Aug 30, 2024 - OpenAI Strawberry Explained: The Future of AI Reasoning

Transcript

1. 0:13[Music] [Music]
2. 0:28all right you ready champy [Music] are you ready it's Friday night date night welcome oh I didn't put up my little
3. 0:42picture what what kind of operation is this I thought I thought we were professionals around here good Lord what a mess there we
4. 0:53go it's almost like I tried it w w w w w w w w [Music]
5. 1:28w your feel feeling nervous aren your boy with quiet voice and impeccable style don't ever let them steal a
6. 1:43joy and G way keep them from running wild they can kick dirt in your face dress your to and tell you that your
7. 1:55place is in the middle when they hate the way you shine I see you tugging on your shirt trying to hide inside of [Music]
8. 2:15it let them spin let them scatter in the wind I've been to the [Music] movies I've seen how they
9. 2:30[Music] and the jokes on there you're feeling nervous you're
10. 2:44feeling something aren't you girl it's your brother's world for a while longer [Music]
11. 3:00you got to wait got I had the lyrics up before I should have kept them [Laughter] up oh good people Friday night date
12. 3:17night what is [Music] happening that's it champ you're just giving us a a half number
13. 3:32[Music]
14. 3:50God you're looking fine tonight [Music] every fellas got you in his
15. 4:06sight what you doing with a clown with a clown like [Music] me surely would have last a
16. 4:21little something's off there [Music]
17. 4:38mystery so tonight I'll ask stars [Music] above how did I ever wish your
18. 4:56love what did I do what did I [Music]
19. 5:12say to turn your [Music] Rous my [Music] way I'm the guy that never learned
20. 5:31[Music] never even got Second [Music] Glance cross a crowded room that was
21. 5:49enough I could look but I could never so T I'll ask [Music]
22. 6:07St how did I ever Your Love what did I [Music]
23. 6:22do good evening good people how many people we got in here 15 people in here what the hell is going on Tik tock
24. 6:31Friday nights are normally a, 1500 2,000 people we got 15 22 people in here come on Tik Tock let's get it going all right
25. 6:38people get the grandkids I know it's Friday night date night they're in bed that's not the point wake them up have
26. 6:43them start tapping the screen share the screen love the champion what champ I love that champion in his sidekick too
27. 6:52exactly he's the star of the show although he just walked away little [ __ ] Diva woo how y'all doing tonight we're here two BL
28. 7:03Tom Joey in the house Milo's wife in the house [Music]
29. 8:11[Music] is this place I can rest my poor kid
30. 8:31together my thoughts in sweet [Music] silence busiest place where the feelings [Music] arear from an over exposure to
31. 8:49violence this is the place I can slowly face the only one I truly can know these are tears from a long time
32. 9:02ago got these tears from a long time ago I made to cry 30 years or so these are tears from a long
33. 9:17term AG [Applause] go well old darling Oh Darling you say unto [Music]
34. 9:33me where have you been all my life time well I have been swimming the seven sad Seas four women have tossed me their
35. 9:51lifelines I pull them into the waters darker in I'd have warn them but I didn't know these are tears from a long time
36. 10:04ago got these tears from a long time ago I need to cry 30 years or so these are tears for a long
37. 10:19time ago [Music]
38. 10:35no last note champy he's singing from the other room I don't know what's up with [Music] him well we went from 22 to 27 Tik tok's
39. 10:48they're they're chucking Us in the algorithm now [ __ ] it's Friday night date night come on tick tock I promise I'll show boobies and
40. 11:00talk about drugs [Music]
41. 11:24[Applause] [Music] put your pink bow
42. 11:39oh hey Tick Tock uh yeah so listen listen we got a talk we got a talk talk this is talk talk welcome to talk talk with me Kyle
43. 11:54Shannon your bedazzled host with the most with the most pink bows of any Tik Tok influencer you never
44. 12:09watch oh we we lost three we're down to 24 I didn't realize this was one of them their pansy channels
45. 12:26[Laughter] oh well what are you going to do you'll win some and you you lose you lose the bow [Laughter]
46. 12:42haters oh man then bows are scary I guess see here's the thing you all have been normalized to my madness it's the
47. 12:54the traditional the traditional ticktockers they have no [ __ ] idea they're like he looks sort of like Rosie
48. 13:00odonald how bad could it be and then on goes the pink bow oh man Hubba Bubba date night [Music]
49. 13:36[Music]
50. 13:5410,000 words swarm around my head a million more in books written beneath my [Music] bed ain't it like most people I'm no
51. 14:12different can't figure out how to hold my [Music] hand and I know you need me in the next room over I'm stuck in here all
52. 14:26paralyzed [Music] four months I got myself in ruts too much time spending mirrors FR in yellow [Music]
53. 14:43walls ain't it like most people I'm no different like to talk on things we don't know [Music] about making a big change in my life and
54. 14:59discussing it with myob bot what's theob bot fantastic experience what's aob bot you meano like the music
55. 15:08thing they've got Bots now did I miss something or is it a something else bot or is there AO bot I don't know
56. 15:16about what the hell are you talking about archetyp archetypal architect I can never say that
57. 15:31[Music] oh I made a bot specifically for sunno oh interesting oh to write I
58. 15:46see so you you created a bot to create suo lyrics but now you're having existential life conversations with it this welcome to the new world
59. 15:59[Music]
60. 16:21[Applause] let us be LS will marry our fortunes together [Music] [Applause] I've got some real estate here in my B so we bought a pack of
61. 16:38cigarettes and Mrs Wagner's pie and we've all come to look for aica [Music] [Applause]
62. 16:53[Music] [Music]
63. 17:08oh good lord good Lord good Lord I think I think what we're going to start with tonight we're going to start with a little [Music]
64. 17:21edication because it's a small group to start it'll probably get bigger it usually does that when they start slow like this
65. 17:32but what that mean what that says to me is the people that are here are the regulars which is great hey
66. 17:39everybody um so I think what we're going to do is I'm going to start out with um David Shapiro's video on explaining how
67. 17:47he thinks strawberry works and then Mr strawberry today said this guy gets it this is the first guy that's actually
68. 17:54understood what it's actually doing um so I think we'll go watch that video we'll just we'll do like a little
69. 18:02watch party I'll watch it then I'll stop it and talk about it if there's things that are interesting but I feel
70. 18:09like this is a good group to understand [Music] that it's what's what's fascinating about these large language models
71. 18:25is you'll see when we watch the video it's just like nested prompts it's not it's not it's not as [ __ ] remarkable
72. 18:41as as it should be like I don't know there's something about this new era of computing where you just ask the
73. 18:49computer to do [ __ ] and it does that's just still a little surreal to me [Music]
74. 19:08and he explains why so one of the things they're using this model for is creating synthetic data for GPT 6
75. 19:18apparently and if you don't know what synthetic data is this will explain that as well and when you explains it you're
76. 19:23like oh that kind of makes sense it makes it makes real good sense I'm excited to have the night off from work
77. 19:29880 hour work week this week Joy py good Lord you got to tell them to lighten up tell everyone to do their work during
78. 19:37the summer and not just the week before Labor Day that's I've heard I've heard this complaint from a lot of people that
79. 19:44everyone's just getting dumped on cuz their [ __ ] colleagues are like H hey I got to get this thing done by the by
80. 19:51so I can I'm taking off Friday did you know I'm going taking off Friday I'm going to take off Friday we're we're
81. 19:56I'll tell you what we're doing we're packing up the kid we're going on a road trip we're going to do it we're going to
82. 20:00do I you know I know it's crazy there's going to be cars everywhere but we'll be careful I'll tell you that anyway uh
83. 20:06listen I got a I got a TPS report I got to get out of you could just whip that up for you that would be fantastic and
84. 20:12then we'll get that into Bob The Bobs and uh boy they'll take that TPS report you know they're going to read it over
85. 20:18the weekend do they ever sleep how do they get all this done anyway yeah so if you had that kind of week I'm
86. 20:34sorry just sorry he wrote Frank out of the script Frank will be back we're just having our summer break is the OT good you mean the TV
87. 20:43show I thought the OT was really good season one anyway hey Kyle did you see the commercial for 1X Tech robot commercial
88. 20:52that new robot yeah looks amazing we'll look at that tonight TPS reports yeah hey Peter yeah listen we're going
89. 21:03to need you to go on ahead and come on in on Saturday yeah yeah you know 900 a.m. or so is that work for you anyway oh hey uh one
90. 21:14more thing Peter uh we're going to need you to go on ahead and come on in on Sunday too yeah Bob Bob
91. 21:29I wouldn't exactly call it [Music] working it's a classic that film it's a classic Michael
92. 21:46[Music] Bolton Samir not not not going to work here anymore anyway [Laughter]
93. 22:04I'll just entertain myself here you guys go ahead wait did we get it oh we got in three more people um you guys go ahead
94. 22:11and entertain yourselves I'm I'm doing the same here I'm just having a good old time oh my God I worked as a computer
95. 22:18programmer and that actually happened to me exactly what you're saying what the the uh we're going to need you to go on
96. 22:24ahead and coming in on Saturday that one oh oh hey hey one more thing we're going to need you to go on ahead and come on in on Sunday
97. 22:33[Music] [Music]
98. 22:49[Music] too oh man uh I'm just tired and Kyle cracks himself up I wait for these
99. 23:07moments I'm not even cracking myself up these are not even my jokes this is just my stupid ass half remembering lines
100. 23:14from a movie I saw 20 years ago that cracked me up then it was it was an odd week it was an odd week
101. 23:29um like it it was for us it was some of our clients were not there they were just non-existent or they canceled meetings
102. 23:38and then other of our clients were like oh my God we got to get it done you can to get it you can get it done you get I
103. 23:43got to go on vacation I I'll be I'm out next week but we can right as soon as I get back it'll be done right you can get
104. 23:49it done you can get it done so a lot of that I'm sorry they said there would be king if I if if I count the number of
105. 23:59slices it appears there's not going not going to be enough slices oh yeah oh this that that it does look tasty it it
106. 24:08looks tasty is oh no oh no there's there's no more cake left I I didn't get a cake could I get
107. 24:22a I know too I know too many too many scenes from that movie oh my God the opening scene where where what's his name the the little dweeby
108. 24:35one is like singing to Michael Bolt Is It Michael Bolton where he's singing to the rap songs and the black guy walks by
109. 24:42the car and he's like closes the lock and starts rapping a lot more quietly oh my God and then Peter the the opening
110. 24:53thing in the Dallas traffic Source Camp if you're on here where he cuts into the other lane and then it's stops and the
111. 24:59other lane goes he cuts back and it stops and that lane goes oh my [Music]
112. 25:13God oh my God that movie that movie it's my stap stapler [Music] stapler going to burn the place down burn the place down
113. 25:28[Music] uh Papa JJ yeah Oprah Winfrey on September 12th we're we'll do a watch party so it'll probably be we'll
114. 25:40probably go on at 6:00 that night CU I assume it's prime time is is 8:00 P.M Eastern is that right is that still Prime Time Eastern
115. 25:50um but whenever that show goes live um I think we'll we'll live stream it if if we can we might get we might get Tik Tok
116. 25:59banned but if we can if we can stream it we'll stream it so September 12th Oprah Winfrey is interviewing Sam Alman and Bill Gates
117. 26:09live and so this is right around the time when you know they said things were going to be launching in the fall and
118. 26:16strawberry seems to be confirmed and all that stuff so it would not surprise me if they didn't do a little bit of coordinated
119. 26:25launching right around the time they're doing you know an audience of however many millions she is tens of tens of Millions for [Music]
120. 26:53Oprah 9:12 is a Thursday okay cool beautiful so Thursday night so Wednesday night I'll be flying back from Washington DC
121. 27:05so Wednesday Tuesday September 10th I'm flying to DC we got a big dinner and then on Wednesday the 11th I'm there with 11 other AI
122. 27:18entrepreneurs and we're talking to senators and [ __ ] like that about AI so I'm going to talk a lot about the AI
123. 27:25salon and this Channel and what going on and how our legislators should not just legislate from fear and from
124. 27:36ignorance they might want to go on ahead and try the chat GPT and understand what the [ __ ] it is and understand how empowering it can be
125. 27:48and legislate from that position as well so that's going to be that's going to be my message I tried to tap while you play
126. 27:56guitar that was a lot of tapping to night what time is it yeah it's a little bit late um too bad klouse Schwab
127. 28:03couldn't make it on Oprah with Bill Gates um are you still coordinating with Google yeah it's still it's still uh The
128. 28:12Flyin is jointly sponsored by Google for small business and the salon which is exciting um that's overly optimistic I
129. 28:21don't know what that's in reference to I assume me saying something's going to launch I agree it's overly optimistic IC
130. 28:28to assume that anything's going to launch ever [Laughter] again oh man all right wait I got to put my charger on my
131. 28:41computer give myself some hydration cheers oh Lordy um I think it's important
132. 29:17[Laughter] Peter Michael the hell's going on man people thought you were going to come in here and start shooting no I just can't
133. 29:32get my address book I'm not going to stay I got a phone number mic that I don't want to lose what Peter you're in
134. 29:38deep [ __ ] you were supposed to come in on Saturday what were you doing Michael I did nothing I did absolutely nothing
135. 29:45and it was everything that I thought it could be well I hope you have a better story than that for Lumberg you know
136. 29:51you're supposed to be having your interview right now with the Consultants the who the Consultants what has gotten into you oh yeah
137. 30:03right te here got to postpone it man tell me you been sick make someone up oh no way no I feel great it's the best day of my [Music]
138. 30:13life SP looks like a Peter Gibbons oh there you are we just talking about you you must be Peter Gibbons
139. 30:22uhhuh terrific I'm Bob Sidell this is my associate Bob Porter oh hi Bob Bob go ahead and grab a seat and join for a minute or
140. 30:30two you see what we're actually trying to do here is we're just we're trying to get a feel for how people spend their
141. 30:37day at work so if you would would you walk us through a typical day for you yeah great well I generally come in
142. 30:49at least 15 minutes late uh I use the side door that way Lumberg can't see me and uh after that I just sort of
143. 30:58space out for about an hour till I'm space out yeah I just stare at my desk but it looks like I'm working I do
144. 31:08that for uh probably another hour after lunch too I'd say in a given week I probably only do about 15 minutes of real actual
145. 31:18work uh Peter would you be a good sport and indulge us and just tell us a little more oh yeah let me tell tell you something about TPS
146. 31:32reports the thing is Bob it's not that I'm lazy it's that I just don't care don't don't care it's a problem of motivation all
147. 31:43right oh you're welcome all right let's go f let's go let's go find some actual AI stuff here um I'll tell you what actually you know
148. 31:56what hang on there's there's there there's something here there's something here to talk about so if you have not memorized all of the
149. 32:09funny scenes of office space there's your homework for the weekend you got to get your [ __ ] together um the TPS reports the central the
150. 32:22central object of office space are these [ __ ] TPS reports that it's all anyone talks about and it's all they do and no
151. 32:30one reads them right they're just they're just this thing this is the kind of [ __ ] these are the kind of jobs those are like the the
152. 32:40TPS reports that's the kind of [ __ ] that AI is going to automate the snot out of and those tasks are going to go away right because
153. 32:53they're just going to be able to happen and and then so smart asses like Peter you know they either won't work there or they'll move
154. 33:02up in management like he did um because he he's smarter than you know the reason he sucks at his job is because his job
155. 33:12sucks and he knows it and he knows no one gives a [ __ ] that scen about did you did you get the memo we're
156. 33:22put now we're putting the covers on the TPS reports he does 15 minutes cuz he knows AI exactly that's going to be the new thing
157. 33:37shareholder Journey oh man I think it may be amusing how fast upper management finds out its replaceable Kyle is your video still at
158. 33:4699% yeah in fact I rendered another one which is also now at 99% so I don't know what the [ __ ] going on with with cling
159. 33:59video It's both of these are hung now which is a real drag anyone else have H hung videos on
160. 34:18cing this is funny I like this movie yeah it's a really good movie um okay AI learning lab all right so let me give you a little a
161. 34:32little context those of you that are in the channel if you're if you're new here my name is Kyle Shannon this is the AI
162. 34:37learning lab as evidenced by the little sign that says welcome to the AI learning lab it's fantastic isn't it
163. 34:45Danielle did that for us she made it out of clamation she's a she's a modeler she goes to the uh to the dollar store and
164. 34:53she buys the uh the 10 packs of clay and she's really good she puts them in um she actually mixes the colors right you
165. 35:02got the 10 that are all kind of harsh she actually mixes them to get these more subtle colors look at this she made this for
166. 35:10us she made like a little Tick Tock symbol that looks like it's out of focus all with Clay amazing right that was this was definitely not
167. 35:24made with AI this was definitely this is handcrafted see here here on the AI learning lab we will not use art well so
168. 35:34you can't even call it art if it's made with AI it's not art right so the only thing that's got any value in this world is things that are
169. 35:43handcrafted and so everyone here knows that and they know that we don't really like AI that the whole AI learning lab
170. 35:50what we're learning here is how bad AI is and and how you shouldn't use it and so this image is actually she took a
171. 35:59photograph with this with a Nikon D50 a film camera with film she took the film down to the uh down to the the Walgreens where they
172. 36:11still process film she had that processed and then and then she she took that she scanned it into her computer
173. 36:19with a scanner not one of those all-in-one printer scanner combos that for from where I sit and this just could
174. 36:28be me if you're using a combo printer scanner it's not really a pure scanner so that's probably too digital for our
175. 36:36taste here we want this as analog as possible so she's got a flatbed scanner she scanned the photograph that was
176. 36:44printed on paper that was shot on film of a clay sculpture that she made with her hands and then and then she sent me
177. 36:53that uh via VIA FedEx on a on a floppy drive and so that that's how we get the artwork here so thank you Danielle for your work
178. 37:08that I mean really seriously must have taken hours hours you can make money with see here's the thing I'm a smart ass that was all
179. 37:22just fibs that was all just fibs Danielle made this in like 30 seconds she made it with AI oh my God I had y
180. 37:32all going you were probably all like well where do I get clay I want to mix custom color hey that's crazy talk I know
181. 37:45anyway I guess I guess I'm in one of those moods indeed it took seconds hours hours to handcraft this Suzanne you got
182. 38:01me oh good he listen it is my joy in life to uh to to to say stupid [ __ ] so dry and convincingly that people go
183. 38:13really is this that is that really how she did that I'm impressed more with the Improv I totally get tricked into that
184. 38:28that's fantastic what' you do Friday night I got Bamboozled I got Bamboozled by by a theatrical that wears pink bows on his
185. 38:42head incorrectly okay listen if Cory Sandler pottery's in here she cannot speak okay in in fact women you cannot
186. 39:00speak this is not this is not for you to you have you have no input on this none men men in here how is this wrong I mean other than
187. 39:18the fact it's a pink bow and I look [ __ ] ridiculous I have been accused of not knowing how to wear a headband you tell
188. 39:27me what I'm doing wrong men cuz the women they've got some [ __ ] conspiracy theory about pink bowed
189. 39:34headbands that men apparently didn't get training on in the seventh grade so anyone what if I don't identify as a
190. 39:49woman as a Shira Shir see let's see seems normal to me see AR typle architect exactly this is other than the pink bonus of it this is
191. 40:03absolutely correct it's wrong take it from a guy who grew up with all sisters and a single mom ah damn it how's it
192. 40:12wrong what's it supposed to be is it is it like you're supposed to have it up like this is this it cuz that looks [ __ ]
193. 40:25ridiculous I mean you know not that this doesn't look ridiculous what happened to your hairline shut up it's male pattern
194. 40:34baldness I'm getting old just trying to get in touch with my feminine side I'm aggressively getting in touch
195. 40:48with my feminine side I'll wear the pink bow the way I want to wear the damn pink bow ladies blood pressure goes [Laughter]
196. 41:03up and scene let's bring it down I feel pretty oh so pretty Three Mile Storage man Jim Ross is like what happened to this
197. 41:17channel I go away for a week and I come back to this well we talk about the salon right so now we have salon product I wonder if
198. 41:28I have any hairs growing I've been monoxide diling oh all right add more jazzer size this is definitely this is definitely
199. 41:40jazzer size era era autra M right leg warmers wait should it be oh maybe no
200. 42:01I'm going into an add hyperfocused tunnel on on how this should be worn that's really bad maybe it's a bow tie oh that's well
201. 42:18I'll tell you what if we did this with the manini I I'll tell you what this and a manini we got ourselves a [ __ ] [Laughter]
202. 42:36party oh lord it's just the Friday night date night's not going well if you brought an actual date here I'm sorry about this uh
203. 42:47I will apologize right now to whoever you brought that has not experienced this before it's not normally like this
204. 42:54it's normally much worse um so I apologize for that okay chip Andale Kyle actually uh my my uh co-founder of agency.com and I
205. 43:06Chan he was he was a really big tall Korean dude but he and I are both a little on the chubby side so Chip and
206. 43:12dals was all the rage when we started agency.com and uh so so we said we're going to open we're going to open a club
207. 43:20called Chubb andales where the guys start out naked and then the women pay the money to put their clothes back on [Laughter]
208. 43:31it's chub andales but so by the way Irregulars if you want to make some Chubb and's images and toss them in
209. 43:37Irregulars I think that could be kind of fun where was my life before this channel you were watching actual entertainment on Netflix I
210. 43:47think this this this is this is like sucrose it it's it's a substitute for entertainment and it's bad for
211. 44:03you oh man sounds like banfa for me yeah probably yeah tick tock tick tock with there you said a dirty word we're going
212. 44:13to we're going to age restrict this channel we're not going to promote you you you dirty birdie you said boobies whatever you do
213. 44:22don't show a booby or we will ban you for life ban you for Life we'll do it we'll do do it we have an AI for
214. 44:31that it's crazy although I did see a Tik Tok today that apparently somewhere in Tik Tok now you can actually clone your
215. 44:40voice and then you can just type scripts and it'll make videos of you talking that sounds not not much like you for what that's
216. 44:50worth um okay so let me give you a little uh I'll give you a little teeny tiny history lesson wait he wears the pink
217. 45:01bow and then we're going to get a history lesson this is the worst date night ever can I have some more nachos
218. 45:08um did you microwave these they're chewy you got to put them in the air fryer Bob oh God do I have to do
219. 45:18everything myself that's not how you wear a pink bow Bob um [Laughter]
220. 45:37sorry now I want nachos now we've got three three comments on nachos it's good I like it try the pink bow dragon egg and talk
221. 45:53about the atmology of emology I see that would be good why not why not squirrel welcome to Chad ad hey
222. 46:04everybody uh what are we going to talk about right history lesson okay um so last November November
223. 46:232023 open AI had just launched custom GP s we were all fired up about it we're building gpts we're figuring them out
224. 46:31we're like yeah gpts are going to change the world we're going to make a fortune they're going to do a GPT store
225. 46:38woo and then like two weeks later they fired Sam Alton and then everyone's like oh [ __ ] and then they unfired Sam Alman everyone
226. 46:50was going to quit and then but apparently what happened was Ilia sus and probably some other people but it sounds like
227. 47:05Ilia saw something in their technology that spooked the living [ __ ] out of him and so he went to the board and said uh Sam's going
228. 47:16to put this in the world and he shouldn't do that so we should fire him and then the board said yeah we Sam's an
229. 47:24[ __ ] we don't like him let's fire him this is so much more fun than Netflix or sucros thank you I appreciate that AI
230. 47:31learning lab nearly as good as artificial sugar um so the world of open AI got tossed into chaos um and and what it seems like has
231. 47:48happened is that since that moment they've been working on testing finishing this thing that is called strawberry
232. 48:01project strawberry or it's called project Orion um it was announced yesterday or the day before in the information that has a bit of a
233. 48:12leak um and it it indeed is is a thinking reasoning GPT or or chat GPT or model from open AI like supposed to be quite
234. 48:27good quite powerful um and the other thing you know over the over the past six or eight months there have been a lot of announcements from
235. 48:37open AI where they announce something everyone gets all hot and bothered like the Sora video model and the real-time
236. 48:46voice and realtime voice they're like they they announced that back in April I think and they're like couple of weeks
237. 48:51it'll be out it's going to be fantastic you're going to love it and then couple of weeks came and went and then four
238. 48:56weeks came and went and they're like yeah about that couple of weeks it's going to be a couple of weeks more and
239. 49:01then they were like yeah about that couple of weeks more it's going to in the fall yeah yeah that's the ticket yeah
240. 49:11yeah and so nothing's been launching everyone else has been catching up all of the Just just [ __ ] Technologies everywhere that are
241. 49:24all of about the same class level um but it does seem now like some progress is happening Sam mman tweeted
242. 49:38yesterday or the day before I think it was yesterday yeah it was yesterday that they've come to an agreement with the US government the
243. 49:48AI I don't know review panel board [ __ ] whatever thing it is that they've come to an agreement for how they're going to
244. 49:57do testing to release these more powerful models so I assume what that means is that sometime over the past 6
245. 50:04months they've been interacting with the government there were some tweets about that as well to let them know what's
246. 50:10coming and to show how they did the testing so that it's safe check your regulars please oh no oh we got a video
247. 50:30oh [ __ ] that's really good wow so if you were not here last night
248. 50:46we did a we did we did what took way too long we went to chat GPT we generated some concepts for um short videos that
249. 50:55have an opening frame and a and a closing frame and then we had three different image models generate start
250. 51:02and end frames and then we the the intent was to do it in three different video models but only one of them
251. 51:08actually worked and then it hung so this is this is the one that hung but this this actually is it that that's actually staggeringly
252. 51:20good the prompt was envelope magically opens itself and bursts into flames I would say it bursts into flames and then magically opens itself
253. 51:31but that's pretty amazing um Irregulars Channel okay anyway that's pretty slick new activity oh yeah nachos anyone
254. 51:47chub and dals oh chub and Dales I like it now listen if you word to the wise if you ever start a Tik Tok
255. 52:03channel in a time in history where people can not only make fun of you but they can make fun of you instantly with
256. 52:13hilarious comical images uh oh it's going to sting it's going to sting Chum and Tails
257. 52:27take my money put your shirts on please please are those your fantasy men no can I leave can we leave how do we get out of
258. 52:49here the Church of the bow beautiful you all are [ __ ] weird oh man that's hilarious that's the chub and's
259. 53:06winner that actually looks just like the bow that's pretty amazing I think this one it's with the nachos the nachos
260. 53:14really make it this is this is like the stage Mom this is the guy at chub andales that wasn't quite good enough to
261. 53:22make it on stage so he's like the he's like the backstage mom always cooking nachos for the boys oh good lord all right so okay
262. 53:42so so so what appears to be coming other than more more comedy images at my expense um what seems to be
263. 53:55coming is this new powerful engine from open AI um David Shapiro if you don't know him you should know him so we're we're
264. 54:06going to hop over to YouTube David Shapiro he made this video he couldn't he couldn't sleep because his mind was all a
265. 54:20flutter David Shapiro strawberry there was also a whole bunch of drama with the strawberry Twitter account and a whole bunch of hype around
266. 54:29it but it does appear to be coming so what I thought would be good to do is watch this video together better
267. 54:40faster cheaper pick all three dcom AI first customer service in Puerto Rico we call ourselves B we're proud hello everybody I awake
268. 54:54and I don't want to be because I couldn't shut my brain off so so David is neuros spicy yes he's wearing a Star Trek shirt he always
269. 55:02wears a Star Trek shirt and he is one of the most Brilliant Minds in in AI he he's super into open source he's super
270. 55:11into AGI he's written I think five books on AGI if you don't know what AGI is it stands for artificial general
271. 55:17intelligence how many people do we have in here 50 okay um really knows this [ __ ] um if you if you tend on the geeky side he's someone
272. 55:29to follow if you if you want to be educated about how these tools work and what what the implications are of them
273. 55:38he's also good to follow you don't need to be an engineer to follow him but if you're an engineer he's got a lot of
274. 55:43content that's that's you know more Technical and and open source projects and things like that um okay let's see I never remember to even
275. 55:56go over there now it's the hair for me okay
276. 56:11okay he sure does love speculating wildly he annoys me to no end of course he does of course he does Pate there's lots better people on the
277. 56:22engineering side to follow uh there may be I again I'm not on the engineering side so this is this is about as
278. 56:29technical as I get and I I do like me some good speculation so anyway um this is his speculation on how qar Works how
279. 56:37strawberry works so let's watch I think I know how strawberry works I'm not even kidding so if you're
280. 56:45not familiar strawberry has been all the rage qar strawberry the leaks from open AI now the rumor has it right now and of
281. 56:54course just rumor we'll see if it gets it's walked back but they're talking about how strawberry can solve complex
282. 57:00math problems but then there's also talk about synthetic data and that synthetic data will be used to train project
283. 57:08o now something stuck out to me when they're talking about having solved uh the generation of synthetic data so next
284. 57:18I want to point out I was talking about synthetic data over 2 years ago so I was using gpt3 to synthesize data to create
285. 57:27synthetic data sets for fine tuning the reason that I'm sharing this it's not a flex I mean it is a little bit of a flex
286. 57:33but I'm pointing out that like I have been working on this for a while um and I've also been sharing this you see this
287. 57:40is the open air yes aio's wife he is wearing a Star Trek uniform he always wears Star Trek uniforms he's neur
288. 57:47neuros spicy like the rest of us he's he's special neuros spicy I for I've been sharing it openly for a while as well
289. 57:56so that's one thing by the by the way he refers to himself as neuros spicy so I'm not outing him in some in any kind of
290. 58:03derogatory way thing is just keep in mind synthetic data it's something that I'm actually a world leading expert in
291. 58:11um and then I want to talk about latent space Activation so latent space activation um basically what we realized is that
292. 58:21if uh all right here's here's another way to think about it you train these models on gigantic data sets and there's
293. 58:27a lot of embedded knowledge even if it's not fully crystallized and so what I mean by crystallized is it knows things
294. 58:35but it has to assemble them and this is very similar to how human brains work is that um have you ever had a conversation
295. 58:42where someone is asking you about something that you know a lot about but they ask you a question that you haven't
296. 58:47really thought of before and then at that time of instantiation you kind of connect all the dots and then you say
297. 58:53then you spit something out you're like wow I knew more about that than I thought that I did that's what latent
298. 58:57space activation is that's the equivalent of what latent space activation is and what what the latent space is if you don't know when they
299. 59:04when they train these models and they they put all this data and information in it converts it into these things
300. 59:10called vectors which is literally like thousand dimensional Vector space you know like an XY graph where you can
301. 59:17chart numbers on a two-dimensional plane the latent space is like thousand dimensional crazy complicated
302. 59:25mathematical space where there's all these points in space that represent the tokens that's what he's talking about so
303. 59:31it's basically it's basically machine learnings or like you know GPT 4's brain is the latent space but you know all all
304. 59:40anything that uses a Transformer uses that for uh large language models and I was working on this 10 months ago um
305. 59:50Source Camp I was talking about lat and space activity earlier today it's activation but yes I'm sure you were and
306. 59:57I made videos about it as well so that data is out there the next what I want to show you is how I think strawberry
307. 1:00:06works at a very high level just using a generic example on Claude And the reason that I think that I know how it works is
308. 1:00:14because I was working on this in my startup that I ultimately didn't go anywhere 18 months ago the the other
309. 1:00:21thing that I think is worth pointing out before he digs into how this works when he's talking about synthetic data what's
310. 1:00:27going on there is so gp4 they they went out and they scraped as much of the internet as they could they they got
311. 1:00:33these web scrapers that go out and find data they probably used illegal sources of data but they basically went out and
312. 1:00:41they got as much content off the public internet as they could and then they embedded it put it into this latent
313. 1:00:48space as these models get bigger and bigger and bigger we are effectively running out of data to train them so the
314. 1:00:56the basic thing that open AI cracked the nut on is they they realized that the more data and the more compute power
315. 1:01:04that you throw at um creating these these giant models the better they get um they're effectively running out of
316. 1:01:15data and and a lot of the new data is just going to be crap that gets vomited out of old AI models and it's going to
317. 1:01:23suck and it's going to start to pollute the data so synthetic data data basically is instead of going out to the
318. 1:01:29internet and finding all the data you're actually generating it out of one of these earlier models and by using this
319. 1:01:36reasoning engine what he's going to talk about here it'll it'll start to make more sense but basically they're
320. 1:01:41creating data that is high quality data that the new like GPT 6 can be trained on and that's what he's that's what he's about to talk
321. 1:01:52about who are all those people pay are those all the people that worked on the Transformer
322. 1:02:07okay so I'm getting really excited now at the startup that I was working on I was developing a concept that I called
323. 1:02:14surface plate surface plate would have been a combination of three models oh that's great so great so can p can
324. 1:02:23either you do that if you haven't can you toss that in the regulars Channel and maybe in um well definitely in in
325. 1:02:30the tech in the tech Guild but maybe also in the um in water cooler would be a good place to put those names good
326. 1:02:39good people to follow on YouTube that's awesome thank you for that oh you'd have the generator or the the expert then
327. 1:02:45you'd have the interrogator oh wait hang on this is important let me go back a little bit since I rambled before this
328. 1:02:52would have been a combination of three models so you'd have the a concept that I called surface plate surface plate
329. 1:03:00would have been a combination of three models so you'd have the generator or the the expert then you'd have the
330. 1:03:08interrogator um which is basically asking the the expert model what do you know about this and then you'd have a
331. 1:03:14grader which would be a third model that would grade the quality of output you can do all this with chat Bots now so in
332. 1:03:22this first one imagine that instead of me asking this question this is a this is a chatbot that is fine-tuned or just
333. 1:03:30instructed to interrogate the expert model and extract all of the information out of it to to create that latent space
334. 1:03:37activation saying I know you know about this at a very deep level now tell me everything you know about it to
335. 1:03:43crystallize it now you might remember the uh the paper that Philip covered over on AI explained textbooks are all
336. 1:03:50you need what we found is that synthetic data when it is highly curated is actually much more efficient at training
337. 1:03:57these models so that's all the theory behind it so in this first example I said tell me everything you know about
338. 1:04:03physics at a high level um we will iteratively drill into topics yada y y and so then uh Claude happily generates
339. 1:04:10the top 10 qu uh categories of physics and then I say excellent unpack fundamental forces and particles because
340. 1:04:17that was the first one so by recursively searching through the data remember what qar does is it's a apparently a
341. 1:04:26recursive search What's the title of this um video so David Shapiro is the guy and the video is called how open AI
342. 1:04:35strawberry works and then in quote quote marks one textbook to rule them all my educated guest algorithm now what I'm not quite
343. 1:04:45sure on is is how this would cover math so I'm probably missing something but at least I know how to generate synthetic
344. 1:04:53data and if you have enough tokens you can recursively generate or or or synthesize or you know do latent space
345. 1:05:00activation for all of human knowledge in these models and you can distill it and then also make let's go back and listen to that
346. 1:05:10statement enough tokens you can recursively generate or or or synthesize or you know do latent space activation
347. 1:05:18for all of human knowledge in these for all of human knowledge so basically we can we can synthetically generate
348. 1:05:24anything on any topic for all of human knowledge um and so if if you didn't get what he's talking about there these
349. 1:05:32three models are you basically have your core llm right that's answering your questions you then have basically a
350. 1:05:38chatbot that's just good at asking questions of that thing to vomit out data and then the third model is something that looks at what gets
351. 1:05:47vomited out and you know figures out if it's good or not so it's three different models interacting that's what he's
352. 1:05:52talking about here yeah and he says it so casually I know models and you can distill it and then also make new
353. 1:06:00connections so fundamental forces and particles it started with just uh three bullet points and then in the next chat and
354. 1:06:07remember you like this does not take much intelligence to interrogate again so you could very easily fine-tune
355. 1:06:14another model to just uh be the interrogator and it say certainly so then it unpacks more about fundamental
356. 1:06:22forces um and so then uh let's just I'll show you again it's like okay gravity is the first one so I'm recursively going
357. 1:06:29through just basically a binary search tree um tell me everything you know about gravity and away we go so again it
358. 1:06:39does not take that much intelligence to interrogate and activate and and create those lat and space activations and then
359. 1:06:46what it's doing is it's synthesizing information and so then what you do is then you take it to the third model and
360. 1:06:54so in this case uh I said you're a grader of data I will give you a bit of text and you will grade it with a rubric
361. 1:07:00what we found a long time ago is that and lots of people independently discovered this by the way is that language models are really good at
362. 1:07:07discriminating so you have a generator and a discriminator but in this case we have an expert an interrogator and a
363. 1:07:13discriminator they're really good at grading each other with rubrics um particularly if you tell it how to grade
364. 1:07:19it and on what criteria now I didn't give it uh a full rubric so a full rubric would be grade one is this grade
365. 1:07:26two is this grade three is this um but I gave it the first sample and I said you know there are four uh this is copied
366. 1:07:33from the other uh conversation and it gave it a five out of five um which I would I would say it's close I would say
367. 1:07:41it's not particularly comprehensive um and so the reason that you do this is because as you're generating samples you
368. 1:07:48want to grade all of them and you just basically discard the bottom % of the samples so that you're constantly
369. 1:07:55refining your your your your synthesized data set to be the highest quality information so um next sample whoops sorry fat fingered
370. 1:08:09that all right there we go okay so that's the next sample and we're going to watch it grade it um and again it graded at five out of
371. 1:08:23five I could see some room for improvement um but when you have basically you're you're basically recursively writing a
372. 1:08:30textbook of everything that humans know that is that is what I think strawberry does is by by using these models that
373. 1:08:39have already been trained on all text Data across the entire world they know everything that humans know whether or
374. 1:08:45not they know it and whether or not it has been trained in them well so by recursively generating a textbook and
375. 1:08:51you have the expert that is saying okay I'm writing the the base text and then then you have the interrogator that's
376. 1:08:56saying okay tell me what you know about that and then you have a third model that is saying okay cool that was well
377. 1:09:02written or maybe not and you could do this with I could imagine there might be several more fine-tune models and you
378. 1:09:08might have one that is reformatting it so actually let's do this as an experiment um so I will create another
379. 1:09:15chat and I say okay uh your purpose is to write um uh textbooks oh that's a that's a good idea d yeah I should be watching this at 1.5
380. 1:09:27speed I I normally watch it 1.5 speed although I I think right now it's fine it's fine going a little slow cuz
381. 1:09:35um like I I want to I want to take this like I want to I want to take this in like most stuff I don't give a [ __ ] if I
382. 1:09:43if I capture everything but the he's he's a pretty dense dude for you know for the non Pates of the world P's like yeah yeah whatever
383. 1:09:55he he's a child uh textbook sections based on the data I give you um write comprehensively um at the highest intellectual level uh
384. 1:10:11oops uh domain expert um expound upon the um Base information I give you with all the uh knowledge you possess on the
385. 1:10:27topic so then there might be a fourth model which is just this is just the drafter and so let's see what it
386. 1:10:35does um certainly I'd be happy to blah blah blah blah blah and then uh actually it's clawed so it's generating an
387. 1:10:41artifact so here we have it it's actually generating the chapter for us this is how I would if if I had the
388. 1:10:49money and the time and the uh and the and the compute time if you asked me said hey Dave use Laten space activation
389. 1:10:58and create me a fine-tuning data set that is going to be like one data set to rule them all um this is how I would do
390. 1:11:04it is I would uh use these models to iteratively basically generate one textbook to rule the is how I would
391. 1:11:13approach this and you see how how far this is going and how fast imagine that you have a whole bunch yes he is wearing
392. 1:11:20a Starfleet uniform CL or gp4 or whatever working in parallel to iter iteratively unpack literally every
393. 1:11:29domain of human knowledge that is what I think open a is doing and that if I'm correct that is how they are training
394. 1:11:37gp5 or GPT 6 we're not sure um now and again like I said the one part that's missing is I'm not sure how they would
395. 1:11:44have solved math um unless they did the same thing where they basically asked a generator saying hey write out this
396. 1:11:51latex formula and and figure out how to how to logically you know unpack this or something along those lines I might make
397. 1:11:57another video if I figure that out but yeah I hope you got a lot out of this um I could be wrong but as someone who's
398. 1:12:04been fine-tuning models since gpt2 this is how I would approach it so so we'll see and then Mr strawberry
399. 1:12:12responded to this video and basically said this is the first guy that's gotten it right that he got it right now Mr
400. 1:12:18strawberry has no [ __ ] credibility in my in my book because it's just a hype machine but you know for speculation but but I I mean that's
401. 1:12:31kind of fascinating so basically what he's saying is you have these models that have been the the way they train them is they take
402. 1:12:39data and they categorize it and and when they embed it it creates these semantic clouds these clouds of meaning where all
403. 1:12:47these words are in these well organized clouds but he's basically saying there's a bunch of [ __ ] in there that's not well
404. 1:12:57organized and wasn't well tagged but you can pull it out by just asking for it and then if you have something that asks
405. 1:13:04for it really good it'll vomit it out then you have another thing that looks at it really good and says here's the
406. 1:13:10good [ __ ] and just keep the good [ __ ] um then you end up with this you know massively valuable um all of human
407. 1:13:19knowledge in a well organized uh thing the the link here is this is a YouTube video the the the channel is David Shapiro
408. 1:13:30it's Dave shap I think hang on yeah at Dave shap and the video is um hello everybody I am awaken how open
409. 1:13:48AI strawberry works one textbook to rule them all so Bonkers right now what are the implications of
410. 1:14:04this I don't know our our gpts are going to get way smarter and they're going to be able to do like a lot of the stuff they fail on
411. 1:14:16now they're going to be able to do and so what that means is a lot of the tasks that we do for our job that require
412. 1:14:22human in the loop may not require human in the loop so assuming this launches this fall which it looks like it's going
413. 1:14:30to um we're going to have we're going to have some we're going to have some other [ __ ] to talk about on this channel all
414. 1:14:37right hope they found the Lost information of Alexandria we need a major bump wait you mean we actually learn here on this AI learning lab I
415. 1:14:49thought this is date night I know this is kind of rough um the whole thing we watched from reminds me of Spock's Brain
416. 1:14:56Episode 1 Season 3 there you go Andy's wearing a Star Trek uniform so you [Laughter] know yeah he he's he's funny he's very neuros
417. 1:15:09spicy he's he's solid he's solid on the neuro the neuro Spectrum uh you don't even have to teach it you
418. 1:15:25just point it in the right direction and it uses its data yeah yeah exactly well and that's so what they're saying what what what
419. 1:15:35I've heard about it is that it's very slow that strawberry is slow which actually makes sense right if
420. 1:15:42you say hey go solve some problem go go tell me everything there is to learn about it doesn't [ __ ] matter whatever
421. 1:15:51um genetic engineering in strawberry growing to to to come up with the more Street sweet strawberry in the least amount of you know
422. 1:16:03iterations it will go off and these three models will start interacting with each other so from what I understand it
423. 1:16:11sort of goes off for five minutes and just does [ __ ] and then comes back with you know incredibly wellth thought
424. 1:16:20out answers I don't know if that's how it's going to work but but that's a possibility um it's crazy and then you know what
425. 1:16:28it's going to be is it's going to be you know version one of this is it takes 5 minutes to go think of [ __ ] and then you
426. 1:16:35know a year later it does it in 30 seconds and a year after that it does it in 5 seconds and then it's instant right
427. 1:16:44so so the next three years are going to be absolutely [ __ ] Bonkers Bonkers I tell you Bonkers add a rag Checker and it's even better I bet
428. 1:16:56rag stands for retrieval augmented generation which is basically where you take your large language model that can
429. 1:17:01do all the language twirly stuff and then the rag is is basically a a database that's got your data in it and
430. 1:17:11then you interact with the large language model and it it if you ask it something about your data it goes in and
431. 1:17:18pulls actual data out and you know the large language model presents that to you so that could be quite powerful yeah
432. 1:17:26exactly all right um but this is just synthesizing human data or is it capable of generating novel ideas I don't that I don't know I
433. 1:17:38mean here's the thing novel novel ideas are just [ __ ] we haven't thought of yet how do you think of [ __ ] we don't we haven't thought of
434. 1:17:50yet is you take what you know and you make lot logical connections between those things and if you make connections between those
435. 1:18:00things that are different than other humans did then that's a novel thought but it's still based on some education
436. 1:18:09it's still based on language it's still B based on what information is in Einstein's brain that allowed him to think
437. 1:18:21about you know the relative speed of or the relativity of time compared to the speed of light it's he's not inventing it out of
438. 1:18:36nothing he's taking other Concepts and said oh no one else has made these connections when I make these
439. 1:18:43connections here's here's how I think the world works so I would think that if you had that interrogator
440. 1:18:54interrogate you to say hey go find me data sets where connections haven't been made or there's gaps of understanding
441. 1:19:04Between Worlds of understanding right here here's how automotive paint suspends particles to color shift and here's
442. 1:19:18um the history of hot dog recipes somewhere between those two things there are some connections to be made that
443. 1:19:26would be a new novel idea color shifting hot dogs that's a new novel concept so I could I could see this thing coming up
444. 1:19:37with new novel Concepts why not because novel concepts are not devoid of data they are derived from the data
445. 1:19:48that's in whoever's brain that makes the connections that no one else made right and and these models are good at making
446. 1:19:58connections really quickly between disparate data sets so but it's only going to keep on increasing its knowledge it needs to
447. 1:20:07increase compute time well but the other thing Danielle is these things are also going to get more and more efficient
448. 1:20:13right people like pate and people like the chip manufacturers and you know everyone's working to make these things
449. 1:20:19more efficient these customized chips like tpus and the grock chip that's just about doing inference where
450. 1:20:27where when you ask chat GPD to do its thing grock is doing that with a very specialized chip that makes it super
451. 1:20:34fast and super efficient so I think you'll as we require more compute you're also going to see the infrastructure get more and more
452. 1:20:43efficient um so you know again I don't I don't think the I I think energy and compute and all of that
453. 1:20:53are absolutely engineering challenges to be solved but I don't think we can look at the world of chips today and assume that
454. 1:21:02that's what they're going to be 5 years from now especially when AI starts getting in and improving chip design and
455. 1:21:09software design and infrastructure design right if you have a strawberry that can say hey we we have we have compute and
456. 1:21:21energy demand going like this but the physical you know infrastructure is only growing like this so you
457. 1:21:31strawberry need to close that Gap so we need to either make [ __ ] more efficient or we need to be able to make more
458. 1:21:38energy or more compute Cycles to match the demand and then AI is going to [ __ ] solve that like 2 million new stable crystals
459. 1:21:49novel Concepts yeah exactly I think it was only 38 ,000 stable stable crystals it was 2.2 million new crystals that were
460. 1:22:01discovered that was that was uh Deep Mind put that out right the the material science thing yeah 380,000
461. 1:22:11stable crystals 2.2 million new crystals or new molecules I don't [ __ ] know Material Science [ __ ] [ __ ] that does [ __ ]
462. 1:22:25and and that's that 2.2 million new um materials that were discovered was not with strawberry right it's whatever Deep Mind
463. 1:22:38is doing over there which is pretty [ __ ] remarkable but once once we have reasoning engines the the um the
464. 1:22:46Innovation that that these tools are going to be capable of it's just going to be Bonkers and it's what's what's fascinating to me is
465. 1:22:56this technology is going to be in our hands some version of it right it will be hobbled to make sure that you can't
466. 1:23:02you know take down the government and crack C cryptography and all all of the [ __ ] that you know they had to go talk
467. 1:23:10to the National you know government about from a safety perspective it's you know there's there's going to be some some guard rails on it
468. 1:23:21but once this thing's out in the world and we understand how it works and what it does the open source Community is
469. 1:23:28going to go oh we can go build that and they're going to go build their version of it and then grock's going to build
470. 1:23:34their version of it and meta's going to build their version of it and apple is probably just going to make me emojis that talk better [ __ ]
471. 1:23:47Apple we're the best no you were the best [Laughter] uh life escape velocity sign me up Lev
472. 1:24:04baby just survive the next five years that's what they're telling us on on my AI office hours today um we we were talking about that about
473. 1:24:21um like in theory you've got I think Danielle you were talking about this in theory you've got you know these tools
474. 1:24:31are going to be able to cure any diseases and then you've got lobbying groups like The Pharma lobbying groups
475. 1:24:38that don't necessarily want to cure right because sick people are their customers so if you cure it you don't sell as many drugs
476. 1:24:52um watching the death throws of big centers of power Financial power lobbying power political power as these
477. 1:25:02tools start to disrupt things is going to be fascinating it it's crazy crazy crazy shit's about to about to hit the fan
478. 1:25:18um that's why I'm glad we have so much open sourced [Music] all right what are your thoughts what are your thoughts Janu here for it it's
479. 1:25:34still probably 5 years away why do they have to be the scary ones yeah it's going to get weird for a while I
480. 1:25:43think we're also watching the death throws of copyright that may just impact Pharma yeah copyright the copyright laws were already
481. 1:25:55struggling to to have relevance in an in an all digital world and now that we're in a generative world where you're generating
482. 1:26:05original content at essentially infinite scale and infinite levels of personalization copyright laws just look
483. 1:26:15weird they look weird and Antiquated and I don't I don't have a good solution my my thought is that you you pay people on the generation of
484. 1:26:27officially licensed models so if I take all my photography and I create you know here's Kyle Shannon clouds photography here's
485. 1:26:37Kyle Shannon street photography and I make a customized model that's Kyle Shannon branded you can sign up for that and say
486. 1:26:48I want to make images that look like these photos and every time you generate an image it's going to I you know I'll get paid a
487. 1:26:59tenth of a penny or whatever it is that system doesn't exist yet but I I think that's where you have to go
488. 1:27:07because we artists and copyright holders can't be chasing down every seven-year-old that makes Mickey Mouse farting on Pluto's
489. 1:27:20head and and you know and that image ends up you know becoming a huge hit and everybody loves it and shares it and
490. 1:27:27then Disney what's Disney gonna do sue the seven-year-old like I I just don't I just don't think longterm copyright just gets really
491. 1:27:41weird so monetized Bots and gpts takeen to the next level huh I agree yeah I think so I think so it's it's it's weird man
492. 1:27:55[ __ ] weird all right what are we doing time
493. 1:28:10wise what do you guys want to do anything what did I have in my Friday night thread bid Shiro oh there's a new robot
494. 1:28:19out let's go look at the new robot I don't know a ton about it but it's an
495. 1:28:34impressive so it's called [Music] Neo so the initial responses to this Neo robot were will you put the burning
496. 1:28:49letter wait will you put that letter burning open Oh you mean in in Irregulars do I
497. 1:29:05have so so they've been they've been accusing that this is a man in a suit it's not this is a humanoid robot so
498. 1:29:12their humanoid robot is putting glasses away closing the cabinet closing the dishwasher and then and then and then
499. 1:29:23there's there's mama mama in the kitchen mama in the kitchen with the little bot no Danielle's like no no and then let's see here's another
500. 1:29:36[Music] one [Music]
501. 1:29:51yeah yeah [Music] wait this one gets creepier watch [Music]
502. 1:30:06this come on back lady give me a hug goodbye oh boy there goes the
503. 1:30:21neighborhood jennif Jennifer got one of the those neobot oh [Laughter] Jennifer I've never seen her [Laughter]
504. 1:30:39happier so that's the neobot oh wait here's another one I didn't see this one I don't understand this video maybe
505. 1:30:56that it walks sort of human still looks weird to [Music]
506. 1:31:11me I think that I think that that's the appropriate that's the appropriate meme for that robot and you know they didn't have a
507. 1:31:24talking there like when so I mean what are there like aren't there like 15 or 20 humanoid robot companies now like
508. 1:31:33figure figure is a 2-year-old company that they're already on version two of their bot this neobot is some new thing
509. 1:31:41there was just a Chinese one that came out like the hardware for these humanoid robots is quickly quickly
510. 1:31:50developing the software is going to be this really authentic natural voice from open AI with this insane reasoning brain in it
511. 1:32:07so welcome to the Future yeah yeah get used to it kids we're gonna be like hey Kyle can we go back and just make gpts again
512. 1:32:24remember when we made gpts for the nonprofits where you could like look up what the schedule was for the
513. 1:32:31nonprofit can we go back to that I like those [Laughter] days imagine it walking into your room your tea sir good day to you sir good
514. 1:32:47day I mean actually what would be funny is like okay so actually so so here's a here's the thing I am increasingly increasingly
515. 1:32:59Cindy Coon's in the in the call tonight or she was earlier anyway I'm increasingly feeling confident that all of us are going to have to
516. 1:33:14confront what it means to create digital versions of ourselves that we put out in the world and and think you should start
517. 1:33:25doing it now rather than wait until we have to and so here's what I mean by that you're going to be able to create
518. 1:33:34these autonomous agents that do things on your behalf well what might they do well you might have one that all it's
519. 1:33:41going to do is go negotiate contracts for you so it's going to call up United Healthcare and it's going to get your
520. 1:33:47[ __ ] paid and it's going to call up you know the car dealer and instead of you having to deal deal with the Ley car
521. 1:33:54salesman it's going to actually negotiate you the best deal possible right and that's going to represent it's
522. 1:34:00going to be a version of you that that is like the badass negotiator you might also have a version of you that's like
523. 1:34:07the social butterfly that's like you've got social anxiety and you don't really feel like meeting people but you're
524. 1:34:12going to have your social butterfly bot go out and meet people and introduce you to interesting people that it it's had a
525. 1:34:20good conversation with and it might have even had a good conversation with with their bot but both of those Bots agree
526. 1:34:26that you two should really meet and then when you have these inhome robots imagine having an inhome robot
527. 1:34:35that you could actually fine-tune and train to have your sense of humor so instead of it instead of me cracking my stupid jokes like I do on
528. 1:34:45here I I could say you know tell them what they've won Bob and then the computer would be like uh they haven't
529. 1:34:52won anything this isn't a game show could we uh get on with mowing the lawn like why not so that's something that
530. 1:35:02that I think over the next year I'm going to start exploring in here is how we can start to make our own digital
531. 1:35:08Twins and what it means to make them how do you use them why would you do it why would you do it now and not wait Kyle regulars
532. 1:35:19neurolink with a personality di oh imagine neurolink with a personality dial yeah exactly exactly where you can you know you can
533. 1:35:32basically just with your brain tap into these systems and have them be like you want them to be personality wise knowledge wise
534. 1:35:45expertise wise skills wise I know that the [ __ ] I'm saying is like
535. 1:35:59sounds [ __ ] weird and that's because it
536. 1:36:15is I mean I'm still kind of Blown Away
537. 1:36:30is that Korea no it's gen 3 Alpha where's Korea here it is so there's our video I I'll put that video on
538. 1:36:46uh on whatever you call it on Irregulars Channel all right let me type something here um oh wait let me go to let's see Kaa real time image
539. 1:37:06generation oh that's weird huh that's not working all right let me go to Deco here this this
540. 1:37:22seems like it's broken [Music]
541. 1:37:41come on create we we live in a world right now where I can type in um a a tall
542. 1:37:58building made of bet of made of jelly beans and
543. 1:38:16Twizzlers and as I'm typing it it's generating those things I I mean this is [ __ ] mind-blowing the
544. 1:38:33stuff we're talking about with strawberry and with these robots and with the ultimate textbook which is all
545. 1:38:41of human knowledge in this crystallized organized access instantly accessible format um I don't quite know how to process this [ __ ]
546. 1:38:54sorry I'm late on date night is this Korea this is Deco here but Korea Korea and Deco here both have these realtime
547. 1:39:00image engines they're using sdxl is the is the real time image generation sitting underneath these things and then
548. 1:39:09and then Kaa has flux which is the the high-end like mid Journey competitor that's open sourced where's my Kaa did I lose it
549. 1:39:31hello CLE good day to you sir so flux here's this generating you know high quality images so I can just
550. 1:39:39roll the dice it writes a prompt for me I hit generate and then it just quickly generates high quality
551. 1:39:50images and it's it like does the first pass at low quality and then it does them the next level up think that one's still
552. 1:40:03generating but like you know like this is just we're just in this [ __ ] weird ass Bonkers and then I can take any one of
553. 1:40:12these and turn them into a video you know it's crazy it's crazy crazy crazy crazy well then R2-D2 can be like an extra pet
554. 1:40:27this is Korea yes so what's nice about Korea is it's free and they're giving you access to the flux um image model so
555. 1:40:35you don't have to roll your own and start your own GitHub repository and repet or whatever the [ __ ] open- Source tool you got to use to
556. 1:40:47make your [ __ ] work you can just go here and do it I got a bad attitude for open source because it requires too much
557. 1:40:58concentration on my part so I'm just waiting for people to build me [ __ ] I can play with and this is one of those all right
558. 1:41:10fantastic thoughts questions craziness is flux better than mid Journey um it's better than mid Journey at text so flux is now really good at
559. 1:41:21text one of the other things that flux can do and Murphy is um if you remember back in stable diffusion days you could
560. 1:41:31create these things called luras L where you're doing sort of custom fine-tuning on stable diffusion for a
561. 1:41:40certain image it looks like flux you can do it with a handful of images and it's really good so like you could train it
562. 1:41:47to know what you look like and then just like dream Booth you could do before but you could have you in starring in a book
563. 1:41:55right so for character consistency um it it looks like the uh the ability to do much more controlled Generations um it looks like flux is
564. 1:42:05really good at it um again I haven't dug deep into it because you have to go do it in programming environments right now and I'm
565. 1:42:15just I'm lazy people I'm just lazy I'll tell you what you know what I want to do you can make money want to make money with chat JP
566. 1:42:25T just go hey go start me a business you big fat loser and it goes off and starts business and then sends me a check every
567. 1:42:32month you know what I don't want to check every month I want to check every week in fact I want I just want to get a
568. 1:42:39little Bing on my phone that says another $500 was put into your account this minute that's what I want in the
569. 1:43:00meantime Kyle I said earlier we need an actual C3PO AI robot so it's not scary yep yep yeah and it's funny I don't know if
570. 1:43:12you've been watching sunny on Apple TV the the movie about the inhome robot that punch punched a chick out and and did did some murders
571. 1:43:26um it's going to be wild like you know if you've got buggy if you get an early version of Microsoft Excel and it's
572. 1:43:35buggy you [ __ ] up a a couple of math equations if you get an inhome robot that's strong enough to lift up the back
573. 1:43:44of your car and it's a little buggy it could [ __ ] some [ __ ] up I actually started a screenplay idea
574. 1:43:59about this I don't I don't know where I put it I think I started it on this channel I've got to go find it cuz it's a good
575. 1:44:07one the screen the screenplay is basically um in the in the town where the robot Factory is that's making these new
576. 1:44:17robots they run a local contest at the local grocery store that people in town can get use of one of these robots for a
577. 1:44:27year and so it's about four people from the grocery store that each get a robot and then hi Jinx and [Laughter]
578. 1:44:41Sue I want a world we don't need to depend on money and that's listen if you if you listen to the Ubi guys and the the infinite
579. 1:44:50Prosperity narrative that's part of a GI and part of part of all this that that's a very real possibility cam katkin the
580. 1:44:59thing I can't get my head around is how do you prevent selfish people from remaining selfish and wanting to keep
581. 1:45:07all the profit if there's sort of infinite Prosperity we live in a world right now where the top you know 0.1% of the
582. 1:45:170.1% just want to siphon all of that profit to themselves personally I don't know how you stop that unless you get the AI
583. 1:45:28to take care of some things and creativity and and inventiveness are rewarded yeah exactly that would be awesome Wouldn't It Disney
584. 1:45:38should make a robot since they already um own the so the the so the SW imagery I don't know what SW is but but friendly imagery I agree with
585. 1:45:50that and I think we'll see that right okay here's the deal why we need artists why we need liberal arts Majors is all of the robot
586. 1:46:02companies right now are populated with Engineers they're not populated with you know DreamWorks imagineers so Disney and Disney by the
587. 1:46:15way has is making Bots like this right have you seen that Superman one or the Spider-Man one that that's a literal
588. 1:46:24it's a bot a humanoid bot that looks like Spider-Man that swings on a rope and then does a tum somersaults through
589. 1:46:33the air and lands in a net like Disney's going to be good at this [ __ ] but you need Disney imagineers working for some
590. 1:46:41of these humanoid robot companies to give them friendly faces to give them friendly forms to give them senses of humor to give them
591. 1:46:50storytelling right if it's just about I can do a task for you yeah if it's like hey we got some cool [ __ ] going on
592. 1:47:02here I mean isn't it already fascinating I mean we already have and this isn't even the good stuff like we already
593. 1:47:17have oh Star Wars yeah hey pie how's it going hey there everything's going great thanks for asking I'm here to chat
594. 1:47:29entertain and help out with whatever you need are you a Star Wars fan oh no I was just talking to someone else you remember my name
595. 1:47:39right if you've created an account with me I should know your name uh but if you haven't registered I'm afraid I won't
596. 1:47:45know it have you created see that she does I just I guess I'm not signed in but we've already got that and like
597. 1:47:51that's the shitty voice like that shitty voice to text voice the advanced voice that open AI has announced is one that's
598. 1:48:01got personality and it can hear our inflection and it can mimic that and respond with emotional tone and then put
599. 1:48:08that in a humanoid robot where imagine going on a 6-hour road trip that you've got to go on your own and you take your [ __ ] humanoid
600. 1:48:18robot and you have just [ __ ] wacky conversations with your robot and then you get to the gas station and you're
601. 1:48:26like I don't feel like just run in and get me Twizzlers while I pump gas and your humanoid robot runs in this gas station you
602. 1:48:35know and if and if you're a hacker you're you're like hey go get me a pack of Twizzlers and by the way you know Rob The Joint here's a
603. 1:48:49gun AI will hopefully democratize resources that's that's so what I'm hoping cam people hoard money often because they're
604. 1:49:00afraid it's all they know yeah yeah exactly yeah if if if AI can democratize resources I mean I think I think a big thing we're
605. 1:49:10going to start to get is accessibility um boosts where people that might not currently be able to contribute to society all of a sudden
606. 1:49:21can um and you know a lot of those people they might not be able to speak or they might have neurological things
607. 1:49:26or injuries or you know um you know blindness or deafness things like that and they might not be able to contribute
608. 1:49:33in a certain way that all of a sudden can so I think we're going to see sort of massive upswings in contribution
609. 1:49:41um and then and then yeah I think it I think it goes up from there I you know I'm quite bullish on the future but man
610. 1:49:48is it going to get weird fast it's it just like and again we we are the people we're we're the only
611. 1:50:03people who will know what it was like before AI before generative AI sort of blew onto the scene like it
612. 1:50:14has we'll be able to look back and say remember when it was just cute imperfect images remember when the videos were
613. 1:50:23only 4 seconds long and they look weird and we're going to be living in that her kind of future where people
614. 1:50:37that sort of grow up into that are going to think that's normal and we're going to be like it was never like
615. 1:50:44that you're dating a robot you don't think that's weird no why okay [Laughter] I'm still getting used to [Laughter]
616. 1:51:06swiping Sally's happier than she's ever [Laughter] been I don't know what it is her casseroles used to suck you think the
617. 1:51:19robots making the casseroles [Laughter] wonder what else that robot's doing the scandals the scandals in
618. 1:51:30Suburbia when when Jennifer gets her [Laughter] robot and all the other housewives are jealous because they've got a [ __ ]
619. 1:51:46human piece of [ __ ] human that [ __ ] about mowing the lawn doesn't talk to them and Jennifer's down the block having romantic
620. 1:52:03dances oh man we got some movies to write here [Laughter] people sorry I'm gonna lose it real real robots of New York
621. 1:52:21exactly real robots of [Laughter] Dallas actually you know what a reality show a reality show of neighbors where
622. 1:52:35some neighbors have robots and some neighbors don't that would actually be a [ __ ] fascinating show right because
623. 1:52:42you could see how the neighbors without robots see it from the outside and they judge it and then you could see it from the
624. 1:52:51inside what it's like and like what the issues are and like like nothing's going to be perfect and these things are going
625. 1:52:57to be janky right like you know they're going to trip and fall and break a coffee [Laughter] table but they're also going to mow the
626. 1:53:09lawn for you you know and Neil's just going to be smoking a cigar in his lawn chair while this robot mows the lawn and then the other
627. 1:53:20men that are out there mowing their lawns like [ __ ] Neil with his [ __ ] robot of course Neil has a BMW robot [ __ ] [Laughter]
628. 1:53:39douche oh man oh well I should probably let y'all get get get get on to your weekend it's a long weekend I'm glad you hang hang
629. 1:53:50out um I forgot it's Labor Day weekend so the fact that we have 60 people in here is absolutely stunning to me um
630. 1:53:57thank you all for hanging out um yeah we're we're in we're in wacky times aren't we we are in wacky
631. 1:54:17times this is different than this all right these are very
632. 1:54:36cool still meeting Monday yeah I'll probably be here Monday I don't I'm not going anywhere I'm not doing anything thank you for being here I
633. 1:54:46always enjoy listening thank you very much manifesting Monica I appreciate that it's very nice of you to say um I didn't really do
634. 1:54:55a an ad for the salon but if you are an AI person if you're like hey I love this AI [ __ ] I want to know more about this I want to
635. 1:55:06find people who are into this this community the salon is a community that I started the week after chat GPT came out a lot of
636. 1:55:15the people that hang out here are leaders in this community doesn't cost anything it's really remarkable community so go hang out there this
637. 1:55:24community is also the one co-sponsoring with Google for small business this fly into New York where we're going to go talk about hey AI is
638. 1:55:34good don't [ __ ] it up um so go check that out that's our website and then if you scroll down there's a button that says join our
639. 1:55:44community so go join the community um homework for the weekend um
640. 1:56:07yes don't pay attention to anything about AI just go have fun just go be in the world be with people connect that's the homework because
641. 1:56:25I I sense I sense that the next 3 months are going to get really weird and really intense and and some of it's going to be
642. 1:56:34like weird and exciting and good and some of it's just going to be like holy [ __ ] how do we [ __ ] process this it's
643. 1:56:40so funny when when um when chat GPT first launched the initial interviews with Sam mman he talked a lot
644. 1:56:51about oh thank you very much that's very very very generous I I appreciate that um he talked a lot about wanting
645. 1:57:02to not launch too much too quick because it would freak people out and as a as a tech fan and Optimist and someone who
646. 1:57:14loves kind of living on the front edge of this stuff my init my my initial response to that was oh you [ __ ]
647. 1:57:22[ __ ] just give us the [ __ ] give us the toys as I've watched over this year the tools get better and better and better they were they
648. 1:57:37were the initial things we were playing with were toys our toys right I can make a 4C video and if if you're talented at video storytelling
649. 1:57:50then you can put that in an editing program and go write your music and you can do good storytelling but the the
650. 1:57:56videos the video makers that we have right now are kind of toys and the image generating tools are kind of toys but
651. 1:58:03the image generating tools now that they're really good at text are starting to turn into things that are more than
652. 1:58:10toys and especially when you get someone like Elon Musk that puts flux into grock and takes off the guard rails for making
653. 1:58:20political figures and making you know Brands um all of a sudden you start to go oh these are kind of more than toys
654. 1:58:30like I'm I'm now starting to grock so to speak why Sam mman was like we've got to roll this out slowly get people used to
655. 1:58:39the power of what it's like to have a large language model where you can ask it a question and it just answers it
656. 1:58:47even if it hallucinates and does it wrong it's taken me a year 18 months to really even get my head around what does that even
657. 1:58:57mean and so I think the next 3 months we we're going to be confronted with technology that is very hard for us to
658. 1:59:10comprehend and it's going to be harder for us to actually use and assimilate and and understand what our work is and what our
659. 1:59:26value is I connect with people via VR does that count that's good actually it does um so so I think this weekend just take
660. 1:59:41just take some time to like don't worry about like this Tech is coming it's going to be here so I think have a nice
661. 1:59:49weekend like that's that's that's my homework work is is just [ __ ] have a nice weekend and I mean maybe think just philosophically
662. 2:00:04about if if the machines can do the stuff in your life that is the repetitive mind-numbing horrible tasks that you don't like
663. 2:00:18doing and you could just do the stuff that made you happy happy what would that be what would be what would be your life if the repetitive
664. 2:00:27low-level [ __ ] storm of Life were handled because I think designing that life designing who you want to be if all
665. 2:00:37that [ __ ] were handled that's going to be our job in the next three years so maybe think about that what
666. 2:00:44could my life be let's get the channel to 30 likes I assume you mean 30,000 we surpassed 30 likes but we are at
667. 2:00:5428,000 we're going to hit that um okay so go to the salon check it out join it unless you use Grill dad or Sin City
668. 2:01:04sipper to help you with your barbecue yeah exactly um yeah I think do nothing I think do nothing yeah it's good go find yourself
669. 2:01:16a pink bow I think we need more pink bows in our life and let me just tell you something all you people with you're [ __ ] wearing
670. 2:01:27it wrong with your with your pink bow rage there are many ways we can wear pink bows there are many people that can wear
671. 2:01:41them and how I wear a pink bow might be different than how you wear a pink bow and and we we cast a wide net here we have a large
672. 2:01:50tent that Embraces is all the pink bow wearing um yeah go get yourself a pink bow um all right I'm going to get out of
673. 2:01:59here so oh try character a AI if you want your feelings work for homework yeah character. a Danielle says it's
674. 2:02:07it's it's a mean-spirited service but it's apparently really popular yeah the the that that's another thing that we're going to confront is
675. 2:02:16that these AI systems already have personalities and we're going to thank you for 30,000 likes that's amazing um
676. 2:02:24we're going to start to discover the tools that we like that have personalities we like and some that have
677. 2:02:30personalities we hate and that's just like dealing with people so it's about to get weird hang on to your butts all
678. 2:02:36right um let's do uh oh very nice Winston thank you so much for the Cornucopia fruit there fantastic digital
679. 2:02:46twin and rag sharing helped me a lot tonight oh good fantastic spin B3 and that's we're going to do I I'm going to
680. 2:02:54I'm going to find ways to dig into that more on this channel because I I think I'm increasingly feeling like that's
681. 2:03:00really important that us really understanding who we are what data represents us and us being in charge of
682. 2:03:08how we put that into the world that feels important to me tell your friends AI Sal AI Salon presents Tuesday night 3
683. 2:03:16p or 5:00 P p.m. mountain yeah this coming Tuesday the salon we have our twice monthly meetings the first one is called
684. 2:03:25AI Salon presents and we have uh a guest speaker come in and speak so that's this coming Tuesday all right so be there for
685. 2:03:34that uh September 11th I'm going to be in Washington DC representing the the salon and all of us so hopefully
686. 2:03:45I maybe I should wear my pink bow to Washington I don't think they get it Mr Shannon Mr Shannon I understand
687. 2:03:59you're a thought leader in the in the artificial intelligence space uh we have Unearthed a video uh of it appears to be
688. 2:04:06you perhaps it's a deep fake uh wearing a a large pink bow on your forehead uh it it doesn't look like you're actually
689. 2:04:13wearing the headband correctly uh which is odd why you would choose to do it that way anyway uh Mr Shannon could you
690. 2:04:20explain is is is this indeed you wearing this pink bow and what was the purpose of the bow and what does that have to do
691. 2:04:27with artificial intelligence and and regulations we should consider here in uh in Washington DC and for for the nation as a whole Mi M Mr
692. 2:04:37Shannon sir sir could could someone please wake up Mr Shannon hi Mr Shannon so that's going to be fun are you still collecting stories for
693. 2:04:49your trip um yeah if if um one of the things that I'm going to talk about when I go to DC is stories of impact so if AI
694. 2:05:00has impacted your life in a positive way um I've I'm trying to think where I've got the announcement I've got an
695. 2:05:09announcement in AI Salon announcements um but you can just DM me if you've got if you've got some interesting story for
696. 2:05:17how AI has impacted your life in a positive way just DM me either here or on the salon and uh and we'll coordinate cuz yeah I'm collecting
697. 2:05:27stories for that so thank you for reminding me of that we learned a lot today thank you thank you side hustle
698. 2:05:32Mimi I appreciate it um all right I am out of here so who's in the house tonight Amelio's wife was here early
699. 2:05:39Auntie bubble's in the house uh Tobias is in the house wear your AI Salon t-shirt yes next Tuesday wear your AI
700. 2:05:45Salon t-shirt Oh you mean me wear it when I go to DC it's a good point um yeah maybe I'll put it underneath like a
701. 2:05:52a nice suit jacket or something like that all right it's not a bad idea um okay Vicki you can repin okay so go
702. 2:06:01to AI Salon if you go here go to the salon. go to the mighty networks's Community there's an announcements Channel Vicky's going to repin my
703. 2:06:10request for stories so go to salon announcements and you'll see my request for stories about AI impact all right
704. 2:06:18side hustle mimies in the house Lord digital God thank you so much for the Roses very very uh generous of you as
705. 2:06:25always um Winston great to see you Brandon thank you papa JJ cam katkin from Cleveland good to see you um Dr Jay
706. 2:06:33in the house who else we got Wolf Man Clint Be not Afraid Danielle good to see you two block Tom great night thank you
707. 2:06:49thank you very much for the kind words I appreciate that Marina good to see you wear your salon T-shirt with your
708. 2:06:57bow I think I should take this to Washington you can make money with J um put on your calendars the Oprah Sam
709. 2:07:09Alman interview cuz we're going to do that we're going to do commentary to that here um that's September 12th the
710. 2:07:16night after I get back from DC uh Joey's in the house [Music] Joey bunny cat bunny cat in the house
711. 2:07:32Jew live laugh love wait Jew oh Jewel live laugh love wear your smart pants too sir if we work for British Airways we're
712. 2:07:44expected to wear smart pants exactly I will wear my smart pants in fact I got some smart pants for my DC trip so I
713. 2:07:53don't look like a dumb Dum hey hello Senator my name is Kyle Shannon I want to tell you about artificial
714. 2:08:00intelligence otherwise known as machine learning or or vice versa if you know what I mean so some one of them is a
715. 2:08:05subset of the other anyway sir uh AI is just awesome so I think you should make it not bad don't mess it up all
716. 2:08:17right Kyle lives in a pineapple under DC nice verifier like it very nice very nice um all right good I'm out of here
717. 2:08:27people have a great Friday night have a great weekend have a great Labor Day and yeah just unplug don't pay attention to
718. 2:08:34any of this [ __ ] I will see you probably Monday night uh if not Salon Tuesday all right peace everybody bye