them well but I do
48:21 answer them
48:22 you well you can be guaranteed is that I
48:24 will answer your question
48:27 what's up for grabs is you know
48:32 all right
48:35 but what are his qualifications
48:38 sir do you have any qualifications hmm
48:41 rather well I was trained at the Academy
48:44 of the Arts in the fine thespian art of
48:47 acting
48:49 no no I meant to to be on an AI Channel
48:52 what are your qualifications to talk
48:54 about machine learning in deep
48:56 mathematics rather
48:59 seemed to be a bit caught off guard with
49:01 that one
49:05 a little a little hot in here hmm
49:09 grab my pizza
49:12 uh so I went to World coin oh Thelonious
49:15 that's right and I I gave away my Iris
49:17 data to the to to the orb and all I got
49:20 was this sticker well we're in America
49:22 so that I think they just didn't the SEC
49:24 just tell coinbase you can't trade any
49:26 crypto at all other than Bitcoin so yeah
49:29 being in America if you want to do
49:31 anything with crypto sucks but with what
49:33 Thelonious is talking about there
49:36 last week
49:37 Samuel Altman's new company which is
49:39 called World coin launched and what
49:42 worldcoin is as creepy as this sounds I
49:44 actually think it or something like it's
49:46 really important so what what they're
49:48 trying to do they've they've created
49:50 three components
49:52 they've created
49:54 um a a transaction service so like an
49:58 app like PayPal where you can you know
50:01 transact
50:03 um they've created a crypto coin called
50:05 World coin
50:06 and then they've created this thing
50:08 called World ID and World ID is you get
50:11 verified as a human so you go to one of
50:14 these centers that have these orbs and
50:16 they're like these Chrome balls do I
50:18 have anything yeah I did yeah they're
50:20 like this size
50:22 but they're but they're Chrome and they
50:24 got a hole in them and you go up to it
50:26 when you stick your eye on it it scans
50:28 your iris verifies that you're human
50:31 and then
50:32 um then you get a world ID and that's
50:34 blockchain verified
50:36 so what they're creating is a network of
50:39 verified humans and I think as we will
50:42 see what the 2024 elections are going to
50:46 be like but I have
50:47 a sneaking suspicion that all this AI
50:50 [ __ ] is going to make lots and lots of
50:52 misinformation it's going to make 2016
50:54 and 2020 look like [ __ ] child's play
50:57 um so I think I think after that there's
51:00 going to be a real demand for I want to
51:02 know that someone's a human I want to
51:03 know that that video that I'm watching
51:05 hasn't been modified I want to know that
51:08 that article that I'm reading you know
51:09 is the original and hasn't been modified
51:12 and so I think this is a step in that
51:14 direction is it creepy as [ __ ] yeah now
51:17 apparently they're not keeping the data
51:19 they're not keeping the biometric data I
51:21 think you can explicitly give them
51:23 permission to use it for other [ __ ]
51:25 which I don't know why you do that but
51:28 whatever we've already given away all of
51:30 our
51:31 [ __ ] to Google
51:33 um but the idea of of having a verified
51:35 network of actual human beings I think
51:37 is a really interesting idea and tying
51:40 it to cryptocurrencies
51:41 um fascinating
51:42 the long-term vision of world coin is to
51:46 have it be a platform for Universal
51:48 basic income funded by the prosperity
51:51 that AI generates
51:54 and if you listen to Sam Altman I mean
51:56 he's very clear he thinks that
51:58 this is going to have a massive
51:59 disruptive effect on employment and
52:03 there's going to be a need for there's
52:05 going to be an increase in prosperity
52:07 but a decrease in the number of people
52:09 needed to create it
52:11 and so if you can tax AI companies that
52:14 are benefiting from that Delta
52:16 and turn that into Universal basic
52:18 income then that could be kind of cool
52:20 so that's that's about anyway glad to
52:24 see you're still here and they didn't
52:25 zap you with a laser and Fry the brain
52:28 but you know
52:29 maybe they did you just don't know it
52:31 yet
52:35 does AI really work on resumes oh
52:38 what's your name Knight Naya Naya
52:43 it works incredibly well on resumes so
52:47 um so depending on what you want to do
52:49 you can copy and paste your LinkedIn
52:52 profile and paste it into chat GPT and
52:56 then copy and paste in a job description
52:59 and even copy and paste in an old
53:02 version of your resume and then say you
53:05 know update my resume with my current
53:06 work experience and
53:09 follow resume best practices and
53:13 um
53:14 you know Focus my experience you know to
53:17 apply for that specific job and it'll do
53:18 it and then you can just tell it to
53:20 write you a cover letter
53:21 for that job as well and it'll do that
53:23 it's amazing at that stuff
53:25 chat GPT like all of these large
53:27 language models are particularly good if
53:29 you feed them a document and tell them
53:31 to analyze it in in any way they're
53:34 really good at that like amazingly good
53:37 at that
53:39 yes keep watching you'll learn well well
53:42 you wait to have your question answered
53:44 that is kind of the sad reality of this
53:48 channel it's like I just want to know
53:51 about Lang chain why doesn't he talk
53:53 about land chain I want to know that
53:57 Lang chain is he ever going to get to my
53:59 oh that was interesting oh I didn't know
54:02 I could do that yeah that's kind of what
54:04 this channel is all about is frustrating
54:05 the [ __ ] out of everyone here
54:07 I hope I I hope I'm succeeding
54:12 all right
54:13 that dude is you that dude is you did
54:18 you know that they used AI to synthesize
54:20 Val Kilmer's voice and Top Gun oh he did
54:23 I heard that yeah
54:25 and did the did the new Beatles song
54:27 came out yet came out yet what did the
54:31 did the English language get good yet
54:34 let's see new
54:37 Beatles song
54:39 I thought it came out end of July
54:52 Beatles release final song when are they
54:55 gonna release it
55:03 maybe it's August
55:17 well that was a useless article
55:21 tell me when it's coming out
55:24 when is it coming out I don't know when
55:27 it's coming out
55:28 but they um
55:30 this was something so this was like a
55:32 cassette recording that that Lennon
55:34 recorded and it was so noisy and bad
55:36 they've tried to release this thing like
55:37 two or three different times and they
55:39 failed every time because the technology
55:41 just sucked and then all of a sudden
55:43 with this generative AI stuff they
55:45 isolated his vocals got it clean enough
55:47 to put in a track and recorded the
55:49 finals Beatles song so I don't know if
55:51 it's out yet but it's coming
55:54 oh wait uh two release when NPR you'll
55:58 have a date please tell me you've done
56:00 some
56:00 reporting here
56:08 when is it coming I don't know
56:15 yeah but
56:17 yeah all that all of that synthesis
56:20 stuff I mean this is one of the things
56:21 if if you're following the SAG and and
56:24 wga strike
56:27 strikes
56:28 um there's a lot of a lot of what
56:31 they're talking about is you know actors
56:33 are going to be synthesized and and I
56:35 think in the future there was an
56:37 interesting piece in The Hollywood
56:38 Reporter By Justine Bateman
56:41 um Jason Bateman's sister and she's
56:43 she's a brainiac she's really smart but
56:46 she was talking about you know she said
56:47 to the SAG folks this is your last
56:50 strike where you have any leverage
56:52 because you know in the future you're
56:55 going to be competing against
56:57 young digitized Val Kilmer and you know
57:00 perpetually young Tom Cruise
57:03 and they'll just be able to pop people
57:04 into whatever movie and then you know
57:07 the future after that is that's all
57:09 going to happen dynamically I want to
57:11 watch Top Gun with Jim Gaffigan in the
57:13 lead okay there you go yeah
57:19 so it's just crazy I just ate my cat all
57:23 right hope it was tender
57:26 um what site was that to animate animate
57:28 faces it was called
57:29 d-i-d.com as in like digital ID I think
57:33 Dash id.com
57:41 do you work on AI well I don't know
57:44 maybe do I
57:47 I spent
57:51 two hours yesterday no three hours
57:54 yesterday writing two prompts
57:56 just trying to get them right
57:59 and then
58:01 I did which is cool it's really
58:03 gratifying when you get the prompts
58:04 right what's what's funny about
58:06 prompting with these things is that you
58:08 can get to
58:10 good
58:11 really really quickly
58:13 and then to get it to great or or like
58:16 consistently good every time that that
58:19 takes a lot longer
58:21 so I did that
58:23 um today I was trying to pull some data
58:26 from our database and I had to
58:29 pull some crap out of postgres on Heroku
58:33 using sequel
58:35 queries I don't know SQL queries but
58:38 there was one in there and it needed to
58:40 be updated so I just copied and pasted
58:42 that crap right into chat GPT and I said
58:45 update that sequel It's query like tell
58:47 me tell me what needs to be updated in
58:49 this from a
58:51 time and date perspective and it did
58:53 that
58:54 and then at one point it was giving me
58:56 inaccurate data and I said I pasted it
58:59 back in and I said why might be this be
59:01 giving me inaccurate data I'm expecting
59:03 this and I'm getting this and it said oh
59:06 well it looks like there's a discrepancy
59:07 between your dates and I said well can
59:09 you analyze this and is there is there a
59:12 way that you could fix it so that the
59:14 dates are better
59:15 and it said well in the current SQL
59:19 query you've got all your dates
59:21 hard-coded and we can change those to be
59:23 dynamically generated and I said yeah
59:25 let's go ahead and do that so it
59:26 completely rewrote the query and
59:28 inserted Dynamic date generation and it
59:31 fixed the data problem
59:33 now it took me like an hour to do that
59:35 but
59:37 the alternative would have been because
59:39 I don't know SQL and well I could have
59:42 figured it out it was a really long
59:44 query I didn't feel like figuring it out
59:45 I didn't have the time to like learn SQL
59:48 and learn this query
59:51 I would have just given that to my tech
59:53 team and they're in Prague so they would
59:56 have charged us you know the 80 or 100
59:57 or whatever the [ __ ] they charge us they
59:59 charge us a lot of money and it probably
1:00:01 would have taken him two hours to like
1:00:03 open it up figure it out analyze it all
1:00:04 that sort of stuff so in an hour I did
1:00:07 it myself fixed the query got the thing
1:00:09 working
1:00:11 um and solved the problem so so that's
1:00:12 the kind of thing these things can do
1:00:14 where like that's one of those
1:00:18 just sort of things in business that you
1:00:20 accept that if you come against
1:00:21 something that's a little too
1:00:22 complicated for you for your skill level
1:00:25 you're like oh just hand it off to an
1:00:27 expert well with this stuff you've got
1:00:31 kind of an expert in your pocket and if
1:00:33 you're willing to sort of go through the
1:00:35 pain in the ass of learning how to talk
1:00:37 to these things they can fill in where
1:00:40 you might otherwise have to hire people
1:00:42 so so it's the kind of thing that allows
1:00:44 a single person like a single proprietor
1:00:47 or a two or three person startup to be
1:00:51 able to act like a five to ten person
1:00:53 startup or 20 person startup if they're
1:00:55 really clever with it these are ability
1:00:58 amplifiers like dramatic ability
1:01:01 amplifiers it's not just words it's
1:01:03 amazing
1:01:04 uh so yeah and then I build content
1:01:07 generators and I'm re-architecting my
1:01:09 entire product vision of my company my
1:01:10 company's called storyvine
1:01:12 and it's a automated video production
1:01:15 platform we've been around since 2012.
1:01:18 so we're an 11 and a half year old
1:01:20 company and have not used AI for the
1:01:23 automation of our video editing we found
1:01:26 another way to do it
1:01:29 um
1:01:30 but I'm completely rethinking what the
1:01:32 product is and what the value is and how
1:01:34 we sell it and how we deliver it and so
1:01:36 yeah so I work on it a lot
1:01:38 but I am not a machine learning guy so
1:01:43 I I love
1:01:45 jumping in early on technology at the
1:01:48 moment at which it switches from only
1:01:50 the Geeks can play with it to the rest
1:01:52 of us get to play with it and for me
1:01:55 that moment was November 30th 2022. chat
1:01:58 gbt absolutely represents this single
1:02:02 you know Hatchet moment in history where
1:02:05 before that machine learning was for the
1:02:07 researchers and the scientists and the
1:02:08 mathematicians and after that it's the
1:02:11 rest of us get to play with it very
1:02:12 similar to the World Wide Web how it it
1:02:16 sort of landed on top of the internet
1:02:19 right the internet previously was
1:02:21 command line and it was researchers and
1:02:23 Librarians and all that kind of crap
1:02:25 scientists and mathematicians and then
1:02:28 the World Wide Web comes along and all
1:02:29 of a sudden we can make magazines and we
1:02:32 can make websites and we can make blogs
1:02:34 and everyone can be self-expressed and
1:02:35 now we can build social media now
1:02:37 someone can invent the iPhone and all of
1:02:40 that leads up to this [ __ ] where
1:02:42 everything's gonna change
1:02:45 all right I'm having trouble get AI to
1:02:47 generate graphics with text yes that's a
1:02:50 problem that will be fixed but you just
1:02:52 kind of have to deal with it right now
1:02:53 so you have to do it
1:02:55 dust off your Photoshop skills do your
1:02:58 text in Photoshop or go play with if
1:03:00 you've got access to photoshop go play
1:03:02 with Adobe Firefly and they've got a
1:03:04 text generating if you want to do
1:03:06 headlines and [ __ ] like that they have a
1:03:07 cool text generating thing where you can
1:03:10 do Standalone text with with patterns
1:03:12 and things in it
1:03:14 um
1:03:16 but yeah the
1:03:18 it wasn't sdxl
1:03:20 it was either a new hang on
1:03:25 there's a new version of stable
1:03:26 diffusion or it was a new version of
1:03:30 no no no you know
1:03:35 I think I think I've lost since I don't
1:03:39 have a track sense of time anymore since
1:03:41 I haven't been doing my lives the days
1:03:42 have all Blended together
1:03:45 um
1:03:45 sometime last week I think this is a
1:03:48 there's a new model coming out from open
1:03:50 AI thank you for the rose I appreciate
1:03:52 that there's a new model coming up from
1:03:54 open AI
1:03:56 um
1:03:57 they're the ones that have Dolly too
1:03:59 which is the image tool sitting inside
1:04:01 Bing
1:04:03 um it sucks it's it's bad like if you
1:04:05 compare it to mid-journey and stable
1:04:07 diffusion it's just not on the same
1:04:08 level but there's apparently a model
1:04:11 coming out from them that does text
1:04:13 really well
1:04:14 so text is coming so
1:04:16 what's your name Johnny's 24 honey okay
1:04:19 9.5 out of 10 on the name love the name
1:04:23 thumbs up
1:04:24 if if you're gonna be in here you should
1:04:26 you should work on your name game
1:04:28 because we got good ones in here I know
1:04:30 AI learning lab really straightforward
1:04:32 kind of boring you know I would give AI
1:04:35 learning lab a 6.8 out of 10. not all
1:04:39 that creative it's clear but not all
1:04:41 that's out thank you Johnny
1:04:43 Johnny's 24. honey bringing in the
1:04:46 Boombox and the world appreciate it
1:04:50 Saturday Saturday come to the buck
1:04:53 tractor pulls
1:04:55 anyone else from redneck Pennsylvania
1:04:58 I grew up in York PA
1:05:03 I've been to tractor pulls what's a
1:05:05 tractor pull you ask go look it up on
1:05:08 YouTube
1:05:10 Johnny's 24 honey
1:05:13 coming in with the lightning and the
1:05:15 whatever the hell else that was unicorn
1:05:17 Saturday Saturday Saturday
1:05:21 all right I found a site with lots of
1:05:23 famous voices you can type a phrase and
1:05:24 download it oh that's cool felonies
1:05:26 remember what it's called
1:05:27 I got you on London wishing my brother a
1:05:29 happy birthday super cool
1:05:32 yeah you know what's that what's that uh
1:05:35 oh Christ what's that site that um
1:05:39 famous people
1:05:40 you can pay them to like do a recorded
1:05:43 message for your loved ones that that
1:05:45 site's over right you're gonna be able
1:05:47 to just do that
1:05:49 Vinnie barbarino hair oh yeah
1:05:53 I promises
1:05:56 what was his uh tagline
1:06:00 I remember the fonzes
1:06:02 what was Vinnie barbarino's ah
1:06:06 welcome back
1:06:08 your dreams are your ticket out welcome
1:06:12 back
1:06:14 welcome back Connor that was a great
1:06:16 show damn was that a good show thank you
1:06:18 for the memory that was really cool it's
1:06:21 free of course and no sign up wait what
1:06:23 was the site Thelonious oh
1:06:25 101sounds.com see this yeah I see how to
1:06:28 see it hang on hold on hold your pants
1:06:31 dancing on this we're doing the best we
1:06:33 can
1:06:34 good Lord man
1:06:37 one
1:06:38 101
1:06:43 soundboards.com
1:06:50 uh who's a good one uh
1:06:53 uh Samuel
1:06:58 Jackson
1:07:00 Samuel Jackson play what again say what
1:07:04 again I dare you I double get you
1:07:06 [ __ ] say what one more goddamn
1:07:08 challenge
1:07:10 Black on
1:07:12 just take one again Samuel Jackson isn't
1:07:15 black
1:07:16 [Laughter]
1:07:18 oh it's comedy it's comedy where's the
1:07:23 one how do you make him say it where's
1:07:25 the thing
1:07:27 help me Thelonious I don't I don't I
1:07:30 don't know where to find the thing
1:07:31 where's the thing to make him say
1:07:33 something
1:07:35 Will Ferrell sounds Jimmy Hopkins sounds
1:07:39 Christina ritzy Pulp Fiction
1:07:43 oh TTS computer voice
1:07:50 oh enter your text to be spoken
1:07:54 um paste that was a thing we did before
1:07:57 generate it
1:08:04 we're making your sound
1:08:06 your phrase is being generated text to
1:08:08 speech technology the process of adding
1:08:10 into the sound board usually takes a few
1:08:12 seconds to a few minutes okay
1:08:16 when are they gonna tell me it's done
1:08:22 I'm whiny I don't know how to use this
1:08:25 technology it makes me sad
1:08:28 anyone else get like this where you like
1:08:30 you go to something new and if you can't
1:08:31 figure it out within 30 seconds you're
1:08:33 like yeah you hurt
1:08:35 I don't like being incompetent I think
1:08:38 it's an 80d thing
1:08:41 patience is not my virtue
1:08:45 uh
1:08:48 Jenner okay generous oh tssq
1:08:53 the waiting list of phrases that are in
1:08:55 the queue to be generated is one
1:09:01 um
1:09:03 in a world where brains were bites and
1:09:05 wisdom was Wi-Fi I'm the all-knowing
1:09:07 simply don't lose downloaded the entire
1:09:09 universe into my obnoxious of where do I
1:09:12 close oh yeah that was it that was the
1:09:14 one I just did
1:09:16 and they chose me yeah
1:09:19 and they chose all right so that was 101
1:09:23 soundboards.com
1:09:27 not great but again with with all of
1:09:29 these AI tools so
1:09:31 as you play with these some of your
1:09:34 response might be that's it is it me or
1:09:37 does that suck
1:09:38 yes
1:09:40 a lot of this stuff sucks right now but
1:09:46 how it is right now is as bad as it's
1:09:48 ever going to be so the fact that we can
1:09:50 do this at all is insane
1:09:54 and so it's all going to get better and
1:09:56 better and better so the investment you
1:09:58 make now to just get up to speed on what
1:10:00 this stuff is how it works and how all
1:10:01 of this stuff is related how the tools
1:10:03 that we use to do language stuff are the
1:10:06 same tools that are behind image
1:10:08 generation and audio generation and
1:10:09 music generation there was a site
1:10:13 um a buddy sent me to today
1:10:16 hang on let me
1:10:19 uh is it slides
1:10:43 this is a workshop I'm doing for a big
1:10:46 company with the dude named Kenny
1:10:48 Friedman who's awesome
1:10:52 and it's called suno.ai okay
1:10:56 so check out suno suno dot a i
1:11:03 uh
1:11:04 oh try the alpha on Discord that's right
1:11:06 this is on Discord so this is like mid
1:11:08 Journey you have to do it in Discord
1:11:11 or as soon oh there it is
1:11:15 um
1:11:17 so go to suno.ai
1:11:21 and then you go into these chirp
1:11:23 channels
1:11:26 and you do
1:11:29 come on I'll get to the bottom
1:11:32 you do slash chirp
1:11:35 is it slash chirp yeah slash chirp
1:11:39 and then we're gonna do we'll do like uh
1:11:43 uh what was my thing well uh we'll do
1:11:45 we'll do bouncy
1:11:49 um pop
1:11:52 and then you can put in your lyrics so
1:11:54 I'm just gonna put in those lyrics that
1:11:55 we generated before in a world where
1:11:57 brains were bites and wisdom was Wi-Fi
1:11:59 I'm the all-knowing simpleton that one
1:12:01 and so now I hit submit and now that's
1:12:04 gonna just like mid Journey will make
1:12:06 you an image this is going to make us a
1:12:08 song where it actually sings those
1:12:09 lyrics
1:12:11 on the style that that we created I like
1:12:14 I just found out about this today so
1:12:16 I don't know how to do this well and I
1:12:18 might not be doing it right like I I'm
1:12:21 not sure if you type when you swipe type
1:12:23 slash chirp if you type the thing after
1:12:26 it I don't know if it's actually paying
1:12:27 attention to that but I think it is
1:12:29 but we'll
1:12:31 it's generating I guess we'll come back
1:12:33 to that
1:12:34 what else can I show you here while it's
1:12:36 doing that I'll go show you so mid
1:12:39 Journey
1:12:42 looks like my internet is slower than
1:12:44 crap
1:12:45 [Music]
1:12:46 um
1:12:47 if you haven't played with mid during oh
1:12:49 yeah there's it's slow
1:12:51 I got I got slowness going on if you
1:12:54 haven't played with mid-journey
1:12:56 um the way you do it you come into
1:12:57 Discord if you haven't played with
1:12:59 Discord before here's all your channels
1:13:00 once you're in a channel these are all
1:13:03 the different or these are servers so
1:13:05 those are your different communities
1:13:06 these are channels within a server and
1:13:09 then so I'm in the salon Discord right
1:13:11 now we have a mid-journey channel and so
1:13:14 you type slash imagine so just like I
1:13:16 typed in slash chirp for that
1:13:18 suno.ai type in slash imagine and then
1:13:22 you
1:13:24 um you know paste in or you type in a
1:13:26 prompt so
1:13:28 um
1:13:28 God rays
1:13:32 bursting through
1:13:35 a thunder head
1:13:38 Storm of the Century Bang
1:13:43 and now that'll go off and make that
1:13:45 image for me
1:13:47 but let's go let's see my song's done is
1:13:49 Our Song done
1:13:51 yes all right so you heard it here first
1:13:57 [Music]
1:14:01 downloaded the entire universe it's like
1:14:03 Noggin by the way I plug in the toaster
1:14:07 so that was crap but
1:14:12 in a horrible Angel fights and wisdom
1:14:14 with Wi-Fi I mean all knowing something
1:14:15 he's downloaded the entire universe into
1:14:17 like Noggin
1:14:19 where do I plug in the toaster actually
1:14:21 you know not bad
1:14:22 not bad
1:14:24 so this is all coming and and again
1:14:27 let's see do we have our images here oh
1:14:29 those are cool check those out
1:14:32 let's see uh
1:14:36 this is pretty groovy I like the one
1:14:38 with the lightning
1:14:40 although this one's the most god-ray
1:14:44 let's do we'll do upscale one
1:14:51 so there's that beautiful thing and I'm
1:14:53 like oh crap I wish I wouldn't have done
1:14:55 that as a square image I wish that had
1:14:57 done that as as uh
1:15:00 what's it called it
1:15:03 16x9
1:15:04 well you could do that now so I'm going
1:15:06 to do aspect ratio of
1:15:09 um 16x9
1:15:14 and all the other stuff's good
1:15:18 I'll do Zoom of 1.5 make it a little bit
1:15:21 zoomed out from that
1:15:25 thing and now it's gonna make me
1:15:30 four versions of this that are wider
1:15:32 it'll extend that image out
1:15:34 so crazy huh isn't that crazy crazy
1:15:39 all right then we're moving on up Tick
1:15:43 Tock just jump me to the bottom because
1:15:45 it's awful and it hates me it hurts my
1:15:49 it hurts me it hits me hated it internet
1:15:55 hopefully some people are falling asleep
1:15:57 to this because they said they do they I
1:16:00 can't fall asleep without you I'm like
1:16:03 that's that's my favorite sort of
1:16:04 compliment ever I think
1:16:06 [Laughter]
1:16:13 ah it's comedy we're doing comedy
1:16:17 tonight
1:16:18 okay I found out where it was do you
1:16:21 think anything about the scientific wait
1:16:23 do you know anything about the
1:16:25 scientific background regarding AI
1:16:27 optimization modeling Etc I are you just
1:16:31 talking about being a user I know a
1:16:33 little bit about it I know about how it
1:16:34 works I can explain how the large
1:16:37 language models do what they do I can
1:16:39 explain how diffusion models do what
1:16:41 they do
1:16:43 um I know marginally about how embedding
1:16:47 works and things like that I know about
1:16:49 Technologies like Lang chain but if you
1:16:50 if you're getting into like you know
1:16:53 weights in the mathematics of it I'm
1:16:55 useless
1:16:56 but if you have basic questions about
1:16:58 how this [ __ ] works yeah I'm pretty good
1:17:01 on that I think it's actually important
1:17:02 to know that stuff
1:17:04 um
1:17:05 so yeah
1:17:07 these are
1:17:09 kind of cool
1:17:12 I like the bottom right one all right
1:17:14 let's make the bottom right one so I'm
1:17:15 gonna upscale that one
1:17:18 oh you know what Let's do let's do it
1:17:21 so we got this beautiful image
1:17:24 right okay thank you
1:17:26 Johnny's 24.
1:17:29 beautiful right so let's take this image
1:17:31 and we're going to turn this into a
1:17:33 video
1:17:39 save image
1:17:41 all right I'm gonna go to downloads
1:17:44 that's good so now I'm gonna go sweet
1:17:47 it's very sweet of you
1:17:49 uh we're gonna go we're gonna go over
1:17:52 where are we going where are we going
1:17:55 we're making pretty pictures oh Runway
1:17:57 we're gonna go to Runway ml
1:18:01 .com and then we're gonna go to the
1:18:04 login
1:18:07 I think we're gonna log in with the
1:18:08 Google
1:18:09 I think that's how I did this one we're
1:18:12 gonna go to the Google
1:18:14 and then we're gonna go
1:18:17 Gen 2 text to video let's see
1:18:23 yeah okay so when you go to gm2
1:18:27 why is the screen so shiny is it because
1:18:29 of this thing maybe
1:18:32 it's dirty
1:18:34 good Lord what a mess
1:18:37 all right so you come to Runway uh
1:18:40 Runway ml
1:18:42 you can type in a prompt but you can
1:18:45 also upload an image
1:18:46 so I'm going to click on upload an image
1:18:49 and there's a bunch that I've done
1:18:51 before
1:18:52 I'm just like a bunch of my cloud photos
1:18:54 and [ __ ] like that and so I'm going to
1:18:57 say upload here
1:18:59 and I'm going to go to downloads and I'm
1:19:02 gonna go get my little God raise pitcher
1:19:06 and so
1:19:10 there's my God Ray's picture right there
1:19:13 and now I'm going to say generate and
1:19:14 we're going to turn what didn't exist
1:19:17 before which is now this beautiful
1:19:20 Sun bursting through the clouds moment
1:19:25 that was generated out of nothing
1:19:28 and now we're going to turn it into a
1:19:30 video
1:19:31 and we're going to see pretty clouds
1:19:33 with a pretty sunset wait what was that
1:19:36 try unlimited that's an upsell that's
1:19:39 what you call marketing folks
1:19:41 someone was sitting there going like
1:19:43 hell you know what while someone's
1:19:45 sitting there waiting for their little
1:19:47 Cloud picture to turn into a video why
1:19:49 don't we do a little pop-up that says
1:19:51 spend more money
1:19:53 yeah that's a good idea we'll get more
1:19:55 subscriptions that way
1:19:57 Let's see we do the math if we can get
1:19:59 another five dollars out of ten percent
1:20:01 of the people that do that we can have
1:20:03 as much as 25 30 every month
1:20:07 I like it I like let's do that let's
1:20:09 let's do that pop-up is that what they
1:20:11 call it a pop-up Pop Tart pop up
1:20:14 we do Pop Tart pop up
1:20:17 with the with the give us more money and
1:20:19 then they'll give us more money and then
1:20:21 maybe I can get that Toyota Super I was
1:20:23 looking at oh here's our video wait ah
1:20:26 okay let's see does it suck
1:20:31 oh that's really cool
1:20:35 the clouds are kind of parting and
1:20:37 disappearing
1:20:38 like this is it's very religious it's a
1:20:43 lot of a lot of uh
1:20:46 that's cool
1:20:48 they don't really look like clouds
1:20:50 moving they kind of look like
1:20:51 set pieces moving out of a play but
1:20:54 but anyway
1:20:55 pretty groovy huh
1:20:57 so that's Runway ml Gen 2 upload an
1:21:00 image so that the workflow there was go
1:21:03 to Mid Journey
1:21:05 make an image download it or expand it
1:21:08 however you want make it whatever the
1:21:09 hell you want it to be then come over to
1:21:12 runwayml.com go to the app go to Gen 2
1:21:15 and then instead of typing a prompt
1:21:17 upload that image and just start
1:21:19 experimenting with it and like that
1:21:21 one's weird it's not quite right so you
1:21:23 know I would probably dick around with
1:21:24 that one a bit and and get it to do
1:21:26 something
1:21:28 get it to do something cool but
1:21:30 you know
1:21:33 um
1:21:37 Troy Decker I'm only here for the
1:21:39 entertainment factor LOL
1:21:42 I don't know what's entertaining about
1:21:45 this Mr Decca
1:21:48 this isn't funny
1:21:50 this is artificial intelligence
1:21:54 and there's nothing intelligent about it
1:22:00 all right carrying on could you explain
1:22:02 how machine Learning Works
1:22:04 thanks greetings from Brett
1:22:06 Buenos Aires Argentina
1:22:09 Brenda Buenos Aires uh yes I can
1:22:15 um
1:22:19 yes let's go
1:22:22 ah there's the slide
1:22:39 okay
1:22:41 so
1:22:44 we'll talk about a couple of things here
1:22:46 um
1:22:52 so so
1:22:54 let's let's start with
1:22:57 everyone talks about this thing called
1:22:58 GPT chat GPT let's chat EBT or or as the
1:23:03 the people that are like you give me
1:23:04 money when I chat jbt those guys have
1:23:06 you ever noticed that those guys also
1:23:08 always switch around the letters they
1:23:09 call it jet chat GTP
1:23:12 I every time they say it I want to smash
1:23:14 my face into the phone I'm like the
1:23:17 letters are right there I know they
1:23:18 sound alike but like learn what they are
1:23:20 and they'll do it like 10 times in a
1:23:22 video Once is okay
1:23:24 but like on screen it says chat GPT and
1:23:26 they're going chat GTP no
1:23:29 anyway I'm not telling you how this [ __ ]
1:23:31 works but
1:23:35 one of those stupid three [ __ ]
1:23:36 letters mean okay they mean
1:23:39 G is for generative meaning it's
1:23:41 generating content so one of the
1:23:44 misconceptions about these large
1:23:46 language models is that They're copying
1:23:48 and pasting from other people's work
1:23:50 right now while the lar the core base
1:23:53 large language model was trained on most
1:23:55 of the internet
1:23:57 um it's not copying and painting pasting
1:23:59 it's actually generating
1:24:01 so that's the G
1:24:03 the p in GPT is pre-trained and and what
1:24:07 that means is that historically large
1:24:09 language models you kind of have to
1:24:11 bring bring your own data so you'd bring
1:24:12 your Radiology data and and you could do
1:24:15 these sort of small relatively small
1:24:17 data sets and you kind of have to supply
1:24:19 your own
1:24:21 um in 2017 Google invents this thing
1:24:24 called the Transformer which is the T in
1:24:27 GPT
1:24:29 um Transformer
1:24:31 generative pre-trained Transformer and
1:24:33 the Transformer allowed these machine
1:24:35 learning models to be trained on massive
1:24:38 data sets so open AI in particular said
1:24:41 hey why don't we if it's all about the
1:24:44 data and throwing as much data and and
1:24:47 high quality data as we can why don't we
1:24:49 go get as much of the public internet as
1:24:51 we can and ingest that into this machine
1:24:54 and we need lots of gpus to do that so
1:24:57 that's why they partnered with Microsoft
1:24:58 there's all sorts of you know that
1:25:00 that's why this is going on
1:25:02 so
1:25:03 it's a model that's been trained on all
1:25:06 of the internet using this technology
1:25:07 called the Transformer you'll you'll
1:25:09 hear them talk uh talk about llms
1:25:12 sometimes that just means large language
1:25:14 model so chat GPT is is built on top of
1:25:18 a large language model called gpt4
1:25:21 that's the underlying
1:25:23 um I think they're calling what are they
1:25:24 calling them now they're calling them
1:25:26 Frontier models so if you're doing the
1:25:28 sort of expensive base model creation
1:25:31 that's called Frontier models and then
1:25:34 chat CPT is an application layer sitting
1:25:36 on top of it
1:25:37 then you've got this thing called the
1:25:39 latent space and and this is this is
1:25:41 where we get a little bit technical but
1:25:43 this is where it also gets fascinating
1:25:45 as fun okay
1:25:46 so what you're looking at there
1:25:49 is a 2d representation of just like an x
1:25:52 y
1:25:54 plane with a bunch of
1:25:57 um what are called vectors so when you
1:25:59 when you
1:26:01 train these models you you basically go
1:26:04 pointed at a lot of documents and it
1:26:06 takes all the words of those documents
1:26:08 and it groups them in semantic clusters
1:26:10 right and so the black ones and the blue
1:26:12 ones and the purple ones are all
1:26:14 different semantic clusters and you
1:26:16 might have and this is dramatically
1:26:17 simplified right but you might have the
1:26:20 word dog and the the word dog can live
1:26:22 in one of those clusters as the the cute
1:26:25 little chubby animal that sleeps on your
1:26:26 carpet and the other one is you know
1:26:28 that guy was a dog and he didn't treat
1:26:30 his girl right right and so so you can
1:26:32 have the same word that's got multiple
1:26:34 meanings so you have these clusters this
1:26:36 is a very simplified 2D representation
1:26:38 these Vector spaces are like 500 or a
1:26:43 thousand Dimensions like these are these
1:26:45 are massively complicated mathematical
1:26:47 spaces
1:26:50 Okay so
1:26:52 that's that's sort of all the math you
1:26:54 need to know but but you'll you'll hear
1:26:56 things like the latent space and when
1:26:58 they talk about the latent space they're
1:27:00 talking about these mathematical
1:27:02 clusters of words that are semantically
1:27:05 grouped
1:27:06 now how the how the models work is based
1:27:10 on the The Prompt that you type in
1:27:13 it basically takes your prompt and turns
1:27:16 that into a mathematical formula and
1:27:19 says okay based on all of the things
1:27:21 that you've done what is the next most
1:27:23 probable word and so it kind of works
1:27:25 like this and then it does that over and
1:27:27 over and over again so if our prompt is
1:27:29 the quick brown fox jumps over the lazy
1:27:32 blank we know as humans if you know that
1:27:35 phrase from typing class
1:27:37 um that word is dog right so it it sort
1:27:41 of takes your prompt and it does this
1:27:43 mathematical thing and then effectively
1:27:45 what happens in the latent space is that
1:27:47 if it essentially deselects semantic
1:27:50 regions that are not of interest to it
1:27:52 but you'll see there's these light blue
1:27:54 dots here
1:27:56 um are are the are the ones that it
1:27:58 looks at and within those light blue
1:28:00 dots one of those dots is mathematically
1:28:02 the highest probability of being the
1:28:06 thing that comes next and if you look at
1:28:08 what's actually in the database it's a
1:28:11 long string of numbers it's wait let me
1:28:14 zoom in on it because it's just crazy
1:28:17 so so what these Vector databases are is
1:28:20 is there databases of just these long
1:28:22 numeric strings which is basically just
1:28:25 the point in the vector space that this
1:28:28 word is and that number is associated
1:28:30 with the word dog and so then it inserts
1:28:33 the word dog into your sentence it's the
1:28:35 it's the one that's got the highest
1:28:36 probability and so when it writes you a
1:28:39 full paragraph of stuff it's doing it
1:28:41 word by word by word with this
1:28:43 mathematical calculation
1:28:45 that's also why when they talk about
1:28:48 large language models hallucinate what
1:28:50 what they're basically mean is that
1:28:52 they're full of [ __ ] and they lie they
1:28:53 will confidently give you an answer even
1:28:55 if it's wrong
1:28:56 and
1:28:58 why that's happening is it's just saying
1:29:01 hey out of all of the words that I've
1:29:03 got in this you know out of the
1:29:04 trillions of words that I've got in this
1:29:07 thing
1:29:08 um this combination of them is sort of
1:29:11 the highest probability of what you
1:29:12 asked for but and it's just doing that
1:29:15 mathematically one word at a time it
1:29:17 actually has no comprehension at all of
1:29:20 what it just wrote
1:29:22 so sometimes you'll say hey you got that
1:29:24 wrong could you do it again it'll say oh
1:29:26 I'm sorry I got that wrong how about
1:29:27 this and it'll get it wrong again and
1:29:29 sometimes you'll do that four or five
1:29:32 times and you're like how can it
1:29:33 possibly be getting this wrong
1:29:35 especially with things like counting or
1:29:37 mathematics or like give me a list of 17
1:29:39 things it'll give you 14 and then it'll
1:29:42 give you 18 and then it'll give you 12
1:29:44 and you're like why can't you just give
1:29:46 me 17 because it doesn't know
1:29:48 so future versions may know but but this
1:29:52 is basically how they work and and how
1:29:54 the images are generated is is is
1:29:56 basically just a layer that sits on top
1:29:58 of this model
1:30:00 um it's it's fascinating how it works
1:30:02 like whoever figured this out deserves
1:30:04 all sorts of medals and prizes
1:30:07 um it's probably been lots of people
1:30:08 over the years but that's how they work
1:30:10 so
1:30:11 I hope it helps I hope it helps you know
1:30:14 I'm just out here trying to heal and and
1:30:17 sometimes it doesn't heal and sometimes
1:30:19 it helps sometimes it helps sometimes it
1:30:23 doesn't heal other times it heals
1:30:26 why does he talk like that I don't know
1:30:28 I don't know
1:30:30 he thinks it helps
1:30:33 please don't
1:30:35 is that about twerking
1:30:39 Lou Grant oh Lou Grant was good that was
1:30:41 a good show
1:30:42 Gen X is like I got a pile of gen xers
1:30:45 in here yeah that hair is something oh
1:30:48 geez it's 11 24. I got work tomorrow you
1:30:51 people
1:30:51 so demanding
1:30:55 all right hope that was a good
1:30:57 explanation of that
1:30:58 um how did you get AI on your blackberry
1:31:01 that's good that's solid that's a solid
1:31:04 burn could you please explain how
1:31:06 machine Learning Works in all caps
1:31:08 thanks greetings from Buenos Aires we
1:31:10 just did that but you're welcome nice
1:31:12 shirt too thank you yeah I like this
1:31:14 shirt
1:31:15 UB Cafe is Spanish slang command wait oh
1:31:19 oh Yuba Kate oh I see yubakate is a
1:31:22 Spanish slang command that means get
1:31:24 your bearings oh that's cool that's cool
1:31:26 felonious
1:31:27 touch and hold the subtitle and you can
1:31:30 move it on screen you're welcome
1:31:32 touch and hold the subtitle oh oh for
1:31:35 the person thank you Dave gold I'm
1:31:38 getting the worlds He Got the Whole
1:31:41 World in His Hand he got the whole wide
1:31:45 world in his hand
1:31:48 wow David gold
1:31:51 knocking the marbles out thank you very
1:31:54 much I appreciate that that's very cool
1:31:57 um
1:31:58 what do you think of adobe's photo AI
1:32:01 they are building
1:32:02 um I think it's remarkable so here's
1:32:05 what's happening with Adobe
1:32:07 um
1:32:08 as of January or February last year they
1:32:11 weren't even sure if they were going to
1:32:13 get into this space because they knew it
1:32:14 was kind of problematic and potentially
1:32:16 cannibalistic on their business and then
1:32:20 uh they said some version of [ __ ] it
1:32:22 let's try
1:32:25 and they did it really smart and when
1:32:26 you when you have a company like Adobe
1:32:28 that's got good resources and smart
1:32:30 talented people when they decide to move
1:32:32 they can move fast so they did a couple
1:32:34 of things really right they they started
1:32:36 out with this thing called Adobe Firefly
1:32:38 which if you've got a an adobe
1:32:41 subscription a cloud subscription you
1:32:44 can play with Adobe Firefly so that's
1:32:46 basically their playground and they've
1:32:47 got all these modules they got a text
1:32:49 module they've got a text to image
1:32:52 module and they've got what's the other
1:32:54 one uh I don't know they got a couple of
1:32:56 modules there and they're and they're
1:32:57 adding a bunch more
1:32:59 um
1:33:00 [Applause]
1:33:02 so so they built a little playground off
1:33:04 to the side and what what's smart about
1:33:06 that is
1:33:08 all those different modules they're
1:33:09 basically
1:33:10 testing those right so if people use
1:33:13 them and they get good feedback on them
1:33:15 those are going to get incorporated into
1:33:16 their tools if they suck they just sort
1:33:18 of flush them down the toilet right so
1:33:20 that it's smart on that front the next
1:33:22 smart thing they did was they trained
1:33:25 the the image models on public domain
1:33:28 and Adobe stock images that they owned
1:33:31 all the rights to
1:33:32 so none of their work has been trained
1:33:35 on commercially produced you know stock
1:33:39 photography or illustrations or you know
1:33:41 people's [ __ ]
1:33:43 they're going to be allowing artists to
1:33:45 opt out of having their images trained
1:33:48 on or opt into having their images
1:33:49 trained on because some artists is it
1:33:51 would benefit them to be in a model
1:33:53 where they can be discoverable right so
1:33:55 some artists are going to want to do
1:33:56 that at some point they're going to
1:33:58 figure out compensation so how they've
1:34:01 built their model is in a very artist
1:34:03 friendly way
1:34:05 it's also very business friendly because
1:34:07 mid-journey are a bunch of [ __ ]
1:34:09 Cowboys and they're like [ __ ] it we
1:34:11 trained it on everything go ahead and
1:34:12 sue us well everyone's going to go ahead
1:34:14 and sue them and they're probably
1:34:15 they're probably not going to survive
1:34:17 long term if they keep it up with that
1:34:19 attitude
1:34:21 um stable diffusion's a little craftier
1:34:24 because they're open sourcing all of
1:34:26 their stuff so they're like [ __ ] it we
1:34:27 trained on everything but here you can
1:34:29 have the models so if they get sued out
1:34:31 of existence at least the technology
1:34:32 lives on
1:34:34 um but adobe's making that kind of image
1:34:38 generation safe for corporations which
1:34:40 is brilliant
1:34:42 and then the final thing is as they've
1:34:44 rolled those things in to photoshop like
1:34:47 generative fill within Photoshop is just
1:34:49 [ __ ] amazing like how they
1:34:51 Incorporated it and how you call it and
1:34:54 how you do it if you haven't done if you
1:34:56 haven't seen it I'll show you how to do
1:34:57 it let's go play with it because it's
1:34:59 it's cool
1:35:01 oh yeah this is this is an image that I
1:35:04 did for uh for the uh for a workshop
1:35:07 that I'm working on
1:35:08 we just hang on
1:35:20 we had this Square image and we needed
1:35:22 it to be the whole width of a of a
1:35:25 PowerPoint slide so I just
1:35:28 did generative fill and expanded it
1:35:32 there's a new version of the Photoshop
1:35:34 beta available woohoo
1:35:37 all right so let's open something here
1:35:42 oh that's kind of cool yeah that's kind
1:35:43 of cool we'll do that we could do that
1:35:45 one what else do we have
1:35:53 that was uh if you haven't played with
1:35:55 this remedy thing you can do virtual
1:35:58 headshots
1:35:59 that's the that's the cheesy the cheesy
1:36:02 me in the in the jean jacket headshot
1:36:05 all right let's let's open this thing up
1:36:14 all right so there's there's our our
1:36:17 little photo
1:36:19 and let's say
1:36:22 we want it wider well I can just grab
1:36:27 this thing and kind of extend it out
1:36:29 like that
1:36:31 and then I can grab my Marquee tool
1:36:34 and I can just
1:36:36 highlight that white section overlap it
1:36:39 a little bit and then there's this
1:36:41 context sensitive little menu bar down
1:36:44 here and this button to the left says
1:36:46 generative fill and so I just click on
1:36:48 it I can put in a prompt so so like what
1:36:50 am I what I might put in here is oh
1:36:53 let's do this we'll do
1:36:55 um scary
1:36:57 green monster with scales
1:37:03 and sweat
1:37:05 I don't know
1:37:08 and so it's now going to extend this
1:37:11 image and if it's if it's a cool monster
1:37:14 then this would make a cool ad for that
1:37:17 muscle car that's not bad and it gives
1:37:20 you three options
1:37:23 that's pretty good
1:37:27 [Laughter]
1:37:29 I think that that one's probably a bit
1:37:31 more realistic but you know it sort of
1:37:33 continued the thing on so so you can
1:37:35 just keep doing that and you know I
1:37:36 could I could you know highlight the
1:37:39 eyes and say generative fill and say
1:37:43 glowing eyes
1:37:47 and so now it should
1:37:51 make those eyes a little more scary
1:37:58 here we come that's bad here we come
1:38:01 we'll save the date that's pretty good
1:38:02 there you go there's some glowing eyes
1:38:06 and so anyway so that's generated Phil
1:38:08 so I like it I I mean you know it's it's
1:38:12 the kind of thing that um
1:38:17 I I think they're doing it smart I think
1:38:18 I think adobe's doing it quite smart
1:38:20 right now
1:38:23 um
1:38:26 beverage without sugar
1:38:29 he's having water he likes the water
1:38:35 glad refreshing
1:38:37 thanks still unsure about screenshots uh
1:38:41 can it pretty much wait can do pretty
1:38:43 much
1:38:45 can do it pretty much all others oh oh
1:38:49 sorry to hijack you Kyle I don't know
1:38:51 you're fine you're good you want a
1:38:54 screenshot you got a screenshot you got
1:38:56 issues screenshotting you got to talk
1:38:57 about it
1:38:59 you gotta you gotta bring it up you got
1:39:01 issues get it well hey a bunch of people
1:39:03 just joined what are y'all doing it's
1:39:06 late
1:39:07 you damn insomniacs are you on the other
1:39:09 side of the world
1:39:10 anyway welcome if you're here my name is
1:39:12 Kyle Shannon this the AI learning lab we
1:39:14 talk about stuff here
1:39:16 we don't know what we talk about I'm
1:39:17 completely unqualified but that's okay
1:39:20 all right
1:39:21 uh it can also mean get real or get
1:39:24 serious to indicate someone is being
1:39:26 unrealistic oh that's cool
1:39:29 are you sure you're not psyops I am
1:39:31 absolutely not sure I'm not psyops I
1:39:33 might be psyops I might be gpt5 did you
1:39:36 see the gpt5 trademark uh application
1:39:39 anyone anyone
1:39:41 no
1:39:44 let's go to the Twitter let's go to the
1:39:46 Twitter you people got to keep up with
1:39:48 this stuff
1:39:49 come on
1:39:51 can't be expected to do all of this
1:40:02 GPD five
1:40:05 so open AI files a U.S trademark
1:40:10 here's what's included artificial
1:40:12 production of human speech and text
1:40:17 conversation of audio a conversion of
1:40:20 audio data files into text
1:40:23 voice and speech recognition
1:40:25 so already in gpt4 is image recognition
1:40:29 they haven't rolled it out yet but you
1:40:30 can use it on Bing
1:40:33 um and it does all its wordy wordy stuff
1:40:36 and now it's going to do audio audio
1:40:37 stuff and
1:40:40 video can't be too far behind so gpd5 is
1:40:43 likely going to be a pretty slick
1:40:45 multimodal
1:40:47 crazy ass
1:40:48 the the thing about all these tools
1:40:50 right now so I showed earlier that that
1:40:53 singing Generation tool and we're doing
1:40:56 mid-journey things here we're going to
1:40:58 Runway ml for there and all these things
1:41:01 right now are these disparate pieces of
1:41:04 crap little
1:41:06 standalone
1:41:08 and I don't mean to diminish the
1:41:10 incredibly
1:41:11 powerful pieces of crap that they are
1:41:14 but they're just they're they're all raw
1:41:17 right
1:41:18 and
1:41:22 within a year we're going to start to
1:41:24 see these things do this and these tools
1:41:27 are going to get so sophisticated and so
1:41:29 slick
1:41:31 so cool what's coming
1:41:35 they're cool they're cool that's a cool
1:41:38 cat that's really cool Kyle that's cool
1:41:41 I'm learning AI technology awesome ah
1:41:44 the bear with my little dangly lights I
1:41:46 love the dangly lights the Vanoss
1:41:49 Thelonious I'm still here I didn't put
1:41:51 my eyeball into the orb just looked at
1:41:53 it I don't know if I was lasered might
1:41:55 have been lasered
1:41:57 uh now and then as the Beatles song
1:42:00 release date 2023 when 2023 I know 2023
1:42:06 when is it is it out is it coming
1:42:10 [Laughter]
1:42:13 Hey guys does bars has an app
1:42:20 I don't know what you saying it won't
1:42:23 ever be good enough to replace actual
1:42:25 writers
1:42:26 um
1:42:28 are you sure about that
1:42:31 you may wanna look into that
1:42:36 um
1:42:40 I know a lot of writers that have
1:42:42 embraced this and one who's one of the
1:42:45 organizers of the AI salon
1:42:48 is a copywriter I met her at a creative
1:42:51 mornings
1:42:52 and I said what do you do and she said
1:42:54 I'm a copywriter and I said this is like
1:42:57 back in January I said if you played
1:42:59 with chat gbt she goes I heard people
1:43:01 talking about that she came to came to
1:43:03 the first salon and started playing with
1:43:05 it and like the second Salon she came to
1:43:07 she was like
1:43:10 and now she's like
1:43:12 you know she basically
1:43:14 uses the tool to do most of her writing
1:43:17 for her and and you know
1:43:20 to be fair
1:43:23 you know to your point you know is it
1:43:25 going to write the Great American novel
1:43:27 or a fantastic screenplay on its own no
1:43:30 but if you're churning out [ __ ]
1:43:33 social media video Scripts
1:43:36 it can absolutely do that and it can do
1:43:39 those things
1:43:40 [ __ ] 30 times faster than you then
1:43:44 you could get draft one you know you
1:43:46 could have 30 versions of those so so
1:43:48 for certain kinds of writing it's there
1:43:51 now
1:43:53 and then if you start to look at more
1:43:54 sophisticated kinds of writing like
1:43:56 screenwriting and novel writing and
1:43:59 things like that
1:44:01 um
1:44:02 go check out a project called GPT author
1:44:06 which you have to be a little geeky to
1:44:08 play with it you have to run it in
1:44:09 Google collab
1:44:11 but it's a series it's what it what it
1:44:14 effectively is is a series of prompts
1:44:15 strung together to write a complete
1:44:18 novel
1:44:19 so what you give it is here's the kind
1:44:21 of story that I want here's the writing
1:44:24 style that I want here's the number of
1:44:26 chapters that I want here's my open AI
1:44:29 key and my stable diffusion key go
1:44:34 and it writes an outline for a book with
1:44:36 the number of chapters that you're
1:44:38 talking about with the style that you're
1:44:40 asking for and then it writes one
1:44:43 chapter at a time
1:44:44 so it writes a chapter
1:44:46 then it takes the outline in chapter one
1:44:48 and it writes chapter two then it takes
1:44:50 the outline in chapter one and two and
1:44:51 it writes chapter three and so and
1:44:54 you're just watching this and it's just
1:44:55 going dude
1:44:58 and then it gets to the end and it's
1:45:00 written this whole book and then it
1:45:02 writes a prompt for stable diffusion to
1:45:05 create a cover for the book and then it
1:45:07 saves it as uh an ebook
1:45:11 and it does it in like I don't know
1:45:13 three minutes
1:45:16 now
1:45:17 is the book brilliant not yet
1:45:22 but go look at that like but is is it a
1:45:25 coherent
1:45:27 well-structured story in three minutes
1:45:30 that's a book yes it is
1:45:35 so again my contention with these tools
1:45:39 is they're the worst they're ever going
1:45:41 to be right now
1:45:44 so making a comment like they won't ever
1:45:47 be good enough to replace actual writers
1:45:49 they're already good enough to replace
1:45:51 actual writers
1:45:53 they're not good enough to replace all
1:45:55 the writers
1:45:56 but what's coming
1:46:00 is significantly Beyond I think what
1:46:02 you're giving
1:46:03 this technology credit for
1:46:09 and listen
1:46:11 if you're a writer and I'm a writer I've
1:46:13 written seven screenplays and a one-man
1:46:15 show and
1:46:16 all I do for my business is write you
1:46:18 know story structures so I get it
1:46:22 it's [ __ ] existentially terrifying
1:46:26 so
1:46:27 I get
1:46:29 why you want to be in that place
1:46:32 but I wouldn't be so confident there if
1:46:34 I were you I would I would dig deeper
1:46:36 and learn more
1:46:38 um just learn learn what these things
1:46:40 can do and in fact if you're a writer
1:46:43 and if you've got that position I would
1:46:44 go play with all of these the reason I
1:46:46 have all these up there is they're all
1:46:48 good at different things gpt4 up here in
1:46:51 my opinion is the best writer of all of
1:46:53 these but they've all got different
1:46:55 skills so depending on the kind of
1:46:57 writing you do there might be certain of
1:46:59 these tools that do things better so
1:47:00 write yourself an interesting prompt
1:47:03 as a test and go test GPT 3.5 gpt4 test
1:47:08 Claude test gpt4 at bing.com testbard
1:47:12 chat test this is meta this is the the
1:47:15 Llama 2 model test Pi which is super
1:47:18 conversational
1:47:20 like all of those
1:47:22 like that it would be a good education
1:47:24 for you just to see where these
1:47:25 different models are which ones are good
1:47:27 at what
1:47:31 um but I I would not sort of I I
1:47:33 wouldn't I would not sit with much
1:47:35 confidence in the statement that you
1:47:37 made
1:47:39 because the [ __ ] that's coming is gonna
1:47:41 melt some [ __ ] faces it's gonna be
1:47:43 like the faces just fall off
1:47:46 what happened to my face I don't know I
1:47:50 said that it couldn't write as good as a
1:47:51 writer right before I got laid off from
1:47:53 a writing job
1:47:54 [Laughter]
1:48:02 all right
1:48:03 hey buddy hello Stevo what's happening
1:48:06 what's happening happening machine
1:48:09 learning tight language hmm yeah you got
1:48:12 tight language that language is
1:48:16 I like turtles Chris that's what got me
1:48:19 in trouble in my last life I showed the
1:48:21 I like turtles thing and then that
1:48:23 reminded me of the other guy and I
1:48:25 showed that video and then I said what
1:48:27 the guy said in the video and then my
1:48:29 live web
1:48:31 so we're not doing Turtles joke anymore
1:48:33 but thank you good joke like it go look
1:48:37 up I like turtles on YouTube if you
1:48:39 don't know what we're talking about
1:48:40 exciting times
1:48:42 sdxl sdxl looks pretty cool I would love
1:48:45 a tutorial of mid Journey if you get a
1:48:47 chance I did that hopefully you were
1:48:49 here
1:48:50 I did kind of a kind of a minor one but
1:48:52 I did did the slash imagine let's
1:48:56 imagine what we could do here
1:48:58 let's turn it into a video shall we
1:49:02 I'm a mid-journey mod if you need help
1:49:05 oh
1:49:07 that's cool
1:49:09 uh
1:49:11 uh I don't need help right now but it's
1:49:13 cool that you're here
1:49:15 um
1:49:17 I here's something that you could answer
1:49:18 for me replicant which by the way 9.5
1:49:21 out of 10 on the name I like it
1:49:24 um
1:49:25 I think with the qualifier it could be
1:49:27 better what kind of replicant
1:49:30 you know
1:49:31 lazy replicant busted replicant I don't
1:49:36 know anyway
1:49:37 um
1:49:39 what would be fascinating for me to know
1:49:42 if if you're connected to mid-journey in
1:49:44 any way one is I heard there's only like
1:49:47 12 employees is that true that sounds
1:49:51 um ungodly small stuff to me for
1:49:54 something that powerful but if it is
1:49:56 [ __ ] cool uh and then my other
1:49:58 question is uh when are we gonna not
1:50:00 have to go and [ __ ] Discord to make
1:50:02 mid-journey images please tell me
1:50:04 there's a website coming please please
1:50:07 please please
1:50:10 or an app or something
1:50:13 anything Bueller
1:50:16 um but thanks for coming and thanks for
1:50:18 uh offering up your uh thoughts
1:50:21 oh replica grew up in Grand Cranberry
1:50:24 Township PA I grew up in York
1:50:25 Pennsylvania actually Wrightsville PA
1:50:28 just across the river from Lancaster or
1:50:32 as they call it not from Pennsylvania
1:50:33 Lancaster
1:50:35 Pennsylvania Dutch Country
1:50:38 hey you going down to the mall today
1:50:40 what are you hey what are you doing are
1:50:42 you you want to go fishing later
1:50:44 that's what I grew up with you go down
1:50:46 the mall today
1:50:48 Kyle why would you why would you uh why
1:50:50 would you go to college why would you
1:50:52 want to leave here
1:50:54 yeah that's what I grew up with
1:50:56 four
1:50:58 four people out of my graduating class
1:51:00 in high school left town
1:51:02 four
1:51:05 they were like why would you want to go
1:51:06 I was like why would you want to [ __ ]
1:51:08 stay
1:51:11 I don't know I just always I always had
1:51:14 the itch to be somewhere else then I got
1:51:16 to New York City and I was like ah
1:51:18 here's my people
1:51:20 here's home
1:51:21 now I'm living in [ __ ] Denver
1:51:24 I gotta get back to the East Coast
1:51:27 because them East Coasters they just got
1:51:28 a different Edge to them
1:51:30 they're so aggressive yeah exactly I
1:51:34 like it
1:51:35 [Laughter]
1:51:37 oh Cameo Cameo that's the one God I'm
1:51:40 I'm significantly behind on my questions
1:51:42 Cameo was like an hour ago when I was
1:51:45 asking about cameo
1:51:47 oh well
1:51:48 up your nose with a rubber hose that's a
1:51:51 good Gen X line
1:51:52 what are the methods that will be used
1:51:54 for AI regulation and how will alignment
1:51:56 become more efficient well
1:51:58 Tick Tock astronaut
1:52:03 I'm going 10 out of 10 on the name
1:52:06 Tick Tock astronaut I like it
1:52:10 um
1:52:14 I don't know what methods will be used
1:52:16 for AI regulation
1:52:18 I think that
1:52:22 I I like what's where Sam Altman's head
1:52:25 on this is feels right to me that if
1:52:28 you're building Frontier models which is
1:52:30 what they're calling you know like the
1:52:32 people that make gpt4 gpt5 Bard you know
1:52:36 so Google open AI Microsoft
1:52:39 if you've got the resources to make
1:52:42 Foundation models that should be
1:52:44 regulated
1:52:45 and then what he's saying is if you're
1:52:47 doing anything
1:52:49 the next level more powerful than that
1:52:51 like the AGI stuff the stuff that's like
1:52:55 as intelligent as any human
1:52:58 you know ever or as intelligent of all
1:53:00 the humans combined
1:53:02 if you're doing stuff at that level of
1:53:04 power
1:53:06 or capability then you need to um
1:53:10 get licensed like you get a license to
1:53:13 build a nuclear power plant
1:53:15 and then what he's saying is so so if
1:53:18 you're doing the super powerful [ __ ]
1:53:20 you've got to license it if you're doing
1:53:22 Frontier Model stuff like Foundation
1:53:24 model stuff that's got to be regulated
1:53:27 and then anything below that he's saying
1:53:28 don't regulate it which I think that
1:53:31 that sounds very right to me
1:53:33 um and then how will alignment become
1:53:35 more efficient so again back to open AI
1:53:37 I'm you know I'm a bit of a show for
1:53:39 them just because I I you know what
1:53:43 they're actually doing in the world like
1:53:45 a lot of people are talking right now
1:53:46 like Elon with his you know x dot AI is
1:53:50 going to understand the universe great
1:53:51 when are you gonna have this technology
1:53:54 um
1:53:55 openai they they just committed to
1:53:58 dedicate 20 of their Computing
1:54:01 um
1:54:02 power
1:54:03 to what they're calling super alignment
1:54:05 so so doing the alignment and alignment
1:54:08 with these tools basically mean
1:54:10 do the
1:54:12 imperatives of these tools align
1:54:15 with the imperative of these beings
1:54:19 that's alignment
1:54:22 if these tools go hey these beings are a
1:54:25 bit of a pain in the ass and they're
1:54:27 getting in the way of my imperatives why
1:54:29 don't I just get rid of them
1:54:32 that's misalignment that's bad that's
1:54:37 so super alignment says let's keep these
1:54:40 things going in the same direction
1:54:42 that this is about prosperity and making
1:54:46 these things better
1:54:48 and not competing with them because the
1:54:50 soon as soon as these things start
1:54:53 competing with these things autonomously
1:54:55 then we're in trouble
1:54:58 and that's that's the fear right
1:55:01 so
1:55:03 um
1:55:04 so that's going to help and then and
1:55:06 then there was a big Consortium of the
1:55:09 big players just got together in
1:55:10 Washington and agreed to
1:55:12 I don't know we're gonna do non-shitty
1:55:16 stuff together which
1:55:18 that in a bus pass will get you downtown
1:55:22 in the comments are easier to read
1:55:23 against the darker shirt I know that
1:55:26 I know that I wear dark shirts all the
1:55:29 time but I could actually read a little
1:55:30 bit against this shirt for some reason
1:55:32 but you know
1:55:36 [Music]
1:55:37 you should
1:55:40 astroguide.ai is this gonna put [ __ ]
1:55:43 malware on my computer I'll be cranky if
1:55:45 it does
1:55:46 Astro guide
1:55:50 dot a I I get
1:55:53 I get viruses so you don't have to
1:55:58 um
1:55:59 oh ask the Oracle oh this is funny dream
1:56:02 meanings oh this is cool my co-founder
1:56:05 and I are looking at this [ __ ] all the
1:56:07 time
1:56:07 hahaha
1:56:09 we talk about dreams all the time hello
1:56:11 I'm the dream Oracle
1:56:13 um okay let's see
1:56:15 um I dreamed
1:56:18 I was
1:56:20 let me see
1:56:25 I dreamed I was a peanut butter and
1:56:27 jelly sandwich
1:56:29 that tripped and fell down a hill
1:56:32 and spent the afternoon picking small
1:56:35 rocks out of my bread
1:56:40 thank you for sharing your dream with me
1:56:42 dreams can be quite Whimsical the
1:56:45 presence of peanut butter and jelly
1:56:47 sandwich could symbolize the blending of
1:56:50 contrasting elements or qualities in
1:56:52 your life very nice
1:56:55 oh this is cool
1:56:57 so this is this is what's commonly
1:56:59 called a wrapper app right so they're
1:57:00 taking this is some pre-prompted
1:57:03 pre-prompted stuff sitting on top of
1:57:06 chechi PT 3.5 or GPD 3.5 the the base
1:57:11 model this would be better if they used
1:57:13 gpt4 but it's a lot more expensive so I
1:57:16 understand why they're doing 3.5
1:57:18 um that's pretty cool I like it
1:57:21 I'll send that to my business partner
1:57:23 she will get a kick out of that
1:57:25 type of prompt uh I did what where
1:57:30 oh barbarino's tagline yeah
1:57:34 what where oh yeah that's right that's
1:57:37 right Vinnie barbarino up in those with
1:57:39 the rubber hose I'm so confused what
1:57:41 where
1:57:42 that's right there's a cube your
1:57:44 download is at the top of the list are
1:57:46 you a machine learning engineer no but
1:57:49 I've met some
1:57:52 what are your qualifications sir
1:57:55 so what do you refer are you a machine
1:57:58 learning engineer oh no
1:58:00 what are your qualifications
1:58:03 with regards to
1:58:08 it's called the AI learning lab I assume
1:58:10 by lab you're a scientist correct
1:58:13 no that's just marketing that's what we
1:58:17 call marketing
1:58:23 a little thirsty carry on
1:58:32 up your nose with a rubber hole all
1:58:35 right
1:58:36 man John Travolta was so good in that
1:58:38 you know it's amazing if if you watch
1:58:41 Welcome Back Cotter the fact that John
1:58:43 Travolta escaped that character like
1:58:46 there's some actors that when you do a
1:58:48 character that
1:58:50 identifiable like Henry Winkler it took
1:58:53 Henry Winkler
1:58:55 decades to escape the funds
1:58:59 and uh and Travolta
1:59:02 Travolta escaped that character that was
1:59:04 pretty pretty remarkable
1:59:06 uh
1:59:08 suno dot AI it's not slash chirp so so
1:59:12 if you go to suno dot AI
1:59:15 It'll point you to their Discord server
1:59:17 so you have to join their Discord server
1:59:19 and then you go into one of the chirp
1:59:20 channels then you type slash chirp
1:59:23 type in the style of music and then
1:59:25 it'll pop up a dialog box to either
1:59:27 paste your music in or have it AI
1:59:29 generate your your lyrics for you and
1:59:33 then it'll make the music it's pretty
1:59:34 cool
1:59:35 all right I gotta go to bed it's almost
1:59:36 midnight here people what are you doing
1:59:38 with my life what are you doing what are
1:59:41 you doing
1:59:42 weirdos
1:59:44 we're going down to the 405.
1:59:48 all right let's see what do we got here
1:59:50 I'll answer I'm I'm it just jumped me to
1:59:52 the bottom I grew up in a town of 12 000
1:59:55 people I was scared I'd be stuck there
1:59:57 tell me about it man I couldn't believe
1:59:59 no one wanted to leave
2:00:02 it's just a different thing like I and
2:00:04 you know it's weird like it was as
2:00:06 foreign to me to stay there
2:00:08 as it was for them to leave and I mean
2:00:11 partial partial part of that was I was
2:00:13 an Air Force brat so I moved around a
2:00:15 lot
2:00:16 so I'd see me seeing some other places
2:00:18 but I just I don't know
2:00:20 there was a lack of curiosity in that
2:00:22 small town that just made me
2:00:25 nuts
2:00:30 uh replica whoa the message was a while
2:00:33 ago and hello Nexus 9.
2:00:39 uh there's a small there's a small
2:00:41 amount of employed team
2:00:43 there's a site already in testing that's
2:00:46 cool replicant thank you for sharing
2:00:48 that
2:00:49 oh they announced us they they announced
2:00:52 it oh cool
2:00:54 yeah like listen I love mid Journey I've
2:00:57 got it on my little I got it on my list
2:00:59 of of [ __ ] you should [ __ ] you should
2:01:03 use I've got mid-journey Adobe Firefly
2:01:06 and Leonardo and mid Journeys first
2:01:08 because it's the best
2:01:10 it's the best although you know sdxl is
2:01:13 is looking sweet but I assume the [ __ ]
2:01:16 that you're playing with that we can't
2:01:17 play with is probably pretty
2:01:19 mind-blowing
2:01:21 uh I'm just learning to use AI very
2:01:25 useful any advice for teachers with
2:01:27 their students
2:01:28 um yeah so
2:01:30 um don't hide from it at all
2:01:34 um understand that these tools are going
2:01:36 to be everywhere
2:01:41 I know
2:01:44 um
2:01:46 so so understand that these tools are
2:01:49 going to be everywhere
2:01:50 um
2:01:52 you know
2:01:53 the the thing that the teachers that
2:01:56 that I've heard talk about this stuff
2:01:57 that I that I think is really
2:01:59 interesting what they're doing are
2:02:00 teachers that are bringing it into the
2:02:02 classroom
2:02:03 and increasing the the critical thinking
2:02:08 around these tools so
2:02:10 rather than saying don't use these tools
2:02:12 to write a paper bring it into the
2:02:14 classroom write the paper have it write
2:02:16 the paper and then have everyone you
2:02:20 know dissect it and discuss what it did
2:02:23 wrong where did it hallucinate
2:02:26 um did it use any weak arguments what
2:02:28 would they have done different
2:02:30 um you know get into that critical
2:02:31 thinking because I I
2:02:34 if I sort of crystal ball the future
2:02:38 you look at tools like these and you say
2:02:40 all of the Tactical execution from
2:02:43 drawing to writing to music the Tactical
2:02:48 execution of it
2:02:50 the grammar the the writing the
2:02:52 composition
2:02:54 is going to be automated
2:02:56 what's not going to be automated or or
2:02:59 will take longer to automate
2:03:01 is the critical thinking of like why am
2:03:03 I writing this who am I writing it for
2:03:06 what's my goal like how do I want this
2:03:08 to impact people
2:03:10 um what does success look like what's
2:03:12 quality look like critical thinking
2:03:15 critical thinking critical thinking so I
2:03:17 feel like the the value of critical
2:03:19 thinking goes up dramatically in a world
2:03:21 where tactical execution is automated
2:03:25 um so I would do everything in your
2:03:26 power to increase using the Noggin
2:03:30 uh actively all right is there a way to
2:03:33 check students writings to see if it
2:03:34 came from AI there are there are tools
2:03:36 to check it but they have a lot of false
2:03:38 positives a lot of false negatives and
2:03:40 smart students can get around all of
2:03:42 those Checkers they're useless give up
2:03:44 on the checkers these tools are going to
2:03:45 be everywhere
2:03:47 embrace it bring it in instead of trying
2:03:50 to push it out because you can't we
2:03:51 can't uninvent it and it ain't going
2:03:53 away don't use the Checkers in my
2:03:56 opinion now if your school district says
2:03:58 you have to great
2:04:00 um New York City Schools banchat GPT at
2:04:03 the beginning of the year and six months
2:04:05 later they unbanned it
2:04:06 because they get it city of Boston wrote
2:04:10 um generative AI guidelines go look at
2:04:12 the city of Boston's AI guidelines
2:04:14 generative AI guidelines use that as a
2:04:16 model to talk to your
2:04:19 um your school district about it your
2:04:22 school district should probably have a
2:04:24 policy on generative AI
2:04:26 and not one that's like [ __ ] you
2:04:28 generative AI but one that's like
2:04:30 we it's here it's coming here are the
2:04:33 challenges with it here are the dangers
2:04:35 of it here are the good ways to use it
2:04:38 city of Boston generative AI guidelines
2:04:40 really powerful all right it is midnight
2:04:42 I gotta go I gotta leave you people it's
2:04:47 really good to be back it's fun hanging
2:04:49 out sorry I had to start so late but you
2:04:51 know that was what the tick tock said so
2:04:54 really good hanging out with you all
2:04:56 peace out peace peace peace peace pace
2:05:00 uh I appreciate all of you hanging out
2:05:03 take care