Dec 5, 2023 - AI Salon's First Birthday: Celebrating a Year of Innovation and Insights!

Description

Chapters

Transcript

0:12 all
0:14 [Applause]
0:18 right all right I'm going
0:22 live the hell am I doing what the hell
0:26 am I doing what the hell am I doing
0:31 Happy First Birthday AI
0:35 salon wait
0:37 it look at
0:40 that oh wait you know where I can go
0:43 here we can just zoom in on the little
0:45 turtle Silver Fox did this cute little
0:49 turtle cutest little turtle ever curfew
0:52 young man lunar stick I know I was I was
0:56 watching the voice and I was like I'm
0:57 not going to go live and then I looked
0:59 at and at the uh the Irregulars Channel
1:02 and everyone's like where is he we
1:04 haven't heard from him like I'll
1:11 go all right we'll do a short one we'll
1:15 do a a short little teeny tiny one I
1:17 probably won't even demo anything we'll
1:18 just go look at some we'll go look at
1:20 look at stuff let's see look here's
1:24 our oneye anniversary
1:29 logo and we
1:33 got Let's
1:36 see we
1:39 have a merch store now for the
1:46 salon so that was pretty cool so I'll
1:49 talk about
1:51 that and what
1:59 else I almost missed you I almost missed
2:02 me I almost didn't go live but then then
2:05 I saw everyone like where is he we
2:06 haven't heard from him damn
2:09 it you just can't sneak
2:14 out added an image to Irregulars for
2:18 this oh awesome hang on let me go
2:29 look
2:43 welcome
2:45 everybody this is the AI learning lab I
2:48 just got back from well I didn't just
2:50 get back went out and had a oh nice
2:53 that's cool went out and had a uh little
2:57 Afterparty after the AI salon
3:02 and then I came home and hung out for a
3:04 little bit
3:07 oh Lun stick all sad that the AI
3:11 learning lab wasn't happening how was
3:13 the party it was nice it was very nice
3:15 we had that very
3:21 cute all right this is this is for you
3:23 Lun stick we can we can have no
3:25 crocodile tears we got to fix that
3:32 oh man uh it was good it was good we had
3:36 we brought back um a lot of speakers who
3:39 spoke spoken over the past year that
3:41 just talked about you know where they
3:44 are with this stuff and you know what
3:47 insights they've had and then we heard
3:49 from Salon members and it was just kind
3:52 of a reminiscing it was just nice it was
3:53 just a nice kind of connecting and
3:56 talking kind of meeting um so so wasn't
3:59 wasn't super flashy which is nice was
4:02 good um yeah and so I'll figure I'll
4:06 come on here and I mean you know I mean
4:10 this is the kind this is the time of
4:11 year I mean we're coming up on the
4:12 holidays we're coming up on the new year
4:15 we just had the one year anniversary of
4:16 chat GPT we just said I just you know
4:18 tonight's the oneye anniversary of the
4:21 salon
4:23 um
4:25 and it's hard to put it in perspective
4:29 it's it's really hard to understand
4:31 where all this stuff is so anyway do me
4:33 a favor share the live even though I'm
4:36 just going to go live for like an hour
4:38 here or whenever I don't know we'll just
4:40 chat if you have any questions about AI
4:42 actually you know what would be cool to
4:44 put in there rather than questions why
4:46 don't I just continue the jam we had
4:48 going at the salon anniversary I'd love
4:51 to hear you know for anyone who's been
4:53 doing this for a while especially from
4:55 the Irregulars you know what's either
4:58 what's a big Insight you've had or like
5:01 where are you right now what's where's
5:03 your head where's
5:05 your you know have you reinvented
5:07 yourselves are you questioning
5:10 everything are you I know lunaticks
5:13 she's dusted off all of her index cards
5:15 and every index card is now a
5:19 GPT uh so hey micro what's happening
5:23 jery good to see you how was the party
5:26 party was nice we did you know what's
5:28 cool about that group is it's like
5:31 you know the party is just like a
5:32 continuation of the
5:34 conversation like we just go out
5:36 everybody grabs a beer some tacos and we
5:40 just sit around just saying well what
5:42 have you played with hey have you seen
5:43 this have you done that like it's just
5:45 it's just it's like this it's like what
5:47 goes on in here so it was
5:49 nice U what I saw of it was noise nice
5:52 Drew glad you glad you get to see a
5:54 little bit AI blocks me as a conspiracy
5:59 researcher yeah Joseph Baker you're
6:01 going to have to go get yourself some
6:03 uh some open- Source uh nonrestricted
6:07 large language model action
6:11 going the safety guard rails and
6:13 conspiracy theories do not go well
6:15 together so so you're going to have to
6:18 you're going to have to geek out I'm
6:19 afraid my friend here's the good news
6:21 you can use the the restricted large
6:24 language models to to train you how to
6:28 set up the unrestricted large Lang
6:29 anguage models you can probably get it
6:31 done in like an
6:33 hour I'm still M navigating my way
6:36 through this AI thing but I'm so
6:39 grateful to have stumbled into this
6:41 place well I'm I'm grateful that you've
6:43 stumbled in here um if it makes you feel
6:46 any better I said this tonight I said so
6:50 like I've been I've been fairly obsessed
6:53 for 2 years I have been clinically
6:56 obsessed for a
6:58 year um you know I started started the
7:00 salon I started this Tik Tok Channel at
7:02 58 I guess I started it at
7:06 57 I have I have aged since I started
7:10 this
7:12 um started a newsletter completely
7:15 ramped my company's product road map are
7:19 trying to figure out how
7:21 to keep my company relevant in the face
7:24 of
7:26 insane competition from infinite content
7:30 generating
7:32 machines and so like I feel like
7:38 I it's not really possible to
7:41 spend
7:43 more of my waking day than I do on
7:49 AI like like you know if I were a
7:52 machine I could but as a human I I am
7:56 pretty much tapped out every [ __ ] day
7:59 jamming [ __ ] into my Cranium about
8:02 Ai and I said this tonight I feel as
8:06 clueless as I've ever
8:09 felt so don't think that just because
8:14 you haven't been you know deep diving on
8:17 this for a year or two or
8:20 10 that that you're way more behind
8:23 you're not way more behind everyone
8:25 right now is trying to figure this
8:28 out and
8:32 the thing the thing that's got me you
8:34 know it's when you worry about the
8:37 future it causes anxiety right the thing
8:38 that's giv me a little
8:41 anxiety is I kind of know what's coming
8:45 right like I know that Gemini is coming
8:48 which is Google's chat GPT 4 competitor
8:52 that's going to be multimodal and crazy
8:55 and hopefully
8:57 good you know I know that grock is
8:59 coming coming I know that you know these
9:03 video models like pico 1.0 looks like
9:06 it's going to be absolutely [ __ ]
9:08 insane um we've been playing around here
9:11 for the past couple of nights with sunno
9:13 the music generating
9:15 thing
9:17 and it's it's like the
9:20 um the core tool Suite so the the way I
9:24 like to think about this is like if you
9:26 could only play with one AI tool
9:31 if you only had access to one what would
9:32 it be the the answer to me is crystal
9:35 clear right now it's chat GT4 period end
9:39 of story nothing's even
9:41 close but it's it's kind of like
9:44 professional tennis right you got
9:47 jokovic
9:48 Nadal and then everyone else you got in
9:51 F1 you got for stappen everyone
9:56 else that's how AI is right now you got
9:59 chachy p pt4 and everyone else the
10:02 problem is the everyone
10:05 else's are
10:07 rumbling they're getting really good
10:09 really fast and so what it means is you
10:13 can't you
10:16 okay she's
10:19 okay um you
10:22 can't ignore them but at the same time
10:25 you can't keep up with them and and
10:28 what's going to happen is the the tools
10:30 as they are right now so if you if you
10:32 look at here let me show
10:35 you is this it
10:38 yeah okay you got chat
10:41 GPT right you're like okay yeah yeah we
10:44 got chat GPT Kyle that's the big one
10:46 great perfect so just if there if there
10:49 was only one you'd do that one right AI
10:52 is my adviser and teacher very cool then
10:55 you got these four Claude Bard co-pilot
10:58 so co-pilot is Microsoft's chat gb4 and
11:00 you have grock grock isn't out yet but
11:01 it's coming so we have to we have to
11:03 keep that on the list and it's going to
11:04 be one of the big
11:06 four so you have to pay attention to
11:08 those four then I think it's worth you
11:10 paying attention to these four Pi
11:12 perplexity Po and llama
11:14 2 now all of these for the most part are
11:17 like chat GPT they started out as large
11:20 language models and now they're going
11:22 multimodal Claude isn't yet um Bard is
11:26 starting to go there with like Bard
11:29 right now one of the interesting things
11:30 they just added is Bard knows all the
11:33 YouTube videos so you can
11:36 take a URL from one of your YouTube
11:39 videos put it in Bard and tell it to
11:40 chapter eyesee it and it will because it
11:43 knows them all um co-pilot from
11:46 Microsoft is multimodal now and and then
11:48 they just added a thing called Deep
11:50 search or something like that some new
11:52 thing I just I just saw it in passing
11:54 tonight before we did the salon um then
11:57 you got these you got Pi is the this
11:59 conversational thing that went from
12:01 having one mode to having four modes and
12:03 getting much more capable and there like
12:06 I have a feeling Pi is just going to
12:08 explode and get multimodal perplexity is
12:10 is an increasingly complicated
12:13 sophisticated research tool po you can
12:16 play with like I don't know a dozen or
12:19 15 different large language models you
12:21 can make Bots like it's it's it's big
12:23 and then llama like I've got labs.
12:25 perplexity here that's just a place to
12:27 play with it llama is part of a whole
12:30 world of open-source models like
12:34 thousands of projects right so if you
12:36 want to if you want to dig
12:37 into large language models there's a
12:40 whole world just in that little thing
12:42 right there so that's just the wordy
12:44 words right that's just words I don't
12:47 even have Amazon they just launched
12:52 Q so so that feels a little overwhelming
12:55 and then you start looking at um the
12:58 image generation tools they're getting
13:00 more and more and more sophisticated and
13:02 faster and and like even the image
13:05 generation tools like in the past two
13:07 weeks two weeks we've gone from you type
13:11 a
13:12 prompt and you hit a button and then it
13:15 the the clock spins and then you get
13:18 back a image or four images and then you
13:21 upscale them right to these models where
13:24 you now draw a shape and it read it
13:27 renders the new image in real time we're
13:30 going from non-real time image
13:33 generation to realtime image
13:35 generation there's a thing called Kaa
13:37 which I actually signed up for um that
13:41 does I can take my webcam as input type
13:43 in a prompt and then it turns me into
13:45 some other thing in real time so even
13:48 these things are how you use
13:51 them is anxiety inducing because if
13:54 you've gotten good at doing it with the
13:55 wordy words or with dolly3 with with you
13:58 know with with you know doing a uh an
14:01 image prompt within the context of a
14:03 chat how you do these is going to get
14:05 different it's changing then you got
14:07 video this world is losing its [ __ ]
14:10 mind these tools are getting so
14:12 incredibly
14:14 powerful that I don't want to ignore
14:17 them I don't want to not do them but
14:19 making movies is hard my company
14:22 storyvine
14:23 exists because video storytelling is
14:26 hard and I thought there's got to be a
14:27 better way
14:30 and so storytelling doesn't necessarily
14:33 get easier with these the production of
14:35 the bits gets easier but you still got
14:37 to tell the
14:39 story why does AI I AI have a so much of
14:43 a problem with spelling words on images
14:45 the reason it's got a problem so in
14:48 general the reason AI has problems with
14:51 accuracy and predictability and
14:54 consistency is that the way the
14:56 Transformer model Works which is the the
14:59 the core underlying technology
15:01 underneath it how do I explain this
15:03 super
15:04 simply what that model does is it turns
15:08 words or pictures or music any sort of
15:11 data but let's just say words let's just
15:13 say like chat GPT
15:15 words it takes all the words that it is
15:17 trained on which is you know trillions
15:19 of them billions of documents like as
15:22 much of the public internet is they
15:23 could get their hands on and it turns
15:26 them all into um strings of numbers that
15:31 live in Thousand dimensional
15:35 space and you're like what yeah exactly
15:38 it's weird it's weird and so when you
15:42 type a prompt in chat GPT and it writes
15:45 you that big ass response and it looks
15:47 like there's like a really smart person
15:50 on the other side of it it's really just
15:52 a calculator that is calculating what is
15:55 the next most probable word after this
15:58 string of of words you're prompt and
16:01 then what's the next most probable word
16:03 after that and the next most probable
16:04 word after that that's that's
16:06 technically what it's doing and it's it
16:07 doesn't even know the words it's putting
16:09 there it's looking in its mathematical
16:13 Universe plucking out a number and then
16:16 that number or a series of numbers a
16:18 vector that Vector is associated with a
16:21 word so it goes I need this set of
16:23 numbers and it pulls it down and says
16:25 okay that set of numbers goes with the
16:26 word dog and it puts that word in and so
16:29 it just spits all that [ __ ] out and it
16:32 doesn't know what it wrote so if if the
16:36 if it's sort of pulling from training
16:38 data that's got inaccuracies in it it's
16:39 going to give you inaccurate stuff the
16:42 image things are the same way it doesn't
16:44 know what it's creating when when these
16:47 tools create the images they create they
16:49 don't go back and look at them and
16:51 analyze if they got them right that will
16:54 be coming like 2024 I guarantee you
16:57 we're going to see that where we're
16:58 going to see see image models that
17:00 generate the
17:01 image analyze the image to see if they
17:04 got it right and keep iterating until
17:06 they get it right and maybe even fixing
17:07 just pieces of it I guarantee you we'll
17:09 see that in
17:10 2024
17:12 so as of 3 months
17:15 ago generative images couldn't do words
17:19 at all so the fact that they're doing
17:20 words even remotely
17:23 good is remarkable let me go look
17:27 at let's see
17:30 like there there's a good one that that
17:32 Sherry did for the for the AI Salon
17:35 happy is right happy
17:37 anniversy sorry AI Salon it got almost
17:41 everything right I don't know what
17:42 you're complaining about but I mean it's
17:45 pretty remarkable that it got it that
17:46 close but all it's doing is it's taking
17:49 a bunch of images sort of smashing them
17:52 together as as this massive pixels and
17:54 then denoising them and then what it
17:57 generates it doesn't even know what in
17:59 there so that's why it does that why
18:01 doesn't it work for foreign languages
18:05 uh I I when was the last time that you
18:08 looked at it what why doesn't what work
18:11 for foreign languages hold
18:16 please all right this is chat GPT this
18:19 is the chat GPT app I'm putting it in
18:22 conversation
18:25 mode I want you to act as a Japanese
18:28 translator I'm going to speak in English
18:30 and I want you to translate in Japanese
18:31 and don't repeat the
18:43 English the AI Salon had its birthday
18:46 tonight and I'm tired but very very
18:57 happy
19:03 [Music]
19:05 come on chat jpt we don't have all
19:19 night now on second thought I want that
19:22 in German
19:27 please
19:36 how about Montreal
19:45 French so they do do languages it knows
19:49 all the languages it knows all the
19:50 programming languages it knows all the
19:52 languages they're translating
19:56 Sanskrit uniform writings
19:59 into English by just taking snapshots of
20:01 [ __ ] like it it knows all the stuff
20:07 now the nature of where this technology
20:10 is as of
20:12 today it's got a lot
20:14 of the technical term for it that we
20:17 talk about a lot on this channel is
20:19 janky pieces of [ __ ] all of this
20:21 technology right now will get you to
20:23 about
20:24 80%
20:26 of good and then the last mile the last
20:30 20% oh it's
20:32 rough but just watch 2024 A lot changes
20:36 just you're going to see they they're
20:38 going to start putting they're they're
20:40 going to start this is what qar is to my
20:42 understanding you've got chat GPT down
20:44 here they got this thing called qar
20:46 which is this um game mechanics training
20:50 system basically they they've merged two
20:52 Training Systems together that basically
20:54 say okay chbt you went and you took an
20:57 action I'm going to look at that action
20:58 and give you a score on how well you did
21:01 and and the score is either going to go
21:02 up or it's going to go down and if it
21:04 goes down then you got to do it better
21:05 and if it goes up then you keep going in
21:07 that direction so you're going to start
21:09 seeing in
21:11 2024 the typos are going to get way less
21:15 the hallucinations are going to get way
21:16 less these things are going to get way
21:18 better and all of the stuff people are
21:20 bitching about right now it doesn't
21:22 spell right
21:24 hallucinates are all of a sudden going
21:26 to get dramatically good dramatically
21:29 David Shapiro talks about this he goes
21:31 the the the thing about machine learning
21:33 and this kind of technology is it's bad
21:35 it's bad it's bad it's bad it's bad it's
21:36 bad it's bad it's okay it's okay it's
21:40 excellent we're we're still in the it's
21:43 bad it's bad it's bad it's getting
21:44 better it's getting better it's like
21:45 we're we're just like there and we're
21:47 about to
21:48 go if qar the the rumor is that there's
21:53 a thing called qar that they that they
21:56 put together at open AI and it made some
21:59 sort of advancement and the rumor is
22:01 that's why they fired Sam Alman cuz [ __ ]
22:02 was moving too fast and the nonprofit
22:04 board said
22:08 ah if that's true shit's going to get
22:11 faster fast or better fast what can you
22:13 tell us about Tree Learning when it
22:16 refers to AGI um let me go see if I can
22:20 find that David Shapiro video hold
22:26 please you mean q learning
22:29 I think it's Q learning so qar is a
22:31 combination of a-learning which I think
22:34 is tree
22:35 logic
22:37 and a star and Q learning
22:41 qar
22:44 um let me go see if I can find this
22:46 video David Shapiro if you if you're
22:49 interested in Ai and you don't follow
22:51 David
22:53 Shapiro you are
22:57 missing you are missing stuff by the way
23:00 his his um AI predictions in 2024 video
23:05 is really
23:07 good
23:08 um bigger than you
23:12 think that wasn't it it was what is qar
23:15 hang
23:23 on what is qar you know a lot of people
23:27 think that in order to conceal
23:32 good morning today's video is brought to
23:35 you by
23:36 insomnia that's not a product that's
23:38 just I woke up at 3:00 a.m. today and I
23:41 woke up to see some really cool news so
23:43 okay so let let me show you a couple of
23:45 things thought Chain of Thought okay so
23:49 so there's there's this this uh graphic
23:55 here so how do I explain this so so this
23:58 is this is the this is the logic behind
24:00 qar so you've got a starting State and
24:03 the starting state is I don't know I
24:06 want to find the best prices on travel
24:08 or whatever it is and then you have the
24:10 ending State like here's your goal your
24:12 goal is to go there and then what
24:15 happens is given this starting
24:19 State the
24:22 the you're you're going to um take an
24:26 action and you're going to take action
24:29 one in state one and the combination of
24:32 the action you take in that state is
24:34 going to give you the quality so Q is
24:36 stands for quality and then what happens
24:39 is so so given this state maybe you
24:42 maybe there's four actions you could
24:43 take you could you know launch a Google
24:45 search you could whatever it is you
24:48 could you could do things and the way
24:49 qar works is it says okay you took that
24:52 action I'm now going to look at the
24:53 results of that action I'm going to
24:54 score it I'm going to give you a plus
24:58 027 or a minus point4 right depending on
25:03 the dramaticness of how bad you did and
25:05 then we're going to move on and then
25:07 given the new state we're going to move
25:09 on to state two and and for state two
25:12 you got a set of actions and then you
25:15 take an action and I judge that and I
25:17 add you know I either add or subtract to
25:20 your Q score and then you just basically
25:22 keep going around so that's that's just
25:24 to understand the kind of math behind it
25:26 and then there's this other drawing here
25:27 that he
25:29 does where so you're sort of you start
25:33 down here and and the state is the
25:35 little red dot in the middle and then
25:37 you could take you know one of nine
25:38 actions around it so it takes this
25:40 action your goal is up here and and
25:43 things are going good you know the the
25:45 your Q rating is going up it's going up
25:47 it's going up and then you get here and
25:48 something's wrong right and your Q score
25:51 starts going down so it says okay I
25:53 think I need to take different actions
25:54 and so it takes a different action and
25:56 it basically learns as efficiently as
25:59 possible to navigate itself around
26:01 whatever the you know logistical
26:03 roadblock is and it and it gets here so
26:06 let's say um the the what they
26:10 apparently used qar
26:12 on like three weeks ago four weeks ago
26:15 was simple grade school level math
26:18 problems and what normally happens with
26:22 large language models is they can
26:25 fine-tune them and fine-tune them and
26:26 fine-tune them and maybe they can get
26:27 them from like 70% correct to 80%
26:31 correct to maybe 85 maybe 90 but they
26:33 can almost never get to 100 well they
26:35 apparently got to 100 because if it got
26:38 the problem wrong at some point or some
26:41 part of the problem wrong it would sort
26:43 of navigate it its way around it and and
26:46 it would get the problem
26:48 correct that's
26:51 reasoning that's what chat GPT and these
26:54 things can't do right now all they are
26:56 right now is dumb calculators they're
26:59 dumb prediction engines they're the
27:01 smartest dumb calculators I've ever
27:03 experienced in my life like they're
27:06 beyond anything I've ever seen but
27:07 they're not
27:08 reasoning this [ __ ] is reasoning now I
27:11 don't know if that's exactly what you're
27:13 talking about qar is just speculation
27:15 yeah that's right Pate Pate Pate is
27:17 keeping me honest here Pate works for
27:19 Google qar is totally speculation but it
27:23 it's going to be some version of this
27:24 right it's going to be some version of
27:26 look at the [ __ ] you did look at the
27:29 action you took did you get it right if
27:31 not try
27:34 again
27:36 uh how can I make mid-journey type of
27:39 word use
27:41 do you can't right now mid Journey
27:44 doesn't do words they will probably come
27:47 out with something in the next two or
27:49 three months that does it because you
27:51 can now do words in stable diffusion
27:53 which you couldn't do before her Herod
27:55 who's in here a fair amount wait is that
27:57 just an image of an a pathfinding
27:59 algorithm yeah this is a part of the
28:01 video where he's talking about AAR AAR
28:04 so I don't I don't I don't know he
28:07 describes in here something about how
28:11 qar is different than a star was
28:14 but that's as much as I know you
28:16 probably know more than I well the fact
28:18 that you said wait isn't that just a
28:20 diagram for AAR says that you know more
28:22 about this than I do Pate by the way
28:24 Pate is the Guild Master of the um AI
28:29 mechanics Guild over in the AI Salon
28:32 where we just had our first year
28:33 birthday
28:35 tonight and um so if you want to like
28:39 hang out with people that know a lot
28:41 about this stuff join the salon so
28:43 here's the if you go to the
28:46 salon. um we just updated the link tree
28:49 there's a link to the mighty networks
28:51 Community there's a link to the meetups
28:53 we meet every other week our our meeting
28:54 was tonight and then there's a link we
28:56 just added a new merch store
28:58 um and we've got a one-year anniversary
29:01 um special edition t-shirt it's only
29:03 going to be up for 48 hours so that's
29:05 the one with the cool graphics in it it
29:07 looks
29:10 like it looks like
29:13 that it's
29:15 cool it's a cool graphic see the one and
29:19 the A and the I and the road and the Sun
29:22 and the you
29:24 know
29:26 um yeah all right Right Moving On Moving
29:30 On We're moving on up to the east
29:34 side to a Delux apartment in the
29:38 sky yeah we're moving on
29:41 up all right new chat limit per hour
29:44 made me upset on
29:46 gp4 oh the their rate limiting I know
29:50 they're their problem is their problem
29:53 is too many people are using their [ __ ]
29:55 cuz it's so good and everyone's addicted
29:56 to it now anytime someone gets rate
29:58 limited I get like 10 pings I just got
30:01 rate
30:01 [Laughter]
30:06 limited I don't know what to tell you
30:08 people all I can tell you is keep stay
30:11 curious keep playing keep building [ __ ]
30:14 assume it's going to
30:15 break assume that the tool that you're
30:18 using that seems like it is a
30:22 miracle is just a janky piece of [ __ ]
30:25 just you you got you have to accept
30:27 those two realities this is the most
30:30 miraculous thing I've ever seen and and
30:33 no one should use this software both of
30:36 those are
30:37 true where's the Llama I got llamas I
30:41 got llamas in here
30:42 hold uh uh uh there llamas llama llamas
30:48 you know why I like llamas llamas is
30:50 tasty M
30:54 llama he said qar will be a combination
30:56 of a star and Q
30:59 learning well that's the again as P
31:02 pointed out that's the speculation they
31:04 have not confirmed that now in an in an
31:06 interview with I think Vanity Fair last
31:09 week Sam wman did say that the letter to
31:14 the board about said algorithm was a
31:19 leak he confirmed that it was a leak um
31:22 but we don't know any more details other
31:24 than that he said it was an unfortunate
31:27 leak
31:29 um will we be homeless next year I do
31:31 not think so I do not think so so one of
31:36 the things that's very easy to fall into
31:38 is the robots are going to kill us um I
31:42 think there's going to be a lot of
31:44 disruption I think there's going to be a
31:46 lot of job loss I think there's going to
31:48 be an adjustment period and then I think
31:51 as these tools get more sophisticated
31:53 and we learn how to wield them better
31:57 there's going to be new kinds of jobs
31:59 created
32:02 um I
32:05 think project
32:07 managers are going to be in demand I
32:10 think that
32:13 um people who have software experience
32:16 are going to be in demand
32:18 because what AI is doing is
32:20 democratizing software development and
32:23 so what the the the the beauty of that
32:27 is lots of people get to make
32:29 applications the absolute [ __ ]
32:32 disaster of that is lots of people get
32:34 to make applications and they don't
32:35 necessarily have the discipline of say
32:39 like QA testing and planning all of the
32:43 things you know keeping track of
32:44 different versions and the thing was
32:47 working good and then they made some
32:48 changes and now it's work not working
32:50 good if you don't have a good system in
32:52 place you can't go back right Vanity
32:55 Fair is the go-to publication from my
32:57 text news P's feeling saucy
33:02 tonight I don't it might not have been
33:04 Vanity Fair which I don't know what it
33:06 was but whatever it was it was some
33:08 interview with Alman where he said that
33:10 the uh the leak was an unfortunate it
33:13 was an unfortunate
33:16 leak oh
33:19 man it's not going to be the robots AI
33:21 can just use existing things against us
33:23 I that's really true save save on
33:26 Manufacturing it's so true yeah the
33:29 robots when the robots you know what
33:31 it's going to be the robots are going to
33:33 just toy with us like a cat for a while
33:35 they're be like they're
33:39 cute they're just gonna
33:43 go
33:47 flick got my t-shirt confirmation
33:50 awesome AI Bob that's great
33:53 fantastic um The Jeffersons yeah moving
33:56 on up to the
34:00 [Music]
34:05 side hi I just got here let me tell you
34:07 about
34:08 [Laughter]
34:10 qar
34:13 oh what did I miss tonight uh you missed
34:16 a birthday party we had a birthday party
34:18 for the AI salon and I just came
34:22 on
34:25 um 20 minutes ago 30 minutes
34:28 ago uh I'm just as much fun at virtual
34:31 parties as real
34:39 ones he's got all the
34:43 friends oh
34:46 man any updates on Gemini not that I saw
34:49 before I was PL I was planning for the
34:51 salon all day we were we were putting
34:54 the store together and the um
34:58 the link tree together and the what's it
35:01 called Mighty networks together we were
35:02 editing Mighty networks like we were
35:04 just trying to get all that because
35:05 that's that's now live so if you go to
35:07 aith
35:09 salon. and go to the community site
35:12 That's mighty networks it's now public
35:14 so um you can just go join that have you
35:16 read about Mamba apparently it's an SSM
35:19 versus an llm a small a small what model
35:23 an you mean an slm small language model
35:26 versus a large language model
35:28 um I've heard of Mamba I haven't I
35:30 haven't um played with it
35:33 um Microsoft
35:37 on dev day on open ai's Dev
35:40 day Microsoft talked
35:44 about being into small language
35:48 models which basically is to say open AI
35:51 ain't the only game in town and then a
35:53 week later Sam Alman got
35:56 fired um so so yeah I think we'll
35:58 probably hear something
36:00 about is small language just slang no
36:05 Lun the I Pate may know this better than
36:07 I do but small language models are just
36:09 basically tighter trained they're not
36:11 trained on massive amounts of language
36:13 they're they're trained on specialty
36:15 pieces of of language so so you don't
36:17 have as much hallucination if you're
36:19 training things on smaller data sets
36:21 basically it's like I what I don't know
36:24 is I don't know the difference between a
36:26 small language model and say an expert
36:29 um AI system from seven years ago like
36:35 like maybe just because it's the
36:36 Transformer and it's a Slicker version
36:38 of it I don't
36:40 know that I do not know supposedly it
36:44 doesn't have the limitations of the
36:46 large language model yeah but it also
36:47 doesn't have the general capability of
36:49 the large language model so everything's
36:52 a trade-off
36:53 okay I get it I just didn't know
36:57 it's slm I think I
37:01 think paint is it slm or S
37:04 SSM so an
37:06 slm so would slm be better at doing a
37:09 particular type of scientific research
37:12 yeah I think
37:13 so oh SSM is a unique method it's not
37:17 slm see you know what
37:19 we're it's it's almost 11:30 I'm
37:22 exhausted I am significantly out of my
37:25 depth here so I don't [ __ ] no uh go to
37:29 the AI Salon join the mechanics Guild
37:32 and and hang out with Pate he will tell
37:35 you all this [ __ ] he can also teach you
37:38 the difference between a TPU a GPU and a
37:42 CPU in in like a short amount of time
37:45 and you actually get it it's pretty cool
37:47 yeah I don't know much about it but we
37:49 can figure it out together and that's
37:50 the whole idea of the salon if if you
37:52 haven't been to the salon it's it's a
37:54 really cool group of people it's it's
37:56 very much about everyone's trying to
37:57 figure all this [ __ ]
37:59 out
38:01 um we're all trying to figure it out
38:03 nobody [ __ ] knows nobody [ __ ]
38:05 knows nobody knows nobody knows even if
38:08 you understand how the [ __ ] Works
38:10 technically okay so now you know the
38:13 difference between SSM slm and
38:16 llm how does that benefit your pizza
38:19 shop I don't know that's the [ __ ] we got
38:22 to figure
38:23 out but I'm a gardener what do I do with
38:26 this stuff
38:29 hang out in the salon hang out with
38:30 people that are looking at things like
38:33 you're
38:37 not State space
38:39 model SSM State space model oh you know
38:43 what we can do
38:47 duh oh chat
38:49 GPT why are they asking a man with an
38:52 acting degree about an
38:56 SSM
38:58 be because he's a mansplainer he talks
39:01 about things as if he knows about them
39:03 even when he
39:05 doesn't this is a man
39:11 [Music]
39:13 spainer he is known Cally as
39:19 Kyle
39:25 um can you explain the difference
39:29 between an SSM an slm and an
39:34 llm in the context of AI as if I had the
39:38 education of a
39:41 [Laughter]
39:45 seven-year-old sure imagine you have two
39:47 different kinds of smart robots one is
39:50 an
39:51 SS wait SS SM
39:55 NSL good Lord which which stands for
39:59 small scale model but that's not what
40:01 wait what was the thing that you said
40:02 something State simple State what was
40:06 it small scale wait I
40:09 think
40:13 um SSM stands
40:16 for small state
40:25 something small state machine
40:32 I don't
40:38 know I can't
40:41 I I don't know I don't even know enough
40:44 ask chat gbt to tell us what's the
40:47 difference this is what it's like
40:50 sometimes you know what I can do you can
40:52 make Mone
40:55 with
40:59 [Laughter]
41:02 next moving
41:04 on moving on up to the next
41:11 topic the
41:13 Simons happy birthday AI
41:17 Salon all right I'm going to do like
41:19 another 15 minutes if the next 15
41:22 minutes go as good as the past 15
41:24 minutes woo what a night it's relatively
41:26 knew it's complicated it's part of a
41:28 push to get past Transformers oh that
41:30 makes sense cuz every everyone's like
41:34 well seems like open AI is won on the
41:36 Transformer game why don't we invent
41:38 something new maybe we can beat him that
41:40 way Kyle can you show us Runway that I
41:43 can show you do you want to see Runway
41:44 TV or Runway ml cam cam cat cam cat
41:49 can cam K him from
41:53 Cleveland you want to see the TV show or
41:55 you you want to see the TV network
41:58 that's going to put Netflix out of
42:00 business if they ignore it or do you
42:02 want to see something
42:07 [Music]
42:10 else State space model let me go back
42:13 and ask that state
42:18 space model look it up what is
42:25 it
42:27 doing research with Bing
42:29 okay all the letters single short
42:33 [Laughter]
42:35 message uh can you show us Runway Tobias
42:39 has a birthday this weekend
42:41 awesome wait State space model is a type
42:43 of mathematical model used in various
42:45 Fields such as control engineering
42:47 Finance economics ecology to simulate
42:50 Dynamic systems basic concept a state
42:53 space model represents physical systems
42:57 by defining a set of inputs outputs and
42:59 internal States well that's kind of what
43:01 we were talking about with the qar
43:04 thing these internal states are key
43:06 variables that describe the system's
43:08 Behavior at any given time yeah the
43:11 relationship between the inputs the
43:13 outputs in the states is typically
43:14 expressed as a first order differential
43:17 or differential equations you know them
43:19 first order differentials I I'll tell
43:22 you what I you know I was at a party
43:24 once I was talking about first order
43:25 differentials and I was like like
43:28 man the wait till you see the second
43:31 order different differentials in the
43:34 it's in the transmission right because
43:36 those the the first order goes fast but
43:39 then you put it in the second order and
43:41 the differential spins and the tires
43:43 just grab and so I know I know math
43:47 pretty good I know it pretty good I'm
43:49 it's I got a solid handle on this so
43:51 yeah come back here AI learning lab
43:54 you'll learn good stuff f fantastic math
43:58 stuff here at the learning lab that's
44:01 what we're all about
44:06 [Laughter]
44:10 qualifications a sure way to get the
44:12 babes yeah listen they uh saw you
44:15 standing there by the punch and I was
44:17 thinking you know I I was just thinking
44:19 about first order differentials and uh
44:23 you're in
44:25 math let
44:30 [Music]
44:32 I've been at parties where order of
44:33 differentials are discussed at length it
44:36 was at a government lab P that's not
44:39 called a party Pate do you know what a
44:41 party is dude we're gonna we're going to
44:43 start a different Guild and we're going
44:45 to invite you to
44:46 [Laughter]
44:49 it you can't discuss differentials at a
44:53 party it's no no
44:58 no that's not a party that's just work
45:01 with
45:05 alcohol a parties where you act like an
45:08 idiot and forget what first order
45:11 differentials are is it differentials
45:15 yes differential
45:18 [Laughter]
45:22 equations so then I says to her how do
45:24 you handle your hidden States and she
45:26 says to me it's a first order
45:29 [Laughter]
45:43 thing Runway okay
45:47 um I was nearly late for the salon cuz I
45:50 got my hand bit by a pitbull no at the
45:54 dog park oh no are you you okay oh
45:57 that's [ __ ]
45:59 awful I'm sorry about that W was it's
46:03 not good pimples and
46:08 hands that's
46:10 horrible horrible
46:14 horrible okay someone wants to go
46:18 see is we're going to go over here we're
46:21 going to go to Runway ml so Runway so
46:24 let me let me show you something first
46:25 okay
46:28 we're going to go here we're going to go
46:30 here we're going to go here is that it
46:32 no no no yes okay
46:37 this this is the world of video right
46:40 now well no the world of video is
46:41 probably way bigger than this this is
46:43 the world of video that I've played
46:46 enough with that I think and heard
46:49 enough about so generally how I pick the
46:53 tools that make this list or like all
46:56 the stuff that I recommend is if I hear
46:59 about something once or twice I don't
47:02 care if I hear about something 10 times
47:05 I'll go check it out I will make a snap
47:08 judgment on
47:11 it because that's how my brain works I'm
47:13 a synthe I'm I got I got ADHD I don't
47:16 have time for your [ __ ] nonsense your
47:18 I don't have time for your [ __ ]
47:20 manual I will go look at it and I will
47:22 snap judge it and then if in snap
47:25 judging it
47:26 it's okay then I'll go look at
47:28 it
47:31 um most stuff does not even come close
47:34 to making this list so what's on this
47:37 list are things that I've actually
47:38 played with or enough people have talked
47:40 about and I've seen output from like
47:42 okay that that's one you got to pay
47:44 attention to
47:45 so the three sites on the left or yeah
47:48 the three projects on the left are are
47:51 all the same basic thing Runway ml Moon
47:54 Valley and pabs Moon Valley seems to be
47:56 drifting into the distance a bit P Labs
47:59 is about to sort of take off in front of
48:03 uh Runway ml but Runway ml right now's
48:05 got the lead Runway ml is for um text to
48:09 video like you know chat GPT is to large
48:13 large language models it's kind of got
48:14 the lead right now and people are coming
48:17 after them
48:19 um but there's a lot of people that I
48:21 know like like um um some of the
48:25 creative directors I know Kenny that was
48:27 on the salon tonight is one of them
48:29 likes P Labs better and Pika 1.0 is
48:32 about to come out but all those three
48:34 things on the left are like you type in
48:36 text and it makes a video or you upload
48:38 an image of a video and it makes video
48:41 why they're getting better is they're
48:43 able to do longer chunk chunks of video
48:46 and they're also starting to give
48:48 directorial control so you can zoom in
48:50 on an image you can pan on an image with
48:53 Runway ml now you can there's a thing
48:55 called motion break brush where you can
48:56 paintt on an image so so I just wanted
48:58 to give you the context of what what
49:00 we're looking at oh and then the reason
49:02 I asked about Runway TV this thing is
49:05 worth checking out if you go to runway.
49:08 TV it's going to look weird and you're
49:11 going to
49:12 see films there videos there that look
49:17 weird and if you're old like me you may
49:21 remember in 1981 this weird TV station
49:25 came out called MTV Music
49:28 Television and nobody knew what to do
49:32 with it it was weird it was like we
49:34 we're used to listening to music and now
49:36 we're watching music what is this that's
49:38 kind of what Runway TV is it's videos
49:42 that are created out of this text to you
49:44 know text or image to video Generation
49:47 stuff and they're weird but they're
49:49 going to get very real very fast so pay
49:51 attention to that thing okay so let's go
49:55 to Runway
50:00 [Music]
50:02 ml we're going to log in all these video
50:04 tools are not cheap they charge you by
50:08 the cred they well well they give you
50:11 they they're there're it's kind of like
50:12 Vegas so you know when you go to Vegas
50:14 and you've got like I don't know
50:17 $1,000 and you're like this ,000 is a
50:21 lot of money and I know what $1,000 is
50:24 and they give you these little these
50:25 little
50:26 things chips and then you get like you
50:30 know 10
50:32 chips and all in a sudden it doesn't
50:34 seem like a lot of money and then you
50:36 just go play with those that's kind of
50:38 what these these companies do they they
50:40 uh they give you tokens and they give
50:42 you like 150 tokens you're like I'm
50:45 going to make 150 videos and then you go
50:47 to make a video and they're like this
50:49 video is going to cost you 15 tokens and
50:52 you're like oh
50:54 [ __ ] that's a lot of money money so
50:57 anyway that's what this is like
51:00 so the interface here is not super
51:03 intuitive but you can uh you can create
51:08 a video from text from an image or from
51:10 an image plus a description so I'm just
51:12 going to start with an image and so I'm
51:15 going to say upload a file so we're
51:16 going to go find an image to turn into a
51:20 movie um where am I going to go I'll go
51:22 here I got a lot of stuff
51:24 here
51:28 [Music]
51:32 that could be a cool one let's grab
51:35 well that's vertical let's
51:41 [Music]
51:45 see yeah let's do this we'll do this one
51:50 okay so I'm going to open that even
51:54 though it's vertical it'll make a
51:55 vertical vertical
51:57 VI
52:00 and then I can do things like I can say
52:04 um how much motion do I want and it's a
52:06 slider from 1 to 10 and then I can click
52:08 on this camera motion Tab and there's
52:11 all these different camera controls
52:13 there's
52:15 horizontal um there's there's
52:19 panning there's tilting so tilting is up
52:22 and down panning is left and right
52:24 horizontal is like tracking the camera's
52:26 doing this panning is it's doing this
52:28 it's rotating and then there's
52:31 rotation um so let's just say We'll zoom
52:35 out on
52:38 this and that's it and then we're going
52:41 to go to this thing called motion
52:45 brush and so now I've got a brush with
52:50 this purple thing on it and I'm just
52:51 going to brush all over the
52:54 leaves
52:56 and maybe a little bit of the
53:00 branches and I'm just G
53:02 to kind of purple my way around this
53:12 thing and you don't have to be really
53:14 precise these things aren't precise
53:16 they're kind of what did I what was the
53:18 term I used oh I know janky pieces of
53:21 [ __ ] so these are all janky now listen
53:24 if anyone in here or works at Runway or
53:27 P laabs or open AI or whatever and
53:29 you're like he K our stuff janky pieces
53:31 of [ __ ] it's it's it's a term of
53:33 endearment I love you you you're good
53:36 people and I'm going to make some leaves
53:38 down here move maybe all right and then
53:40 what do I do
53:42 there's
53:44 proximity see I don't you you you have
53:47 to do something with the so I guess
53:49 we'll go vertical down so maybe we'll
53:52 make some leaves fall I don't
53:54 know and then I'm going to save that and
53:57 then we're going to generate a movie
53:59 Bang so you start with a
54:02 picture you click on some
54:05 [ __ ] and you H generate video you can
54:08 also type in words um I want a picture
54:11 of
54:13 a an artvark
54:15 skateboarding and it'll make that and
54:19 then it takes a
54:20 while my experience with these is that
54:24 because I don't do it every day I'll
54:26 occasionally come in here and do what I
54:28 just did where I just kind of stumble
54:29 into an image and just kind of do it I
54:31 get pretty shitty results with these
54:34 people that spend time doing this
54:37 proximity will come closer to you yep
54:39 okay cool um people that spend time with
54:42 these can get decently predictable
54:44 results it's like anything if you use
54:45 the tool and get better with the tool
54:46 you'll be gooder the 4C limitation is a
54:50 bit annoying yeah but Jerry so um with
54:54 Runway now you can extend end a video
54:56 four times so you can go up to 18
54:58 seconds long so what it what it does
55:00 it's actually kind of cool it'll
55:02 generate a 4 second clip and then if you
55:05 say extend it for 4 seconds it takes the
55:08 last frame of the first clip and that
55:10 becomes the starting frame of the next
55:11 clip and it you can do that four times
55:14 so you can create 18 second sequences
55:17 the problem is they lack coherence right
55:21 which is basically this tree at the end
55:23 of 18 seconds would probably turn turn
55:25 into like a Rob red breasted Robin or
55:30 something um is this our movie where's
55:32 our movie oh yeah there you go look it's
55:39 working we got decent results it erased
55:42 all the letters and down come the
55:45 leaves that's not
55:49 bad so that's Runway doesn't have any
55:52 audio it's just it's you know this is
55:54 not
55:56 fully edited video this is just
55:58 animation right this is this is motion
56:01 sequences but if you go to runway.
56:04 TV you can see what people are doing
56:06 with this and and what they're also what
56:09 the clever storytellers over at Runway
56:11 TV are doing is they're mixing
56:13 traditional footage with digital footage
56:16 with AI footage and they're kind of you
56:19 know munging it all together with After
56:21 Effects and stuff like that I sometimes
56:24 expand to 18 seconds but you can speed
56:26 up the frames to say 9 seconds oh yeah
56:29 interesting cool oh yeah you've been
56:32 doing a lot of this Drew right you've
56:33 been you've been doing a fair amount of
56:35 this
56:38 so but go experiment just know that your
56:42 experiment's going to cost you about 30
56:44 bucks and you're going to get like you
56:46 know 15 shitty
56:49 videos you be like wait that's
56:54 it but what's the alternative go get a
56:58 camera go find a
57:00 tree the right time of year and film
57:03 some leaves falling out of it that's
57:05 your
57:06 [Laughter]
57:09 alternative oh man can we see Runway TV
57:12 absolutely okay so this is how you make
57:15 the piece parts and and to to submit
57:18 something to Runway TV you have to have
57:21 made something in Runway ml which makes
57:24 sense
57:29 [Music]
57:34 runway. so you start out that that thing
57:37 very much reminds me of the the early
57:40 MTV stream offline all right
57:51 fine that's normal footage oh maybe it's
57:54 not
57:59 [Music]
58:06 [Music]
58:10 and I don't know what this is it's
58:11 something but if I go to TV Guide here
58:14 this is where it gets this gets really
58:16 reminiscent of like early days of
58:17 Netflix early days of a lot of the
58:19 streaming services were kind of these
58:22 sort of scrolling lists of videos right
58:24 in in generally you know um curated
58:29 channels AI Film Festival gen 48
58:31 whatever that is you know maybe they
58:33 partnered with someone called gen 48
58:35 this is the staff pick film of the week
58:39 um the creative partners program Runway
58:41 Studios Runway Studios Runway Studios
58:45 pay attention to this who's Runway
58:48 Studios I don't know they took a pile of
58:50 the you know hundred million they got
58:53 invested or whatever it was and and they
58:55 said hey why don't we try to make some
58:57 original content let's go get some
58:59 filmmakers we'll give them our tools and
59:01 we'll just tell them to go make original
59:03 content so that's what that channel is
59:06 um you can submit your work to this
59:09 channel because we're so early just
59:12 anyone can do this like this
59:16 video is just weird it's cool but it's
59:20 just
59:24 weird music video right does this feel
59:28 like early MTV
59:31 days this was all hand animated back
59:43 then um so there's that kind of stuff
59:47 um endless Reflections animation
59:50 worldwide Couture
59:54 24
1:00:17 that's pretty slick so so it kind of
1:00:19 feels like art art films of the 70s like
1:00:21 the Andy Warhol stuff there's some in
1:00:23 here that are true story
1:00:25 telling and what's going to happen just
1:00:28 mark my words this is going to start to
1:00:30 get a little bit of Buzz or something
1:00:32 like it it doesn't have to be this one
1:00:34 like this this is just the first one
1:00:35 I've seen of
1:00:38 these and real filmmakers are going to
1:00:41 piss all over this they're going to be
1:00:43 like it's not real
1:00:46 film that's stupid real film's better
1:00:49 and then what's going to happen is for a
1:00:52 while there's going to be there's going
1:00:54 to be be two different tracks and one's
1:00:57 going to be a really thin track which is
1:00:59 this generative stuff and the other
1:01:00 one's going to be Hollywood and then
1:01:02 just over time that really thin track is
1:01:04 going to get wider and wider and wider
1:01:06 and who are the power players in the
1:01:08 generative side are not necessarily
1:01:11 going to be who the power players are on
1:01:13 the Hollywood side and there's going to
1:01:14 be a really interesting battle so just
1:01:16 pay attention to this stuff cuz right
1:01:18 now this stuff is I wouldn't call it
1:01:21 laughably bad but it is janky
1:01:26 this a lot of this work is the result of
1:01:28 janky technology that made it possible
1:01:31 but it's going to get real good real
1:01:34 fast that's me at the differential
1:01:40 party what will happen with Netflix so
1:01:43 cam that's so okay so
1:01:47 here's this is one of the re one of the
1:01:49 reasons that I get so excited about this
1:01:53 [ __ ] is that
1:01:56 um hang
1:01:59 on happy anniversary AI Salon famam okay
1:02:04 one of the reasons I get so excited
1:02:06 about this stuff
1:02:13 is nobody learns from
1:02:19 history so back in the
1:02:23 80s some researchers at
1:02:28 Kodak created this thing called the
1:02:31 digital
1:02:32 camera and they took it into the powers
1:02:34 that be at Kodak and they said hey
1:02:37 bosses looky at the
1:02:40 future we got a
1:02:43 CCD and where film would be we put a CCD
1:02:46 there and look it made an image that's
1:02:49 like 32 pixels across ain't that
1:02:53 cool and the bosses at Kodak
1:02:59 went you don't understand what business
1:03:01 we're in son we sell
1:03:04 chemicals silver Halli is our game get
1:03:07 out of
1:03:09 here and then Kodak went out of
1:03:12 business
1:03:14 so Netflix has two
1:03:19 options they can say we're
1:03:23 Netflix we make more Academy
1:03:25 award-winning films than Hollywood does
1:03:29 or Emmy award-winning films than you
1:03:31 know Hollywood
1:03:35 does and they're going to laugh at that
1:03:38 [ __ ]
1:03:40 or they're going to acquire that or
1:03:44 they're going to hire someone to say you
1:03:48 go grind Runway TV into the [ __ ]
1:03:52 ground if you have to put us out of
1:03:53 business I don't want anything
1:03:55 resembling one of those things we're
1:03:57 going to we're going to win this
1:03:58 Market in order for them to win the
1:04:02 generative AI
1:04:03 Market they're likely going to have to
1:04:07 threaten their other
1:04:12 business so so why this is entertaining
1:04:15 is we're just going to get we're going
1:04:17 to get to watch in real time company
1:04:20 after company after company in every
1:04:23 industry
1:04:25 be confronted with some snot noosed
1:04:30 28-year-old some someone something
1:04:33 something or gen xer that walks into the
1:04:36 big boss's office and goes hey boss
1:04:39 here's the
1:04:40 future look at this GPT I just made that
1:04:44 does for free what we charge people
1:04:47 $150,000 a month
1:04:49 for why don't we sell these for
1:04:53 cheaper and when when the boss
1:04:56 goes you don't understand what business
1:04:58 we're in
1:04:59 son this is about time and materials get
1:05:02 out of
1:05:08 here down that company
1:05:11 goes and then you know and then you know
1:05:14 other ones will be like there'll be
1:05:15 Senior Management that says [ __ ] it we
1:05:17 got to turn on a dime those will be the
1:05:19 rare
1:05:21 ones and then you know and then there's
1:05:23 going to be a lot of small and medium
1:05:25 companies that start out executing large
1:05:28 companies start threatening them and
1:05:30 they'll either get bought or they will
1:05:33 um take
1:05:34 over they'll become new leaders in that
1:05:43 market time is it I gotta go it's late
1:05:47 people what what are you still doing
1:05:48 here why is anyone still here this I've
1:05:51 been completely useless tonight
1:05:58 I showed you how to watch TV I showed
1:06:00 you how to watch crappy three minute
1:06:06 TV do these videos take a lot of memory
1:06:09 no well they take a lot of memory to
1:06:11 produce not a lot of memory they take a
1:06:12 lot of GPU Cycles to produce which is
1:06:14 why they're
1:06:16 expensive um but the videos that are
1:06:18 produced are actually really small
1:06:20 because it's the technology is janky
1:06:24 right now
1:06:26 they're not generating 4K 60 frame a
1:06:28 second video let's put it that
1:06:32 way and occasionally a face will look
1:06:34 like a
1:06:39 face oh my God you can make money
1:06:43 with what are we going to do people 2024
1:06:47 is going to be [ __ ]
1:06:49 crazy I can't I I'm still reeling from
1:06:53 2023
1:06:56 what I don't what am I supposed to do my
1:07:00 brain can't hold
1:07:02 anymore I'm
1:07:05 cranky I don't want to
1:07:13 learn reminds me of the early flash days
1:07:15 totally like that my life is a never
1:07:19 ending AI field nightmare this is my
1:07:21 free
1:07:23 time
1:07:29 oh
1:07:31 man pretty cool is it free just joined
1:07:34 Runway TV I think is free or or you
1:07:37 talking about the AI Salon AI salon's
1:07:39 free um so how how we're funding it
1:07:44 being free is we just launched a merch
1:07:46 store so if you go to the salon. there's
1:07:49 a merch store but also you can sponsor a
1:07:53 month of of Mighty
1:07:56 networks it's 150 bucks it's not much
1:07:59 money and for the two meetings of you
1:08:03 know that that you sponsor you get
1:08:05 mentioned you're on the thank you page
1:08:08 um and so that's how we're keeping it
1:08:10 free for end
1:08:15 users a guy on Tik Tok has been blowing
1:08:17 up with Runway 4 second clips and
1:08:19 splicing together a few seconds um it I
1:08:23 don't know who's blowing up but Drew
1:08:25 who's in here a lot has been doing a lot
1:08:27 of that he's been taking images I don't
1:08:29 know where he's generating them I assume
1:08:31 on doll or mid journey and he's making
1:08:34 videos of his whole Channel I I run into
1:08:36 his stuff all the time so I don't know
1:08:38 who's who's blowing up but yeah this the
1:08:41 just watch the first 6 months of 2024
1:08:44 there's going to be a lot more noise
1:08:45 about this video stuff because it's
1:08:46 going to get it's going to get good fast
1:08:49 because it happened with images images
1:08:51 were laughably
1:08:52 bad for
1:08:54 two
1:08:55 years and this is what David Shapiro
1:08:58 talks about with this with this AI stuff
1:09:00 it's bad it's bad it's bad it's bad it's
1:09:01 bad it's bad it's bad it's good it's
1:09:03 good it's good it's
1:09:06 excellent so right now video's been it's
1:09:08 bad bad bad bad bad bad oh that's not
1:09:10 bad we're right there and then paa 1.0
1:09:13 is about to come out and you got motion
1:09:15 brush with with
1:09:17 Runway and then all of a sudden it's
1:09:19 just going to
1:09:22 go and what it'll start start out with
1:09:24 is it'll go from 4 seconds to 30 seconds
1:09:29 to 3 minutes to make me a
1:09:35 movie and the mo the mo the movie will
1:09:38 be narratively coherent visually
1:09:43 coherent there'll be a soundtrack that
1:09:46 will
1:09:47 be um musically
1:09:53 coherent
1:09:57 every industry it's not just movies
1:09:59 every industry this is coming
1:10:01 for congratulations on the anniversary
1:10:04 night Kyle thank you so much Robert
1:10:05 Rousy thank you thanks for coming to the
1:10:07 salon tonight and and hanging out here I
1:10:10 really do appreciate it I did see
1:10:12 something about Gemini demos coming this
1:10:14 month anyone hear that so what I heard
1:10:16 cam was Gemini I was supposed to launch
1:10:19 this week like a surprise launch and
1:10:22 then the CEO of Google killed because it
1:10:24 was bad at foreign
1:10:27 languages and then I heard they're going
1:10:29 to demo it this week which that would
1:10:31 make sense if you're going to if you
1:10:32 were going to sneak launch it and you
1:10:34 have events
1:10:36 scheduled and then you pull
1:10:39 it it would make sense to do a demo but
1:10:42 I it's it's all speculation and Pate in
1:10:44 here Pate who is quick to point out how
1:10:48 I don't know an SSM from an SL LM from
1:10:51 an llm from an spt t
1:10:55 P he will not tell us what's going on at
1:10:58 Google at the Google so take that put
1:11:03 that in your pipe and smoke it so I
1:11:06 don't know I have no idea but I but I
1:11:08 did hear they're going to be demoing
1:11:10 something this week my kids would love
1:11:12 that gold Gadget I don't know what
1:11:13 you're talking about you can make money
1:11:16 [Music]
1:11:17 with you mean this it's solid gold a
1:11:22 solid gold I got this from Parker on
1:11:24 Gold Rush I said Parker make me an Ingot
1:11:28 I'm going to make it into a button this
1:11:30 is an Ingot of solid gold from Parker on
1:11:32 Gold Rush he signed it it's signed on
1:11:34 the
1:11:34 bottom you can make money with J that's
1:11:38 Parker's
1:11:44 voice ah Pate cam Ken it seems very
1:11:48 close to coming out based on all the
1:11:50 speculative
1:11:52 articles Pate I think P needs a nap I
1:11:55 think Pate has been um TPU optimizing a
1:11:59 little too much
1:12:03 today to the Moon all right listen
1:12:06 everybody I don't know ssms either I
1:12:09 know ah the
1:12:12 [Laughter]
1:12:13 Google all right listen um it's good
1:12:16 seeing y'all um Happy First Birthday to
1:12:18 the salon
1:12:20 um I don't know I don't know when I
1:12:23 started this channel it was right around
1:12:26 this time we're probably actually coming
1:12:28 up on the the one year anniversary of
1:12:30 this but the first 6 months of this I
1:12:32 just did recorded videos I didn't do
1:12:33 lives the live started like the end of
1:12:36 April so I I don't I don't like this
1:12:39 channel for me is much more about the
1:12:41 lives now than the recorded [ __ ] anyway
1:12:44 good to see you
1:12:45 all um Sher bear good to see you human
1:12:48 being get rest Kyle thank you happy
1:12:50 birthday Salon I know happy birthday
1:12:52 Salon steveo thank you a i l TMZ that's
1:12:56 right all speculation all the
1:12:58 time you can go to other channels for
1:13:01 facts come here for the
1:13:02 fun come for the fun stay for the
1:13:06 speculation we'll make [ __ ] up if we
1:13:08 don't know the answer we'll answer it
1:13:10 just like chat
1:13:14 gbt oh I'm Punchy I'm a little Punchy
1:13:17 people you know what I mean Tyron truth
1:13:19 good to see you Tobias good to see you
1:13:22 thanks for coming to the salon tonight
1:13:24 steo Pate as always very nice Sher bear
1:13:28 uh Christopher Ramsey good to see you
1:13:29 scam scam
1:13:35 scam welcome to Chad add K katkin
1:13:41 jrc he's Punchy he's clueless but
1:13:45 occasionally entertaining all right
1:13:47 everyone have a good night I will see
1:13:49 you tomorrow
1:13:51 bye