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

Transcript

1. 0:11all [Applause] right all right I'm going live the hell am I doing what the hell
2. 0:25am I doing what the hell am I doing Happy First Birthday AI salon wait it look at
3. 0:40that oh wait you know where I can go here we can just zoom in on the little turtle Silver Fox did this cute little
4. 0:48turtle cutest little turtle ever curfew young man lunar stick I know I was I was watching the voice and I was like I'm
5. 0:57not going to go live and then I looked at and at the uh the Irregulars Channel and everyone's like where is he we
6. 1:03haven't heard from him like I'll go all right we'll do a short one we'll do a a short little teeny tiny one I
7. 1:16probably won't even demo anything we'll just go look at some we'll go look at look at stuff let's see look here's our oneye anniversary
8. 1:28logo and we got Let's see we have a merch store now for the
9. 1:46salon so that was pretty cool so I'll talk about that and what else I almost missed you I almost missed
10. 2:01me I almost didn't go live but then then I saw everyone like where is he we haven't heard from him damn it you just can't sneak
11. 2:14out added an image to Irregulars for this oh awesome hang on let me go
12. 2:28look
13. 2:43welcome everybody this is the AI learning lab I just got back from well I didn't just get back went out and had a oh nice
14. 2:52that's cool went out and had a uh little Afterparty after the AI salon and then I came home and hung out for a little bit
15. 3:07oh Lun stick all sad that the AI learning lab wasn't happening how was the party it was nice it was very nice we had that very
16. 3:20cute all right this is this is for you Lun stick we can we can have no crocodile tears we got to fix that
17. 3:31oh man uh it was good it was good we had we brought back um a lot of speakers who spoke spoken over the past year that
18. 3:41just talked about you know where they are with this stuff and you know what insights they've had and then we heard
19. 3:48from Salon members and it was just kind of a reminiscing it was just nice it was just a nice kind of connecting and
20. 3:56talking kind of meeting um so so wasn't wasn't super flashy which is nice was good um yeah and so I'll figure I'll
21. 4:06come on here and I mean you know I mean this is the kind this is the time of year I mean we're coming up on the
22. 4:12holidays we're coming up on the new year we just had the one year anniversary of chat GPT we just said I just you know
23. 4:18tonight's the oneye anniversary of the salon um and it's hard to put it in perspective it's it's really hard to understand
24. 4:31where all this stuff is so anyway do me a favor share the live even though I'm just going to go live for like an hour
25. 4:37here or whenever I don't know we'll just chat if you have any questions about AI actually you know what would be cool to
26. 4:43put in there rather than questions why don't I just continue the jam we had going at the salon anniversary I'd love
27. 4:51to hear you know for anyone who's been doing this for a while especially from the Irregulars you know what's either
28. 4:57what's a big Insight you've had or like where are you right now what's where's your head where's your you know have you reinvented
29. 5:07yourselves are you questioning everything are you I know lunaticks she's dusted off all of her index cards and every index card is now a
30. 5:19GPT uh so hey micro what's happening jery good to see you how was the party party was nice we did you know what's
31. 5:27cool about that group is it's like you know the party is just like a continuation of the conversation like we just go out
32. 5:35everybody grabs a beer some tacos and we just sit around just saying well what have you played with hey have you seen
33. 5:43this have you done that like it's just it's just it's like this it's like what goes on in here so it was
34. 5:49nice U what I saw of it was noise nice Drew glad you glad you get to see a little bit AI blocks me as a conspiracy
35. 5:58researcher yeah Joseph Baker you're going to have to go get yourself some uh some open- Source uh nonrestricted large language model action
36. 6:11going the safety guard rails and conspiracy theories do not go well together so so you're going to have to
37. 6:18you're going to have to geek out I'm afraid my friend here's the good news you can use the the restricted large
38. 6:24language models to to train you how to set up the unrestricted large Lang anguage models you can probably get it done in like an
39. 6:32hour I'm still M navigating my way through this AI thing but I'm so grateful to have stumbled into this
40. 6:41place well I'm I'm grateful that you've stumbled in here um if it makes you feel any better I said this tonight I said so
41. 6:50like I've been I've been fairly obsessed for 2 years I have been clinically obsessed for a year um you know I started started the
42. 6:59salon I started this Tik Tok Channel at 58 I guess I started it at 57 I have I have aged since I started this
43. 7:12um started a newsletter completely ramped my company's product road map are trying to figure out how to keep my company relevant in the face
44. 7:23of insane competition from infinite content generating machines and so like I feel like
45. 7:37I it's not really possible to spend more of my waking day than I do on AI like like you know if I were a
46. 7:51machine I could but as a human I I am pretty much tapped out every [ __ ] day jamming [ __ ] into my Cranium about
47. 8:02Ai and I said this tonight I feel as clueless as I've ever felt so don't think that just because you haven't been you know deep diving on
48. 8:17this for a year or two or 10 that that you're way more behind you're not way more behind everyone right now is trying to figure this out and
49. 8:31the thing the thing that's got me you know it's when you worry about the future it causes anxiety right the thing that's giv me a little
50. 8:40anxiety is I kind of know what's coming right like I know that Gemini is coming which is Google's chat GPT 4 competitor
51. 8:51that's going to be multimodal and crazy and hopefully good you know I know that grock is coming coming I know that you know these
52. 9:03video models like pico 1.0 looks like it's going to be absolutely [ __ ] insane um we've been playing around here
53. 9:11for the past couple of nights with sunno the music generating thing and it's it's like the um the core tool Suite so the the way I
54. 9:24like to think about this is like if you could only play with one AI tool if you only had access to one what would
55. 9:32it be the the answer to me is crystal clear right now it's chat GT4 period end of story nothing's even close but it's it's kind of like
56. 9:43professional tennis right you got jokovic Nadal and then everyone else you got in F1 you got for stappen everyone
57. 9:56else that's how AI is right now you got chachy p pt4 and everyone else the problem is the everyone else's are
58. 10:07rumbling they're getting really good really fast and so what it means is you can't you okay she's okay um you
59. 10:21can't ignore them but at the same time you can't keep up with them and and what's going to happen is the the tools
60. 10:29as they are right now so if you if you look at here let me show you is this it yeah okay you got chat
61. 10:41GPT right you're like okay yeah yeah we got chat GPT Kyle that's the big one great perfect so just if there if there
62. 10:49was only one you'd do that one right AI is my adviser and teacher very cool then you got these four Claude Bard co-pilot
63. 10:57so co-pilot is Microsoft's chat gb4 and you have grock grock isn't out yet but it's coming so we have to we have to
64. 11:03keep that on the list and it's going to be one of the big four so you have to pay attention to those four then I think it's worth you
65. 11:09paying attention to these four Pi perplexity Po and llama 2 now all of these for the most part are like chat GPT they started out as large
66. 11:19language models and now they're going multimodal Claude isn't yet um Bard is starting to go there with like Bard
67. 11:28right now one of the interesting things they just added is Bard knows all the YouTube videos so you can take a URL from one of your YouTube
68. 11:38videos put it in Bard and tell it to chapter eyesee it and it will because it knows them all um co-pilot from
69. 11:45Microsoft is multimodal now and and then they just added a thing called Deep search or something like that some new
70. 11:52thing I just I just saw it in passing tonight before we did the salon um then you got these you got Pi is the this
71. 11:59conversational thing that went from having one mode to having four modes and getting much more capable and there like
72. 12:05I have a feeling Pi is just going to explode and get multimodal perplexity is is an increasingly complicated
73. 12:13sophisticated research tool po you can play with like I don't know a dozen or 15 different large language models you
74. 12:20can make Bots like it's it's it's big and then llama like I've got labs. perplexity here that's just a place to
75. 12:26play with it llama is part of a whole world of open-source models like thousands of projects right so if you want to if you want to dig
76. 12:37into large language models there's a whole world just in that little thing right there so that's just the wordy
77. 12:43words right that's just words I don't even have Amazon they just launched Q so so that feels a little overwhelming
78. 12:55and then you start looking at um the image generation tools they're getting more and more and more sophisticated and
79. 13:02faster and and like even the image generation tools like in the past two weeks two weeks we've gone from you type a
80. 13:12prompt and you hit a button and then it the the clock spins and then you get back a image or four images and then you
81. 13:20upscale them right to these models where you now draw a shape and it read it renders the new image in real time we're
82. 13:30going from non-real time image generation to realtime image generation there's a thing called Kaa which I actually signed up for um that
83. 13:40does I can take my webcam as input type in a prompt and then it turns me into some other thing in real time so even
84. 13:47these things are how you use them is anxiety inducing because if you've gotten good at doing it with the
85. 13:55wordy words or with dolly3 with with you know with with you know doing a uh an image prompt within the context of a
86. 14:02chat how you do these is going to get different it's changing then you got video this world is losing its [ __ ]
87. 14:09mind these tools are getting so incredibly powerful that I don't want to ignore them I don't want to not do them but
88. 14:19making movies is hard my company storyvine exists because video storytelling is hard and I thought there's got to be a better way
89. 14:29and so storytelling doesn't necessarily get easier with these the production of the bits gets easier but you still got to tell the
90. 14:38story why does AI I AI have a so much of a problem with spelling words on images the reason it's got a problem so in
91. 14:48general the reason AI has problems with accuracy and predictability and consistency is that the way the
92. 14:56Transformer model Works which is the the the core underlying technology underneath it how do I explain this super
93. 15:04simply what that model does is it turns words or pictures or music any sort of data but let's just say words let's just say like chat GPT
94. 15:14words it takes all the words that it is trained on which is you know trillions of them billions of documents like as
95. 15:21much of the public internet is they could get their hands on and it turns them all into um strings of numbers that
96. 15:30live in Thousand dimensional space and you're like what yeah exactly it's weird it's weird and so when you
97. 15:41type a prompt in chat GPT and it writes you that big ass response and it looks like there's like a really smart person
98. 15:49on the other side of it it's really just a calculator that is calculating what is the next most probable word after this
99. 15:58string of of words you're prompt and then what's the next most probable word after that and the next most probable
100. 16:04word after that that's that's technically what it's doing and it's it doesn't even know the words it's putting
101. 16:08there it's looking in its mathematical Universe plucking out a number and then that number or a series of numbers a
102. 16:18vector that Vector is associated with a word so it goes I need this set of numbers and it pulls it down and says
103. 16:24okay that set of numbers goes with the word dog and it puts that word in and so it just spits all that [ __ ] out and it
104. 16:32doesn't know what it wrote so if if the if it's sort of pulling from training data that's got inaccuracies in it it's
105. 16:39going to give you inaccurate stuff the image things are the same way it doesn't know what it's creating when when these
106. 16:46tools create the images they create they don't go back and look at them and analyze if they got them right that will
107. 16:53be coming like 2024 I guarantee you we're going to see that where we're going to see see image models that generate the
108. 17:01image analyze the image to see if they got it right and keep iterating until they get it right and maybe even fixing
109. 17:07just pieces of it I guarantee you we'll see that in 2024 so as of 3 months ago generative images couldn't do words
110. 17:18at all so the fact that they're doing words even remotely good is remarkable let me go look at let's see
111. 17:29like there there's a good one that that Sherry did for the for the AI Salon happy is right happy anniversy sorry AI Salon it got almost
112. 17:40everything right I don't know what you're complaining about but I mean it's pretty remarkable that it got it that
113. 17:46close but all it's doing is it's taking a bunch of images sort of smashing them together as as this massive pixels and
114. 17:54then denoising them and then what it generates it doesn't even know what in there so that's why it does that why
115. 18:01doesn't it work for foreign languages uh I I when was the last time that you looked at it what why doesn't what work
116. 18:10for foreign languages hold please all right this is chat GPT this is the chat GPT app I'm putting it in conversation
117. 18:25mode I want you to act as a Japanese translator I'm going to speak in English and I want you to translate in Japanese and don't repeat the
118. 18:43English the AI Salon had its birthday tonight and I'm tired but very very happy
119. 19:02[Music] come on chat jpt we don't have all
120. 19:19night now on second thought I want that in German please
121. 19:36how about Montreal French so they do do languages it knows all the languages it knows all the programming languages it knows all the
122. 19:51languages they're translating Sanskrit uniform writings into English by just taking snapshots of [ __ ] like it it knows all the stuff
123. 20:06now the nature of where this technology is as of today it's got a lot of the technical term for it that we
124. 20:16talk about a lot on this channel is janky pieces of [ __ ] all of this technology right now will get you to about 80%
125. 20:25of good and then the last mile the last 20% oh it's rough but just watch 2024 A lot changes just you're going to see they they're
126. 20:38going to start putting they're they're going to start this is what qar is to my understanding you've got chat GPT down
127. 20:43here they got this thing called qar which is this um game mechanics training system basically they they've merged two
128. 20:51Training Systems together that basically say okay chbt you went and you took an action I'm going to look at that action
129. 20:58and give you a score on how well you did and and the score is either going to go up or it's going to go down and if it
130. 21:03goes down then you got to do it better and if it goes up then you keep going in that direction so you're going to start seeing in
131. 21:102024 the typos are going to get way less the hallucinations are going to get way less these things are going to get way
132. 21:18better and all of the stuff people are bitching about right now it doesn't spell right hallucinates are all of a sudden going
133. 21:25to get dramatically good dramatically David Shapiro talks about this he goes the the the thing about machine learning
134. 21:32and this kind of technology is it's bad it's bad it's bad it's bad it's bad it's bad it's bad it's okay it's okay it's
135. 21:39excellent we're we're still in the it's bad it's bad it's bad it's getting better it's getting better it's like
136. 21:44we're we're just like there and we're about to go if qar the the rumor is that there's a thing called qar that they that they
137. 21:56put together at open AI and it made some sort of advancement and the rumor is that's why they fired Sam Alman cuz [ __ ]
138. 22:02was moving too fast and the nonprofit board said ah if that's true shit's going to get faster fast or better fast what can you
139. 22:13tell us about Tree Learning when it refers to AGI um let me go see if I can find that David Shapiro video hold please you mean q learning
140. 22:28I think it's Q learning so qar is a combination of a-learning which I think is tree logic and a star and Q learning qar
141. 22:43um let me go see if I can find this video David Shapiro if you if you're interested in Ai and you don't follow David Shapiro you are
142. 22:56missing you are missing stuff by the way his his um AI predictions in 2024 video is really good um bigger than you
143. 23:11think that wasn't it it was what is qar hang on what is qar you know a lot of people
144. 23:26think that in order to conceal good morning today's video is brought to you by insomnia that's not a product that's
145. 23:38just I woke up at 3:00 a.m. today and I woke up to see some really cool news so okay so let let me show you a couple of
146. 23:45things thought Chain of Thought okay so so there's there's this this uh graphic here so how do I explain this so so this
147. 23:57is this is the this is the logic behind qar so you've got a starting State and the starting state is I don't know I
148. 24:05want to find the best prices on travel or whatever it is and then you have the ending State like here's your goal your
149. 24:11goal is to go there and then what happens is given this starting State the the you're you're going to um take an
150. 24:26action and you're going to take action one in state one and the combination of the action you take in that state is
151. 24:34going to give you the quality so Q is stands for quality and then what happens is so so given this state maybe you
152. 24:41maybe there's four actions you could take you could you know launch a Google search you could whatever it is you
153. 24:47could you could do things and the way qar works is it says okay you took that action I'm now going to look at the
154. 24:53results of that action I'm going to score it I'm going to give you a plus 027 or a minus point4 right depending on
155. 25:02the dramaticness of how bad you did and then we're going to move on and then given the new state we're going to move
156. 25:09on to state two and and for state two you got a set of actions and then you take an action and I judge that and I
157. 25:17add you know I either add or subtract to your Q score and then you just basically keep going around so that's that's just
158. 25:24to understand the kind of math behind it and then there's this other drawing here that he does where so you're sort of you start
159. 25:32down here and and the state is the little red dot in the middle and then you could take you know one of nine
160. 25:38actions around it so it takes this action your goal is up here and and things are going good you know the the
161. 25:44your Q rating is going up it's going up it's going up and then you get here and something's wrong right and your Q score
162. 25:50starts going down so it says okay I think I need to take different actions and so it takes a different action and
163. 25:55it basically learns as efficiently as possible to navigate itself around whatever the you know logistical
164. 26:03roadblock is and it and it gets here so let's say um the the what they apparently used qar on like three weeks ago four weeks ago
165. 26:15was simple grade school level math problems and what normally happens with large language models is they can
166. 26:24fine-tune them and fine-tune them and fine-tune them and maybe they can get them from like 70% correct to 80%
167. 26:31correct to maybe 85 maybe 90 but they can almost never get to 100 well they apparently got to 100 because if it got
168. 26:37the problem wrong at some point or some part of the problem wrong it would sort of navigate it its way around it and and
169. 26:45it would get the problem correct that's reasoning that's what chat GPT and these things can't do right now all they are
170. 26:56right now is dumb calculators they're dumb prediction engines they're the smartest dumb calculators I've ever
171. 27:03experienced in my life like they're beyond anything I've ever seen but they're not reasoning this [ __ ] is reasoning now I
172. 27:11don't know if that's exactly what you're talking about qar is just speculation yeah that's right Pate Pate Pate is
173. 27:16keeping me honest here Pate works for Google qar is totally speculation but it it's going to be some version of this
174. 27:24right it's going to be some version of look at the [ __ ] you did look at the action you took did you get it right if not try again
175. 27:36uh how can I make mid-journey type of word use do you can't right now mid Journey doesn't do words they will probably come
176. 27:46out with something in the next two or three months that does it because you can now do words in stable diffusion
177. 27:52which you couldn't do before her Herod who's in here a fair amount wait is that just an image of an a pathfinding
178. 27:59algorithm yeah this is a part of the video where he's talking about AAR AAR so I don't I don't I don't know he
179. 28:07describes in here something about how qar is different than a star was but that's as much as I know you
180. 28:15probably know more than I well the fact that you said wait isn't that just a diagram for AAR says that you know more
181. 28:21about this than I do Pate by the way Pate is the Guild Master of the um AI mechanics Guild over in the AI Salon
182. 28:31where we just had our first year birthday tonight and um so if you want to like hang out with people that know a lot
183. 28:41about this stuff join the salon so here's the if you go to the salon. um we just updated the link tree there's a link to the mighty networks
184. 28:50Community there's a link to the meetups we meet every other week our our meeting was tonight and then there's a link we
185. 28:55just added a new merch store um and we've got a one-year anniversary um special edition t-shirt it's only
186. 29:02going to be up for 48 hours so that's the one with the cool graphics in it it looks like it looks like that it's
187. 29:15cool it's a cool graphic see the one and the A and the I and the road and the Sun and the you know um yeah all right Right Moving On Moving
188. 29:30On We're moving on up to the east side to a Delux apartment in the sky yeah we're moving on up all right new chat limit per hour
189. 29:43made me upset on gp4 oh the their rate limiting I know they're their problem is their problem is too many people are using their [ __ ]
190. 29:54cuz it's so good and everyone's addicted to it now anytime someone gets rate limited I get like 10 pings I just got rate [Laughter]
191. 30:06limited I don't know what to tell you people all I can tell you is keep stay curious keep playing keep building [ __ ] assume it's going to
192. 30:15break assume that the tool that you're using that seems like it is a miracle is just a janky piece of [ __ ]
193. 30:25just you you got you have to accept those two realities this is the most miraculous thing I've ever seen and and
194. 30:32no one should use this software both of those are true where's the Llama I got llamas I got llamas in here
195. 30:42hold uh uh uh there llamas llama llamas you know why I like llamas llamas is tasty M llama he said qar will be a combination
196. 30:56of a star and Q learning well that's the again as P pointed out that's the speculation they have not confirmed that now in an in an
197. 31:06interview with I think Vanity Fair last week Sam wman did say that the letter to the board about said algorithm was a
198. 31:19leak he confirmed that it was a leak um but we don't know any more details other than that he said it was an unfortunate leak
199. 31:29um will we be homeless next year I do not think so I do not think so so one of the things that's very easy to fall into
200. 31:37is the robots are going to kill us um I think there's going to be a lot of disruption I think there's going to be a
201. 31:46lot of job loss I think there's going to be an adjustment period and then I think as these tools get more sophisticated
202. 31:53and we learn how to wield them better there's going to be new kinds of jobs created um I think project managers are going to be in demand I
203. 32:10think that um people who have software experience are going to be in demand because what AI is doing is
204. 32:20democratizing software development and so what the the the the beauty of that is lots of people get to make applications the absolute [ __ ]
205. 32:31disaster of that is lots of people get to make applications and they don't necessarily have the discipline of say
206. 32:38like QA testing and planning all of the things you know keeping track of different versions and the thing was
207. 32:46working good and then they made some changes and now it's work not working good if you don't have a good system in
208. 32:51place you can't go back right Vanity Fair is the go-to publication from my text news P's feeling saucy
209. 33:02tonight I don't it might not have been Vanity Fair which I don't know what it was but whatever it was it was some
210. 33:08interview with Alman where he said that the uh the leak was an unfortunate it was an unfortunate leak oh
211. 33:18man it's not going to be the robots AI can just use existing things against us I that's really true save save on
212. 33:25Manufacturing it's so true yeah the robots when the robots you know what it's going to be the robots are going to
213. 33:32just toy with us like a cat for a while they're be like they're cute they're just gonna go
214. 33:47flick got my t-shirt confirmation awesome AI Bob that's great fantastic um The Jeffersons yeah moving on up to the [Music]
215. 34:04side hi I just got here let me tell you about [Laughter] qar oh what did I miss tonight uh you missed
216. 34:16a birthday party we had a birthday party for the AI salon and I just came on um 20 minutes ago 30 minutes
217. 34:28ago uh I'm just as much fun at virtual parties as real ones he's got all the
218. 34:43friends oh man any updates on Gemini not that I saw before I was PL I was planning for the salon all day we were we were putting
219. 34:53the store together and the um the link tree together and the what's it called Mighty networks together we were
220. 35:02editing Mighty networks like we were just trying to get all that because that's that's now live so if you go to aith
221. 35:09salon. and go to the community site That's mighty networks it's now public so um you can just go join that have you
222. 35:16read about Mamba apparently it's an SSM versus an llm a small a small what model an you mean an slm small language model
223. 35:25versus a large language model um I've heard of Mamba I haven't I haven't um played with it um Microsoft on dev day on open ai's Dev
224. 35:40day Microsoft talked about being into small language models which basically is to say open AI ain't the only game in town and then a
225. 35:52week later Sam Alman got fired um so so yeah I think we'll probably hear something about is small language just slang no
226. 36:04Lun the I Pate may know this better than I do but small language models are just basically tighter trained they're not
227. 36:10trained on massive amounts of language they're they're trained on specialty pieces of of language so so you don't
228. 36:17have as much hallucination if you're training things on smaller data sets basically it's like I what I don't know
229. 36:24is I don't know the difference between a small language model and say an expert um AI system from seven years ago like
230. 36:34like maybe just because it's the Transformer and it's a Slicker version of it I don't know that I do not know supposedly it
231. 36:44doesn't have the limitations of the large language model yeah but it also doesn't have the general capability of
232. 36:48the large language model so everything's a trade-off okay I get it I just didn't know it's slm I think I think paint is it slm or S
233. 37:03SSM so an slm so would slm be better at doing a particular type of scientific research yeah I think so oh SSM is a unique method it's not
234. 37:16slm see you know what we're it's it's almost 11:30 I'm exhausted I am significantly out of my depth here so I don't [ __ ] no uh go to
235. 37:28the AI Salon join the mechanics Guild and and hang out with Pate he will tell you all this [ __ ] he can also teach you
236. 37:37the difference between a TPU a GPU and a CPU in in like a short amount of time and you actually get it it's pretty cool
237. 37:47yeah I don't know much about it but we can figure it out together and that's the whole idea of the salon if if you
238. 37:52haven't been to the salon it's it's a really cool group of people it's it's very much about everyone's trying to figure all this [ __ ] out
239. 38:00um we're all trying to figure it out nobody [ __ ] knows nobody [ __ ] knows nobody knows nobody knows even if
240. 38:08you understand how the [ __ ] Works technically okay so now you know the difference between SSM slm and llm how does that benefit your pizza
241. 38:18shop I don't know that's the [ __ ] we got to figure out but I'm a gardener what do I do with this stuff hang out in the salon hang out with
242. 38:30people that are looking at things like you're not State space model SSM State space model oh you know what we can do
243. 38:46duh oh chat GPT why are they asking a man with an acting degree about an SSM be because he's a mansplainer he talks
244. 39:01about things as if he knows about them even when he doesn't this is a man [Music] spainer he is known Cally as
245. 39:19Kyle um can you explain the difference between an SSM an slm and an
246. 39:33llm in the context of AI as if I had the education of a [Laughter] seven-year-old sure imagine you have two
247. 39:47different kinds of smart robots one is an SS wait SS SM NSL good Lord which which stands for small scale model but that's not what
248. 40:00wait what was the thing that you said something State simple State what was it small scale wait I think um SSM stands
249. 40:16for small state something small state machine
250. 40:31I don't know I can't I I don't know I don't even know enough ask chat gbt to tell us what's the
251. 40:47difference this is what it's like sometimes you know what I can do you can make Mone with [Laughter]
252. 41:02next moving on moving on up to the next topic the Simons happy birthday AI
253. 41:16Salon all right I'm going to do like another 15 minutes if the next 15 minutes go as good as the past 15
254. 41:23minutes woo what a night it's relatively knew it's complicated it's part of a push to get past Transformers oh that
255. 41:30makes sense cuz every everyone's like well seems like open AI is won on the Transformer game why don't we invent
256. 41:38something new maybe we can beat him that way Kyle can you show us Runway that I can show you do you want to see Runway
257. 41:44TV or Runway ml cam cam cat cam cat can cam K him from Cleveland you want to see the TV show or you you want to see the TV network
258. 41:57that's going to put Netflix out of business if they ignore it or do you want to see something [Music] else State space model let me go back
259. 42:13and ask that state space model look it up what is it doing research with Bing
260. 42:28okay all the letters single short [Laughter] message uh can you show us Runway Tobias has a birthday this weekend
261. 42:40awesome wait State space model is a type of mathematical model used in various Fields such as control engineering
262. 42:47Finance economics ecology to simulate Dynamic systems basic concept a state space model represents physical systems
263. 42:56by defining a set of inputs outputs and internal States well that's kind of what we were talking about with the qar
264. 43:03thing these internal states are key variables that describe the system's Behavior at any given time yeah the
265. 43:11relationship between the inputs the outputs in the states is typically expressed as a first order differential
266. 43:16or differential equations you know them first order differentials I I'll tell you what I you know I was at a party
267. 43:23once I was talking about first order differentials and I was like like man the wait till you see the second
268. 43:31order different differentials in the it's in the transmission right because those the the first order goes fast but
269. 43:39then you put it in the second order and the differential spins and the tires just grab and so I know I know math
270. 43:46pretty good I know it pretty good I'm it's I got a solid handle on this so yeah come back here AI learning lab
271. 43:53you'll learn good stuff f fantastic math stuff here at the learning lab that's what we're all about [Laughter]
272. 44:09qualifications a sure way to get the babes yeah listen they uh saw you standing there by the punch and I was
273. 44:16thinking you know I I was just thinking about first order differentials and uh you're in math let [Music]
274. 44:31I've been at parties where order of differentials are discussed at length it was at a government lab P that's not
275. 44:38called a party Pate do you know what a party is dude we're gonna we're going to start a different Guild and we're going to invite you to
276. 44:45[Laughter] it you can't discuss differentials at a party it's no no no that's not a party that's just work
277. 45:00with alcohol a parties where you act like an idiot and forget what first order differentials are is it differentials yes differential
278. 45:17[Laughter] equations so then I says to her how do you handle your hidden States and she says to me it's a first order [Laughter]
279. 45:42thing Runway okay um I was nearly late for the salon cuz I got my hand bit by a pitbull no at the dog park oh no are you you okay oh
280. 45:57that's [ __ ] awful I'm sorry about that W was it's not good pimples and hands that's horrible horrible
281. 46:14horrible okay someone wants to go see is we're going to go over here we're going to go to Runway ml so Runway so
282. 46:23let me let me show you something first okay we're going to go here we're going to go here we're going to go here is that it
283. 46:31no no no yes okay this this is the world of video right now well no the world of video is probably way bigger than this this is
284. 46:43the world of video that I've played enough with that I think and heard enough about so generally how I pick the
285. 46:52tools that make this list or like all the stuff that I recommend is if I hear about something once or twice I don't
286. 47:02care if I hear about something 10 times I'll go check it out I will make a snap judgment on it because that's how my brain works I'm
287. 47:13a synthe I'm I got I got ADHD I don't have time for your [ __ ] nonsense your I don't have time for your [ __ ]
288. 47:19manual I will go look at it and I will snap judge it and then if in snap judging it it's okay then I'll go look at it
289. 47:30um most stuff does not even come close to making this list so what's on this list are things that I've actually
290. 47:38played with or enough people have talked about and I've seen output from like okay that that's one you got to pay attention to
291. 47:44so the three sites on the left or yeah the three projects on the left are are all the same basic thing Runway ml Moon
292. 47:54Valley and pabs Moon Valley seems to be drifting into the distance a bit P Labs is about to sort of take off in front of
293. 48:02uh Runway ml but Runway ml right now's got the lead Runway ml is for um text to video like you know chat GPT is to large
294. 48:12large language models it's kind of got the lead right now and people are coming after them um but there's a lot of people that I
295. 48:21know like like um um some of the creative directors I know Kenny that was on the salon tonight is one of them
296. 48:28likes P Labs better and Pika 1.0 is about to come out but all those three things on the left are like you type in
297. 48:35text and it makes a video or you upload an image of a video and it makes video why they're getting better is they're
298. 48:43able to do longer chunk chunks of video and they're also starting to give directorial control so you can zoom in
299. 48:50on an image you can pan on an image with Runway ml now you can there's a thing called motion break brush where you can
300. 48:56paintt on an image so so I just wanted to give you the context of what what we're looking at oh and then the reason
301. 49:02I asked about Runway TV this thing is worth checking out if you go to runway. TV it's going to look weird and you're going to
302. 49:12see films there videos there that look weird and if you're old like me you may remember in 1981 this weird TV station
303. 49:25came out called MTV Music Television and nobody knew what to do with it it was weird it was like we we're used to listening to music and now
304. 49:35we're watching music what is this that's kind of what Runway TV is it's videos that are created out of this text to you
305. 49:44know text or image to video Generation stuff and they're weird but they're going to get very real very fast so pay
306. 49:50attention to that thing okay so let's go to Runway [Music] ml we're going to log in all these video tools are not cheap they charge you by
307. 50:08the cred they well well they give you they they're there're it's kind of like Vegas so you know when you go to Vegas
308. 50:14and you've got like I don't know $1,000 and you're like this ,000 is a lot of money and I know what $1,000 is
309. 50:23and they give you these little these little things chips and then you get like you know 10 chips and all in a sudden it doesn't
310. 50:34seem like a lot of money and then you just go play with those that's kind of what these these companies do they they
311. 50:40uh they give you tokens and they give you like 150 tokens you're like I'm going to make 150 videos and then you go
312. 50:47to make a video and they're like this video is going to cost you 15 tokens and you're like oh [ __ ] that's a lot of money money so
313. 50:56anyway that's what this is like so the interface here is not super intuitive but you can uh you can create
314. 51:07a video from text from an image or from an image plus a description so I'm just going to start with an image and so I'm
315. 51:14going to say upload a file so we're going to go find an image to turn into a movie um where am I going to go I'll go
316. 51:22here I got a lot of stuff here [Music] that could be a cool one let's grab well that's vertical let's
317. 51:40[Music] see yeah let's do this we'll do this one okay so I'm going to open that even though it's vertical it'll make a
318. 51:55vertical vertical VI and then I can do things like I can say um how much motion do I want and it's a
319. 52:05slider from 1 to 10 and then I can click on this camera motion Tab and there's all these different camera controls there's
320. 52:15horizontal um there's there's panning there's tilting so tilting is up and down panning is left and right
321. 52:23horizontal is like tracking the camera's doing this panning is it's doing this it's rotating and then there's
322. 52:31rotation um so let's just say We'll zoom out on this and that's it and then we're going to go to this thing called motion
323. 52:45brush and so now I've got a brush with this purple thing on it and I'm just going to brush all over the leaves and maybe a little bit of the
324. 52:59branches and I'm just G to kind of purple my way around this thing and you don't have to be really
325. 53:13precise these things aren't precise they're kind of what did I what was the term I used oh I know janky pieces of
326. 53:21[ __ ] so these are all janky now listen if anyone in here or works at Runway or P laabs or open AI or whatever and
327. 53:29you're like he K our stuff janky pieces of [ __ ] it's it's it's a term of endearment I love you you you're good
328. 53:35people and I'm going to make some leaves down here move maybe all right and then what do I do there's proximity see I don't you you you have
329. 53:46to do something with the so I guess we'll go vertical down so maybe we'll make some leaves fall I don't
330. 53:54know and then I'm going to save that and then we're going to generate a movie Bang so you start with a picture you click on some
331. 54:04[ __ ] and you H generate video you can also type in words um I want a picture of a an artvark skateboarding and it'll make that and
332. 54:18then it takes a while my experience with these is that because I don't do it every day I'll occasionally come in here and do what I
333. 54:27just did where I just kind of stumble into an image and just kind of do it I get pretty shitty results with these
334. 54:33people that spend time doing this proximity will come closer to you yep okay cool um people that spend time with
335. 54:41these can get decently predictable results it's like anything if you use the tool and get better with the tool
336. 54:46you'll be gooder the 4C limitation is a bit annoying yeah but Jerry so um with Runway now you can extend end a video
337. 54:56four times so you can go up to 18 seconds long so what it what it does it's actually kind of cool it'll
338. 55:02generate a 4 second clip and then if you say extend it for 4 seconds it takes the last frame of the first clip and that
339. 55:09becomes the starting frame of the next clip and it you can do that four times so you can create 18 second sequences
340. 55:16the problem is they lack coherence right which is basically this tree at the end of 18 seconds would probably turn turn
341. 55:25into like a Rob red breasted Robin or something um is this our movie where's our movie oh yeah there you go look it's
342. 55:38working we got decent results it erased all the letters and down come the leaves that's not bad so that's Runway doesn't have any
343. 55:52audio it's just it's you know this is not fully edited video this is just animation right this is this is motion
344. 56:00sequences but if you go to runway. TV you can see what people are doing with this and and what they're also what
345. 56:08the clever storytellers over at Runway TV are doing is they're mixing traditional footage with digital footage
346. 56:16with AI footage and they're kind of you know munging it all together with After Effects and stuff like that I sometimes
347. 56:24expand to 18 seconds but you can speed up the frames to say 9 seconds oh yeah interesting cool oh yeah you've been
348. 56:31doing a lot of this Drew right you've been you've been doing a fair amount of this so but go experiment just know that your
349. 56:42experiment's going to cost you about 30 bucks and you're going to get like you know 15 shitty videos you be like wait that's
350. 56:53it but what's the alternative go get a camera go find a tree the right time of year and film some leaves falling out of it that's your
351. 57:05[Laughter] alternative oh man can we see Runway TV absolutely okay so this is how you make the piece parts and and to to submit
352. 57:18something to Runway TV you have to have made something in Runway ml which makes sense [Music]
353. 57:33runway. so you start out that that thing very much reminds me of the the early MTV stream offline all right
354. 57:50fine that's normal footage oh maybe it's not [Music]
355. 58:06[Music] and I don't know what this is it's something but if I go to TV Guide here this is where it gets this gets really
356. 58:15reminiscent of like early days of Netflix early days of a lot of the streaming services were kind of these
357. 58:21sort of scrolling lists of videos right in in generally you know um curated channels AI Film Festival gen 48
358. 58:31whatever that is you know maybe they partnered with someone called gen 48 this is the staff pick film of the week
359. 58:38um the creative partners program Runway Studios Runway Studios Runway Studios pay attention to this who's Runway
360. 58:47Studios I don't know they took a pile of the you know hundred million they got invested or whatever it was and and they
361. 58:54said hey why don't we try to make some original content let's go get some filmmakers we'll give them our tools and
362. 59:01we'll just tell them to go make original content so that's what that channel is um you can submit your work to this
363. 59:08channel because we're so early just anyone can do this like this video is just weird it's cool but it's just
364. 59:23weird music video right does this feel like early MTV days this was all hand animated back
365. 59:42then um so there's that kind of stuff um endless Reflections animation worldwide Couture 24
366. 1:00:16that's pretty slick so so it kind of feels like art art films of the 70s like the Andy Warhol stuff there's some in here that are true story
367. 1:00:25telling and what's going to happen just mark my words this is going to start to get a little bit of Buzz or something
368. 1:00:32like it it doesn't have to be this one like this this is just the first one I've seen of these and real filmmakers are going to
369. 1:00:41piss all over this they're going to be like it's not real film that's stupid real film's better and then what's going to happen is for a
370. 1:00:51while there's going to be there's going to be be two different tracks and one's going to be a really thin track which is
371. 1:00:58this generative stuff and the other one's going to be Hollywood and then just over time that really thin track is
372. 1:01:04going to get wider and wider and wider and who are the power players in the generative side are not necessarily
373. 1:01:11going to be who the power players are on the Hollywood side and there's going to be a really interesting battle so just
374. 1:01:15pay attention to this stuff cuz right now this stuff is I wouldn't call it laughably bad but it is janky
375. 1:01:25this a lot of this work is the result of janky technology that made it possible but it's going to get real good real
376. 1:01:34fast that's me at the differential party what will happen with Netflix so cam that's so okay so here's this is one of the re one of the
377. 1:01:49reasons that I get so excited about this [ __ ] is that um hang on happy anniversary AI Salon famam okay
378. 1:02:04one of the reasons I get so excited about this stuff is nobody learns from
379. 1:02:18history so back in the 80s some researchers at Kodak created this thing called the digital camera and they took it into the powers
380. 1:02:34that be at Kodak and they said hey bosses looky at the future we got a CCD and where film would be we put a CCD
381. 1:02:46there and look it made an image that's like 32 pixels across ain't that cool and the bosses at Kodak went you don't understand what business
382. 1:03:01we're in son we sell chemicals silver Halli is our game get out of here and then Kodak went out of business so Netflix has two
383. 1:03:18options they can say we're Netflix we make more Academy award-winning films than Hollywood does or Emmy award-winning films than you
384. 1:03:31know Hollywood does and they're going to laugh at that [ __ ] or they're going to acquire that or they're going to hire someone to say you
385. 1:03:47go grind Runway TV into the [ __ ] ground if you have to put us out of business I don't want anything resembling one of those things we're
386. 1:03:56going to we're going to win this Market in order for them to win the generative AI Market they're likely going to have to
387. 1:04:06threaten their other business so so why this is entertaining is we're just going to get we're going to get to watch in real time company
388. 1:04:20after company after company in every industry be confronted with some snot noosed 28-year-old some someone something
389. 1:04:32something or gen xer that walks into the big boss's office and goes hey boss here's the future look at this GPT I just made that
390. 1:04:43does for free what we charge people $150,000 a month for why don't we sell these for cheaper and when when the boss
391. 1:04:55goes you don't understand what business we're in son this is about time and materials get out of here down that company
392. 1:05:10goes and then you know and then you know other ones will be like there'll be Senior Management that says [ __ ] it we
393. 1:05:16got to turn on a dime those will be the rare ones and then you know and then there's going to be a lot of small and medium
394. 1:05:25companies that start out executing large companies start threatening them and they'll either get bought or they will um take
395. 1:05:34over they'll become new leaders in that market time is it I gotta go it's late people what what are you still doing
396. 1:05:48here why is anyone still here this I've been completely useless tonight I showed you how to watch TV I showed
397. 1:05:59you how to watch crappy three minute TV do these videos take a lot of memory no well they take a lot of memory to
398. 1:06:10produce not a lot of memory they take a lot of GPU Cycles to produce which is why they're expensive um but the videos that are
399. 1:06:18produced are actually really small because it's the technology is janky right now they're not generating 4K 60 frame a
400. 1:06:28second video let's put it that way and occasionally a face will look like a face oh my God you can make money
401. 1:06:43with what are we going to do people 2024 is going to be [ __ ] crazy I can't I I'm still reeling from 2023
402. 1:06:55what I don't what am I supposed to do my brain can't hold anymore I'm cranky I don't want to
403. 1:07:12learn reminds me of the early flash days totally like that my life is a never ending AI field nightmare this is my free time
404. 1:07:28oh man pretty cool is it free just joined Runway TV I think is free or or you talking about the AI Salon AI salon's
405. 1:07:39free um so how how we're funding it being free is we just launched a merch store so if you go to the salon. there's
406. 1:07:49a merch store but also you can sponsor a month of of Mighty networks it's 150 bucks it's not much money and for the two meetings of you
407. 1:08:02know that that you sponsor you get mentioned you're on the thank you page um and so that's how we're keeping it free for end
408. 1:08:14users a guy on Tik Tok has been blowing up with Runway 4 second clips and splicing together a few seconds um it I
409. 1:08:23don't know who's blowing up but Drew who's in here a lot has been doing a lot of that he's been taking images I don't
410. 1:08:29know where he's generating them I assume on doll or mid journey and he's making videos of his whole Channel I I run into
411. 1:08:36his stuff all the time so I don't know who's who's blowing up but yeah this the just watch the first 6 months of 2024
412. 1:08:43there's going to be a lot more noise about this video stuff because it's going to get it's going to get good fast
413. 1:08:48because it happened with images images were laughably bad for two years and this is what David Shapiro
414. 1:08:58talks about with this with this AI stuff it's bad it's bad it's bad it's bad it's bad it's bad it's bad it's good it's good it's good it's
415. 1:09:05excellent so right now video's been it's bad bad bad bad bad bad oh that's not bad we're right there and then paa 1.0
416. 1:09:13is about to come out and you got motion brush with with Runway and then all of a sudden it's just going to
417. 1:09:22go and what it'll start start out with is it'll go from 4 seconds to 30 seconds to 3 minutes to make me a
418. 1:09:34movie and the mo the mo the movie will be narratively coherent visually coherent there'll be a soundtrack that will be um musically
419. 1:09:52coherent every industry it's not just movies every industry this is coming for congratulations on the anniversary
420. 1:10:03night Kyle thank you so much Robert Rousy thank you thanks for coming to the salon tonight and and hanging out here I
421. 1:10:10really do appreciate it I did see something about Gemini demos coming this month anyone hear that so what I heard
422. 1:10:16cam was Gemini I was supposed to launch this week like a surprise launch and then the CEO of Google killed because it was bad at foreign
423. 1:10:26languages and then I heard they're going to demo it this week which that would make sense if you're going to if you
424. 1:10:32were going to sneak launch it and you have events scheduled and then you pull it it would make sense to do a demo but
425. 1:10:41I it's it's all speculation and Pate in here Pate who is quick to point out how I don't know an SSM from an SL LM from an llm from an spt t
426. 1:10:55P he will not tell us what's going on at Google at the Google so take that put that in your pipe and smoke it so I
427. 1:11:05don't know I have no idea but I but I did hear they're going to be demoing something this week my kids would love
428. 1:11:12that gold Gadget I don't know what you're talking about you can make money [Music] with you mean this it's solid gold a
429. 1:11:21solid gold I got this from Parker on Gold Rush I said Parker make me an Ingot I'm going to make it into a button this
430. 1:11:29is an Ingot of solid gold from Parker on Gold Rush he signed it it's signed on the bottom you can make money with J that's Parker's
431. 1:11:43voice ah Pate cam Ken it seems very close to coming out based on all the speculative articles Pate I think P needs a nap I
432. 1:11:54think Pate has been um TPU optimizing a little too much today to the Moon all right listen everybody I don't know ssms either I
433. 1:12:09know ah the [Laughter] Google all right listen um it's good seeing y'all um Happy First Birthday to the salon
434. 1:12:19um I don't know I don't know when I started this channel it was right around this time we're probably actually coming
435. 1:12:28up on the the one year anniversary of this but the first 6 months of this I just did recorded videos I didn't do
436. 1:12:33lives the live started like the end of April so I I don't I don't like this channel for me is much more about the
437. 1:12:40lives now than the recorded [ __ ] anyway good to see you all um Sher bear good to see you human being get rest Kyle thank you happy
438. 1:12:50birthday Salon I know happy birthday Salon steveo thank you a i l TMZ that's right all speculation all the
439. 1:12:58time you can go to other channels for facts come here for the fun come for the fun stay for the speculation we'll make [ __ ] up if we
440. 1:13:08don't know the answer we'll answer it just like chat gbt oh I'm Punchy I'm a little Punchy people you know what I mean Tyron truth
441. 1:13:19good to see you Tobias good to see you thanks for coming to the salon tonight steo Pate as always very nice Sher bear
442. 1:13:27uh Christopher Ramsey good to see you scam scam scam welcome to Chad add K katkin jrc he's Punchy he's clueless but
443. 1:13:44occasionally entertaining all right everyone have a good night I will see you tomorrow bye