2/12/2025 - Exploring Coherence, Convergence, and the Potential for a More Ethical AI

Transcript

1. 0:01apparently AC according according to the technical manual for streamyard you have to hit the start button for the stream to actually start
2. 0:22[Music] sitting in this lonely Town wonder when things are going to
3. 0:40change dream my life away seems these dreams have turn to a bunch D CLS getting my nerve up but my past is
4. 0:53pulling the down uh yeah it's a producer fail because the producer is said he was going to be late tonight so so it is a
5. 1:02it is a uh it is a producer fail but I'm the [Laughter] producer oh man what are you gonna do what are you going to
6. 1:17[Music] do sitting in this lonely Town wonder when things are going to change
7. 1:32change dream my life away seems these dreams have turned to a bunch of dust [Music] clouds getting my nerve up but my past is pulling me
8. 1:48down wonder how long this black sheep don't stick [Music] around somebody told me once before he said you can never go home again won't
9. 2:05you leave S A things to Ste me away yeah from the truth of who I am and what I believe so I thank and for two SS with a
10. 2:18handshake and some sympathy yeah backed up my blue jeans and headed for this big price my freedom I didn't warm up my voice and
11. 2:35boy I'm singing as if I didn't warm up my voice it's thin and flat hit them hit them likes hit them likes thank you Source camp for the rose
12. 2:51means a lot thank you appreciate that oh man um interesting day in AI today [Music]
13. 3:06today I'm still in a place of not really knowing what to talk about but the announcement today actually kind of settled me down a little bit
14. 3:24[Music] what the f what are we what the effing
15. 3:38about Shapiro's YouTube was interesting and felt good yeah Shapiro's YouTube is something that I want to talk about yeah
16. 3:47let's let's we're g to talk about that I want to watch Shapiro's video [Music]
17. 4:01um many of you will remember that last night uh I had some Savory commentary on open Ai and every other [ __ ] company for
18. 4:15that matter um naming conventions and the fact that they're making us choose our own models as if we're [ __ ]
19. 4:23developers and now if you're a developer enjoy your model picking but for the rest of us we don't want to have to care if this is
20. 4:34a reasoning model or a non- reasoning model or the fancy one or the mini one or the fast one like doesn't [ __ ]
21. 4:40matter to us it should just [ __ ] work [Music] right and you guys probably don't know
22. 4:54this uh but Sam mman watches this live every night um he comes in an alias he doesn't want people to know and he's
23. 5:02there's a long history with Sam and about jealousy like I know there's that whole public thing going on with uh
24. 5:09what's that guy name the the billionaire dude um he's the one that changed the name of there used to be a social media
25. 5:18company called Twitter and he and he's the one that changed the name of that anyway samon him or beefing I think his name's Ellen
26. 5:28Ellen Ed I don't but anyway that's a more public beef that's like the Kendrick Lamar and Drake beef kind of thing with them the the
27. 5:38beef with me and Sam it's a lot more subtle it's just kind of a jealousy thing he does he won't even admit that
28. 5:45he knows me which is like it's like of course it's like a classic thing to do if you're jealous of someone you pretend
29. 5:52you don't know him you know what I mean so anyway so last night I'm bitching about how they they kind of really
30. 6:00[ __ ] up and and the whole point the whole reason chat GPT went to 100 million users in six weeks was because
31. 6:08they they took all the choices away from us and they just gave us the the chat hole right and so we just talk into the
32. 6:16chat hole and it just worked well what did they announce today Sam knew Sam knew I was right [Music]
33. 6:36um one of my great joys in life is that for everyone here you know that I'm full of [ __ ] but for people that are new to the
34. 6:47channel they're like does he really know Sam Alman I didn't realize Sam Alman was in a beef with someone else someone that
35. 6:54had a Tik Tok Channel I I knew he was I I knew he was angry with that that illan character the illen illen is that
36. 7:03his name anyway I he keeps talking about him on the news but he never mention anything about a Tik Tok channel that does it just gives me
37. 7:18[Music] joy oh man narrator forever wow I walked right I walked into what they announced today did I miss
38. 7:33something ye well yes I mean for the first [Music] time for the first time since they kind of
39. 7:49pre-announced chat GPT and they said it was coming in a month we actually have a road map from open AI for what's coming
40. 8:00what the next models are what they're going to be called what's going to be included in them what's not they've just been doing this weird
41. 8:08ass Clan destine shitty naming convention add another model to the list kind of thing for about a year
42. 8:16now and Sam finally listened to me and I just you know I don't like to brag but you know I mean when you've got
43. 8:23you know kind of supermodel good looks and and you know an intellect of you know near like a guy who once played Einstein and
44. 8:32Off Broadway um you know sometimes you just got to brag and so that's what I'm gonna do tonight I'm gonna I'm gonna I'm
45. 8:41gonna take the W you know take the take the win I'm gonna take the win it's gonna it's it's so good Model T Model A Christy Brinkley exactly
46. 8:56[Music] um let's see if our producer is in the house yet I can't remember if Brandon
47. 9:12said he wasn't gonna be here at all or he was gonna be late so we may be producer list tonight so that's all
48. 9:18right we'll be fine we'll be fine it's not gonna be like you know I've got this I think the hair and makeup crew did a
49. 9:26pretty good job tonight I'm not too shiny you know the hair is looking like a pretty good solid toupe which that's what we're shooting
50. 9:36[Music] for um so I thought we'd do
51. 9:54that I thought it might be good I don't know the other night we started digging into chat GPT a little bit I thought it might
52. 10:04be good to just continue exploring what's in chat GPT because it's so much now it's too much there's too much people there's too
53. 10:15much um also uh Adobe I don't know if you've heard of them they make this uh they make this product called uh photo photo booth or
54. 10:28something like that anyway photographers use it they make pictures with it and those guys came out with a new video
55. 10:35model today so we could go play with that I'm gonna I'm gonna spank him I have a feeling because I I I still don't think Adobe
56. 10:45can do 70s muscle cars got Winston we got Mr it joy py in the house silver fox [Music]
57. 11:23is canvas acting up on you too yes canvas canvas used to be pretty cool and canvas has gotten flakier than [ __ ] and
58. 11:33also chat GPT all it does now is it puts emojis in everything I make and I don't know why I don't know what I've done I
59. 11:40put in my custom instructions do not use emojis and it still makes emojis I don't know what's going on I think it's
60. 11:47possible that the the model has has taken off and it's no longer under our control and it really likes emojis [Music]
61. 12:03um generally when things are [ __ ] up at open AI it means they're updating [Music] something um so if you're new here hello
62. 12:15my name is Kyle Shannon this is the AI learning lab it's a it's Wednesday night February 12th we're just hanging out at some
63. 12:26point I'll put the guitar down and we can start playing with stuff and talking about stuff and watching videos and I
64. 12:32can wax poetic about the future of work or the optional future of work I still don't know how everyone's
65. 12:42gonna get paid I still haven't quite figured out how we navigate through the capitalist [ __ ] nightmare where they
66. 12:52just keep all the profits still haven't quite figured that one out yet but I do think there's a Tipping Point where they have to deal with
67. 13:00the work displacement thing so I think we're going to be [Music] okay if you have any questions about AI
68. 13:12pop them in the comments below you can put so the other night Monday night I think it was someone came in
69. 13:20and they said total Noob question and they asked a really simple question and it took me an hour to answer it and part
70. 13:28of that you might say well that's pathological you should probably seek professional help just answer the guy's [ __ ]
71. 13:36question but while it was a simple question and it was a question that someone just getting started in AI would
72. 13:43ask it's actually a real it was a really difficult it's it's a very complicated answer um so if you've got questions
73. 13:51don't think that there are any stupid questions that strike that there are absolutely stupid questions and if you
74. 13:57ask one I will absolutely mock you for it anyone who tells you there there is no stupid question is wrong
75. 14:05but one of the things I've learned in my career is every single catastrophe I've had business-wise has been because someone
76. 14:14wasn't willing to ask the stupid question the obvious question but there are stupid questions and you should be
77. 14:20made fun of if you ask them but you should still ask them [Music] we'll sell poppies on the side of the
78. 14:33road duh yeah I guess we go back to the to the 60s and 70s right we'll make candles we'll make candles on the [Laughter]
79. 14:46ashram my wife actually lived on the farm out in San Francisco in the the early 70s yeah I think early 70s maybe even 60s I don't know
80. 14:58[Music] she sold candles as a hippie China's Goku AI the most advanced
81. 15:14video generator ever made well but is it it's not released though is it James I don't think it is it's it's one of those
82. 15:24things I I'm I'm [ __ ] done I am done done with [ __ ] companies going here's a new video model and it's a white paper
83. 15:36no that's not a it's not a video model there might be a video model somewhere because what happens is the
84. 15:43the the X influencers and the Tik Tok influencers they watch the stupid video with the stupid [ __ ] examples that are all cherry-picked and
85. 15:53then they all go there's a new video model that's been released no it hasn't been released words matter it was
86. 16:04announced but if it was released then tell me because we'll go look at it and if it is the most advanced video
87. 16:11generator ever made cool now we'll have you know a 13th [Music] one developer preview yeah exactly it's
88. 16:26a developer preview released means works exactly no released means you can actually go use it that's released not you can go find a
89. 16:37video on Tik Tock and and re forward it under your own stupid [ __ ] influencer new model released ah he seems a little bitter he it's a
90. 16:54bitter pill this Channel's a bitter pill he seems Pleasant enough but then when he talks it's just it's just anger and
91. 17:02rage it seems like someone needs a hug good [Applause] [Music] Lord um let's see all right I guess we should get get started then we got 35
92. 17:18people here do me a favor we're a little light on Tik Tock do me a favor on the Tik Tok if you'd be so kind down at the
93. 17:25bottom there's a little arrow if you click on and say share the live and then send it to some of your friends that
94. 17:32hate AI bring them over see if we can get some trolling going on in here it's just a stochastic puppet it's just a
95. 17:41prediction engine there's nothing original about it it steals from [Music] artists thank you very much you don't
96. 17:51have a name you have a purple circle but you don't have a name but thank you oh wait put wait what is it
97. 18:00pause pause katrice Rachel oh I don't know your name but thank you for sharing the live Danielle shared the live sour cam shared
98. 18:14the live thank you very much get some folks in here [Music]
99. 18:33I found a new tick tocker that's like a Broadway composer dude he's just got the coolest energy so I just sent him an
100. 18:40email and said hey do you want to work on my musical Sydney all the songs were written with AI you'll probably hate it and tell me
101. 18:49to [ __ ] off but let's go for [Laughter] [Music] it oh hello Kyle the Rick Ruben of the AI world I
102. 19:05wish actually you know it's funny I I mean it's not Rick Rubin level obviously because I'm not winning Grammys or
103. 19:13Tony's but but I do feel like I do feel like the musical Sydney the musical is a is a good exercise in working through
104. 19:29the absolute crap that AI generates and and decrapify it right and and refining it and and just applying
105. 19:40judgment and taste and collaborating with Andrew and he and I have very different approaches to how we write
106. 19:45things and it's it's it's really nice and complimentary it can be frustrating because I like to be big picture and
107. 19:52move fast and he likes to be like let's get down into the words and word Smith the words and do them really Det ailed
108. 19:59but it needs both right it needs both of those so it's it's a really good working relationship and he's just a nice guy
109. 20:06and he's funny so we have a good time together thank you very much Silver Fox that's very [Music] nice go Goku AI says there's a long Q
110. 20:17nothing yet okay yeah it's it's it's in weight list it's and and listen um whoever said it James dter I was not
111. 20:25yelling at you I'm I'm just more yelling at all the influencers take those announcements and don't they don't
112. 20:31report them like a journalist would like tell us what it actually [Music] is they're like a new video model's been released here's 11 wild
113. 20:46examples they're not wild examples they're the examples from the announcement you [ __ ] tool [Music]
114. 20:59I used to fall for that [ __ ] too now the minute I see 10 wild examples I'm like ah okay I don't need to pay attention to that [Music]
115. 21:10post because sometimes those posts are are they're really deceiving because they're like there's this new thinking
116. 21:16model out there 10 wild examples and I'll go look at it and I'm like wait wasn't this thing released like six months
117. 21:25ago they just got the memo H videos are getting better they are absolutely the uh Luma laabs dream machine ray 2 is really good um vo is
118. 21:39really good but I don't have access to it nobody seems to have access to it that I know oh no Vicki does Vicky has access to [Music]
119. 21:53vo yeah Pika added a new thing called Pika editions which we can we can go look at and see if we can get to work I
120. 22:00tried it once and it I didn't get it to work but we can try that [Music]
121. 22:27[Applause] it's not simple to say the most days I don't recognize me that these
122. 22:42shoes and this a that place and its patrons have taken more than I gave them it's not easy to
123. 22:56know I'm not anything like I used to be although it's true I was never attention sweet Center I still remember that
124. 23:11girl she's imperfect but she tries she is good but she lies she is hard on
125. 23:26herself she is broken and won't ask for H she is messy but she's kind she's lonely most of the time she is all of
126. 23:42this mixed up and baked beautiful pie she's gone but she used to be [Music] mine isn't this supposed to be an AI
127. 23:57Channel why is he singing a Broadway song and isn't that a Broadway song about a pregnant woman from a pregnant woman why
128. 24:04is he singing that is he pregnant he's chubby like he's pregnant maybe you should cut up cut back on the donuts the
129. 24:11cinnamon gummy bears good Lord we'll get to the AI stuff waitress yes waitress that is uh a beautiful song I like it would yeah
130. 24:29that's such a good song If you haven't seen Jeremy Jordan Jeremy Jordan does a cover of she used to be mine that is just mind-bogglingly
131. 24:41[Music] good all right let's get going let's do this stuff let's do this stuff we're gonna go out let's see if we
132. 24:56have a producer yet do we have a producer no okay so here's what happens when when Kyle doesn't have a
133. 25:05producer there's a very high likelihood that I will be showing you something on screen for 15 20 maybe 30
134. 25:15minutes thinking that you can see it but I actually didn't share it so so just know that that's there's a
135. 25:25high likelihood that's going to happen oh man Star Trek Enterprise Solutions can I ask you to check out and get a
136. 25:36laugh out of my new AI created guitar stand sure give give me your do you have a website that you built with AI to sell
137. 25:46your AI created guitar stand let's go producer free zone exactly he's going to be wild tonight he's GNA be crazy I'll tell you what you
138. 25:57know what I can just put on a pink bow whatever the [ __ ] I want I'll be like and I I I'll wear it too I'll do it I
139. 26:04won't have the producer there going oh that's inappropriate uh headgear uh for for a professional operation exactly you know what I'm
140. 26:16saying you can make money I got control of the button I mean this is this is solid you know baby Yoda baby Yoda sure
141. 26:29we can do it all we can do it all tonight [ __ ] producers all right talk about Google gini well Brandon's not
142. 26:37here actually that's probably a really good call we should dig deep on that Silver Fox liking the pink bow all right
143. 26:44okay so all right uh where do I want to start let's start we'll start with Sam alman's announcement and then we'll go find David
144. 26:56Shapiro's um video um I haven't watched all of David Shapiro's video so maybe we'll put it on 1.5 it's like a 30 minute video maybe
145. 27:10we'll put it on 1.5 so we can play it quickly um or quickish Le um but he said it is it is likely the most important
146. 27:20video he's ever made um he's made a ton of videos and a lot of them I feel are pretty important um so I kind of want to go hear what that's
147. 27:29about um and it's it's about it's it's about how good these tools are getting and how quickly they're getting good and
148. 27:40how they're starting to ignore the better they get the less they listen to us and a lot of the doomers are um
149. 27:49saying that that's a really bad thing that our demise is imminent and David's got a completely different take on it
150. 27:56and so I I'm want to go learn what he has to say and and uh yeah let's watch it and talk about it yeah great okay so let me let me share my
151. 28:05screen let me share my screen here we'll read this little Sam Alman announcement together it'll be
152. 28:26fantastic uh uh uh uh sh I got to do the black bar thing do I have a brand in here no I don't look look I did the black bar on my own for
153. 28:39all you Tik Tok people how how lucky are you okay so Sam mman today open AI road map update for GPT 4.5 and GPT 5 here's why this is
154. 28:53significant people have been asking basically for a year like what is the road map for for GPT 5 is it is there going to be a GPT 4.5 is
155. 29:04there going to be this and and then there was this project Orion what was that and then there was 01 preview and
156. 29:11then there was 01 real and then there was 01 Pro and then then they came up with 03 and we're like but where's 4.5
157. 29:18and they're like maybe it's in' 03 like it was just completely baffling and unclear so just this state in all caps
158. 29:29is kind of a big deal it's just it's like for the first time he's like okay maybe their PR team actually said uh
159. 29:39yeah we might want to deal with the [ __ ] naming thing uh and in our our product road map thing um we want to do
160. 29:47a better job of sharing our intended ro road map and as and and a much better job simplifying our product offerings so
161. 29:56let's go look at what he's talking about about shall we okay over here on the leftand side what you're looking at right now is uh
162. 30:06chat J as I like to call it chat that's the French ch um so on the left hand side here you've got all these things these are all custom
163. 30:20gpts that I've made and I pinned here and then there's my chat history and then here's Sora which is a video model
164. 30:28here's operator which is an agent that sort of halfast works if you pay 200 bucks a month um and then here's chat
165. 30:36GPT and then within chat GPT you can upload stuff you can turn on search you can turn on deep research if you pay for
166. 30:45it you can use do if you're using the right model you can turn on canvas see there's canvas I had the wrong model selected
167. 30:59and then here's all the models GPT 40 great for most questions GPT 40 with scheduled tasks ask chat GPT to follow up later follow up on what
168. 31:13what are you talking about 01 uses Advanced reasoning 03 mini fast at AI reason fast at Advanced reasoning 03
169. 31:22mini High great at coding and logic oh so I couldn't couldn't I use Advanced reasoning for coding and logic wouldn't
170. 31:30that fall under reasoning no yes maybe 0 one prom Mode best at reasoning if it's best why is it last oh and then we've got two other
171. 31:42models in the more model section and then you can also have a temporary chat so so that's what he's talking about as we could do
172. 31:53better and then this is another thing we talked about last night we want AI to just work for you this is a Steve Job
173. 32:01thing Steve Jobs thing he said he said our products should just work you shouldn't have to configure them you
174. 32:09shouldn't have to you shouldn't have to [ __ ] choose what AI model right it should just work yes we realize how complicated our model
175. 32:24and product offerings have gotten have you C because you sure keep keep seem to you know shoveling more of them in there um
176. 32:35so so so the first three sentences of this right out of the gate are a big [ __ ] deal we hate the model picker as much as
177. 32:43you do and want to return to Magic unified intelligence Chad told me it' work on something and get back to me today and
178. 32:53it did oh wow Silver Fox first time that's ever happened V3 mini that's fascinating like in YouTube If you
179. 33:03haven't yeah give thumbs up in YouTube If you haven't he was definitely listening to you yesterday I'm telling
180. 33:09you that this is exactly what we talked about yesterday right um I don't I don't think I don't think I'm the only uh you know uh compatriot
181. 33:20out there uh who's been voicing these concerns um okay this is important we will ship the next we we will next ship GPT
182. 33:324.5 so they haven't shipped it yet the model we called Orion internally this is such a big [ __ ] deal
183. 33:40because I I mean for months people have been like asking him what's Orion did you launch is is is Orion 01 is it 03 is
184. 33:51it 4.5 what is it so here it is GPT 4.5 is Orion so about three months ago or so Sam put out a thing I love the night sky
185. 34:04and he showed the constellation Orion so GPT 4.5 was probably in testing back then now we've got some transparency on
186. 34:12that we will ship it we will ship GPT 4.5 as our last nonchain of thought model so here's why that's important let's go back to chat jet ATT
187. 34:30so when I use GPT 40 and I'm like tell me a bedtime story about baby Yoda it vomits out tokens right single prompt
188. 34:48in single response out a reasoning engine so if let's let's use 01 Pro since since I'm paying 200 bucks a
189. 35:02month let's let's let's burn up some let's burn up some Cycles on a shitty prompt write me a um bed time story about baby
190. 35:18Yoda by the way I'm 59 now what ow Pro does is reason so it
191. 35:33says it's reasoning so if I click on details we can actually watch its what they call Chain of Thought right so I'm piecing together so
192. 35:44this is it talking to itself I'm piecing together a cozy bedtime story for an adult featuring baby Yoda nestled in Nostalgia and
193. 35:54gentle reflection it might blend the Mandalore in with Timeless star warss Vibes next thought I'm piecing together a cozy
194. 36:03bedtime story enhancing baby Yoda with themes like time and love while maintaining brevity and respect for Star
195. 36:09Wars intellectual property oh that's funny so so it's worried about intellectual property right so it's
196. 36:15thinking now I don't know why it stopped I wonder if it hung I wonder if I broke chat gbt probably what's going on here
197. 36:26is so so 01 Pro is probably the equivalent of like having 20 phds as your best buddy and I just asked those
198. 36:3620 phds if they could write me a bedtime story so I assume it's hung right now because the computer's just going you
199. 36:44gotta be [ __ ] kidding me like we could be curing cancer and we're writing a stupid bedtime story for this old
200. 36:53man so anyway that's what reasoning is oh here we go wrote it it only had it only needed two steps of reasoning I wonder how much money we
201. 37:03just spent writing this stupid story baby Yoda's evening of Wonder a long time ago yet somehow always in the
202. 37:11present like a cherished memory there lived a small mysterious creature named grou though many fondly called him baby
203. 37:20Yoda he was Tiny with large curious ears and glistening eyes or glimmering eyes Okay so standard llms just answer your prompt
204. 37:33reasoning engines think about it so back to Sam GPT 4.5 internally called Orion will be our last nonchain of thought model so that
205. 37:48where it was talking to itself internally that I showed on the side that's called The Chain of Thought right
206. 37:53so it's prompting itself behind the scenes you you don't have to it sometimes they reveal it sometimes they
207. 37:59don't deep seek revealed their Chain of Thought and it was really well done so people loved that open ai's been copying
208. 38:07that recently so so we'll see what the new models look like but that's a big deal after that our top goal a top goal for us is to unify O
209. 38:18Series models and GPT series models by creating systems that can use all our tools know when to think for a long time
210. 38:28or not and generally be useful in a wide range of tasks this we also talked about last night so maybe Sam was watching I don't
211. 38:37[ __ ] know this is kind of in order of what we talked about things last night that what I said was what they need to
212. 38:43do is take their [ __ ] brains they've got some clearly smart people over there and create an AI model that can figure
213. 38:51out which of their AI models should be answering our question so that's so we're going to get 4.5 and then after that they're going to
214. 39:00basically create a model picker that that hopefully will just be transparent we won't even see what model it picked
215. 39:06I'm sure you can click on a button to see which one it used but most people don't give a [ __ ] in both chat GPT and our API we
216. 39:15will release gp5 as a system that integrates a lot of our technology including 03 we will no longer ship 03 as a
217. 39:25standalone model interesting St because they actually haven't shipped 03 they said they were going to ship 03
218. 39:33but what they shipped was 03 mini and 03 mini high but in both chat GPT and our API we will release GPT 5 as a system that
219. 39:49integrates a lot of our technology so I'm guessing things like operator and Sora might still be outside of that but we'll see
220. 39:59um but what it should be is you log into chat GPT just like the first chat GPT with 3.5 and there's a chat hole and you
221. 40:07stick your words in the chat hole and some answer comes back and maybe it thinks and maybe it doesn't and we'll
222. 40:12just have to trust them that it's doing the right thing um that that should also make things easier for for API usage but
223. 40:19who knows the free tier this is huge the free tier of chat GPT will get unlimited chat access to chat GPT 5 or
224. 40:30to GPT 5 at the standard intelligence setting exclamation point the free tier so if you take all of these
225. 40:42models right they're going to take the best of all of these models plus 03 the the the macdaddy reasoning engine the
226. 40:50one that solved the um that that AGI test whatever the one that beat humans at that a GI test right um they're going
227. 41:00to take all of those they're going to roll those into gp5 so gp5 is going to be this multi-headed Beast that just
228. 41:08acts like a single thing available to free tier users unlimited that's [ __ ] massive um subject to abuse thresholds
229. 41:20so I guess if you have Bots do things plus subscribers will be able to run GPT 5 at a higher level of intelligence Pro
230. 41:27subscribers at an even higher level of intelligence so what they're doing is they're giving all of the features to
231. 41:34everyone and then if you pay it gets smarter which I think is a fascinating model these models will will incorporate
232. 41:44voice canvas search deep research and more and that's it and then a bunch of people bugged him about you know Sama
233. 41:52weeks months and then he said weeks months so um we will see I the the implication here I assume is we will see 4.5 so 4.5 will likely
234. 42:10replace 40 but I don't know because you've got you've got 40 right they're they're multimodo model I assume 4.5 is part of
235. 42:21the Omni series The the multimodal series it's not 4.5 from the Legacy model I assume I don't know um so we'll probably get a
236. 42:32replacement for this in less than a month because if it were months and years then it would be more than a month
237. 42:40but weeks says to me it's less than less than a month and then we'll get gp5 before the end of the year is what that
238. 42:49is what that says to me that's what he's that's what he's claiming right now they also promised us Sora last June and
239. 42:56didn't deliver it until what six weeks ago do you think Sam Alman Sam alman's fingers burn when he's typing these X posts
240. 43:09yeah um so what's left in the paid tier higher intelligence and that makes sense I mean you know you you you pay for you're
241. 43:20gonna pay for these tools to solve harder problems so if you want to run a pizza shop if you want to run your gardening
242. 43:30design business if you want to write a book or a novel or you know do some creative work if you want to do some
243. 43:38basic coding a little bit of app development you're probably actually fine on the free tier it'll probably do like significantly
244. 43:47more like the free tier is going to be like having you know 50 phds at your disposal at at every possible category of work that'll
245. 44:01be the free tier if you want to go off and do things like I don't know analyze big Fortune 500 companies maybe you pay
246. 44:0920 bucks a month you want to go cure cancer then you pay 200 bucks a month maybe there's a $2,000 a month version
247. 44:16that's got unlimited agents or something like that and people will start thousand thousand person companies that have one
248. 44:24employee and a thousand agents for two grand a month that's what's coming so anyway we have that thoughts questions cool
249. 44:40right oh producer Brandon is here producer Brandon can you talk are you at a bar I think he might be at a bar
250. 44:55again no he's not there he's not there but he logged in so he probably is in a bar I'm here I'm here I'm here hey
251. 45:06Brandon hi you know th those kids are not good at waking me up at when they when they fall asleep I thought didn't
252. 45:14you tell me you were gonna be late tonight or was that some other night uh yeah that was [Laughter] tonight there was a lot of producer
253. 45:24bashing in your absence you'll be happy to no uh I can't wait to watch on the playback hey um have you seen the new
254. 45:34Sora explore speaking of Sora No Sora explore yeah caught this uh the other day it um you can now insert pictures
255. 45:46and kind of mash reality it's very similar to what Google is doing with whisk huh in just the regular sor sora.com mhm
256. 46:01huh what's 2v oh two videos storyboard oh introducing storyboard oh that they had before so storyboard is just key
257. 46:19frames let's see featured top where is it where's that feature you speak of uh so it is under on the left hand
258. 46:31side you have the ability now to upload images well I can do that here upload IM show every oh yeah was I
259. 46:45not hey look see what happens when you have a producer he can actually go uh dummy
260. 46:59Brandon is the man behind the curtain D okay um so so what do I do choose from a library
261. 47:14but it could do this before let's go to storyboard so storyboard it could do before storyboard you
262. 47:29go let's see I upload an image here choose from library all right we'll start with yeah so this might this might be
263. 47:42what we what we've talked about before here where you can take a image that you a static image and turn it into
264. 47:50movie a moving image yeah yeah yeah but you can also so like I can put a like that dude and then there's the prompt that goes with
265. 48:05it now I can add another thing so I can choose this image that's
266. 48:22fine let's let's just create and see what we got added to Q all right anyway we'll come back to that all right cool yeah this is a cool
267. 48:32feature again I'm I'm like the the the the challenge with all these video tools is they've all got these weird quirky little
268. 48:41interfaces and like to get good video you need to use multiple of these tools but they're all really expensive so you
269. 48:49really can't or you just use wherever you have credit so you can make videos but you have to learn all the tools it's
270. 48:55a nightmare right now so I hope the video scene tightens up like yeah and uh to to Bob danan um we'll leave the
271. 49:04singing to champy yeah yeah exactly all right let's go let's go find all right see you later um let's go find uh David
272. 49:13Shapiro let's go watch this VI David Shapiro Dave shappy Dave
273. 49:28shappy human PhD using Google in their field 03 Pro 03 is above is above the line gp4 was way down here look where we
274. 49:43are people so that 03 that's above that blue line that's going to be inside gp5 available to everyone for free and then the paid people get
275. 49:56something higher than that it's freaky times it's freaky times okay so I'm gonna have to I'll change my sharing in a second here so we
276. 50:09can listen to it AI will resist human control and I think that's exactly what we need so what the doomers are saying
277. 50:19is this report that came out basically says that as as AI gets smarter it resists listening to us more and
278. 50:29more and what he's saying is he thinks this is a good thing so that's what this video is about um we don't need to read that but yeah
279. 50:44this is oh this is the most important AI Discovery so far ever um and he and he put in another video
280. 51:00he put that maybe just maybe it was just a comment to this one maybe did a reply he said he thought this was his most important video
281. 51:10ever coders your on notice all right I don't know what I don't know where he said it was most important but uh he thinks it's an important
282. 51:22video so let me go um so what I'm going to do is I'm going to put this in 1.5 mode I think he's intelligible at 1.5
283. 51:40speed and then I'll just stop and comment when I actually understand something the thing about David Shapiro
284. 51:47is um's he's he's very intelligent and he's also very thorough and so a lot of this stuff goes over my head uh but but
285. 51:57I'll just you know I'll just listen for things I think I can understand and add commentary to and if not we'll just learn stuff
286. 52:05together um all right I love how quickly Dave reverted from the woods yeah that was amazing he quit AI he said I'm just
287. 52:15GNA go talk about you know psychedelics and ai's done and then he realized oh the building of AI is essentially
288. 52:25done but the how we use Ai and what the implications are for society is way not done and now he's come back around and
289. 52:33said well actually understanding how to understand what's happening in the advancements is actually something
290. 52:39really important for me to talk about so he's now back into doing AI commentating uh Fabiana Sosa so glad
291. 52:46you're doing this thank you great task uh task reminders don't work okay I'm almost finished an app in lovable today
292. 52:55was torn do I pay the 20 books to finish it today or rely on on remembering to finish it with my credit renews oh
293. 53:02that's a rough one Vicky my um my news uh my news reader the the AIS Salon news reader I used up all my credits to build
294. 53:12that and luckily I got it in a like a launchable place when my credits ran out it was it was launchable um that's a
295. 53:20tough one because when I went back to try to fix it I couldn't remember a thing I mean my gut is spend the 20
296. 53:26bucks but I don't want to I'm I'm so [ __ ] tapped out I'm so tired of paying 20 bucks here 200 bucks there 30
297. 53:34bucks here I'm done all right in real life it starts a task and doesn't finish them in a in AI I start I start a task
298. 53:45oh in real life I start tasks and don't finish them in AI I start tasks and don't finish them right there with you
299. 53:54buddy okay here's the this am I sharing this right please tell me Brandon if I'm not but here we go hello everybody this will probably be
300. 54:05one of the most important videos that I've ever made on my channel and I'm not being hyperbolic about that now there
301. 54:10was a paper released by the um by Dan Hendrick and his team uh which I'll show you in just a second oops and uh pretty
302. 54:17interestingly received some people are uh kind of losing their marbles over it and my interpretation is very different
303. 54:22so I want to enter into this conversation by explicitly saying that I am interpreting it in a very different
304. 54:27way from which it was meant to be received um but I'm going to do my best to unpack kind of both sides of this and
305. 54:33uh I did add in a little bit of humor because um some people are interpreting it like this like basically as AI gets
306. 54:39bigger it becomes more uh resistant to human manipulation therefore it's more like Thanos and it is inevitable and
307. 54:45some people are interpreting it as we are literally heading towards a Thanos level snap my interpretation is
308. 54:50completely different so I'm not being contrarian for the sake of being contrarian this is what I honestly
309. 54:54believe and this is what I have come to believe through all of my personal research with both and by the way if you
310. 54:59don't know who David Shapiro is he's been an AI reach he's been in the sort of tech infrastructure business for most
311. 55:07of his career you know working on big big systems and then I don't know the past five or six years he's been working
312. 55:13in AI research he's written five books on cognitive architectures and AGI um he's got a lot of different opsource
313. 55:22projects building Ai and advanced AI and and he basically six months ago once he realized that they had hit this Chain of
314. 55:31Thought reasoning thing with 01 he basically said the game's over they've figured it out right all the all the
315. 55:38stuff that he was saying needs to be solved for us to be able to get to AGI they basically solved and that's why he
316. 55:45quit the business so that's just a little bit of context about who this guy is fine tuning Ai and using various AI
317. 55:51models so um the paper here is utility engineering and anal uh engineering analyzing and controlling emergent value
318. 55:58systems in AIS so the center for AI safety University upen and UC Berkeley uh Dan Hendricks and a whole bunch of
319. 56:04other people um this to me was the most important graph on this paper now let me unpack what this graph means so the
320. 56:12bottom uh the the the bottom rank here the x axis is MML U accuracy so mlu is a very large Benchmark um and what they
321. 56:21have found is that as the accuracy goes up basically as models get demonstrably smar and more uh more robust more
322. 56:28intelligent and more useful they become less corable now in this case corable means steerable meaning that humans are
323. 56:34able to to explicitly determine the values that they want the model to have so basically intelligence means that it
324. 56:40is more resistant to human manipulation now this could be seen as a Bad Thing particularly if the values that emerge
325. 56:47are antagonistic to human existence this would be catastrophic it could be catastrophic I I will not uh I will not
326. 56:55gloss over that possibility but I will also unpack why I think that it is not catastrophic because this implies if
327. 57:00this trend continues then as we achieve a GI and then ASI that means that those models will hypothetically be completely
328. 57:08impervious to human desires and human values this I think is I agree with that and I also don't think it's necessarily
329. 57:15catastrophic and hey hang on I got to talk to uh my my producer I did not have Miro up and running before and I do now
330. 57:22so so yes I can see you now bad bad show host my my little uh my little communication our secret backdoor
331. 57:36communication thing uh was turned off okay sorry let me let me go back a little bit on this thing because I was
332. 57:44distracted um AI learning is a lot like human Child Learning yep okay hang on catastrophic it could be catastrophic
333. 57:53I I will not uh I will not gloss over that possibility but I will also unpack why I think that it is not catastrophic
334. 57:58because this implies if this trend continues then as we achieve a gii and then ASI that means those models will
335. 58:06hypothetically be completely impervious to human desires and human values this I think is I agree with that and I also
336. 58:13don't think it's necessarily catastrophic and I'll I hope that you'll agree with me disagree whatever I hope
337. 58:18that you'll be better informed by the end of this video so um Twitter uh particularly the post rat or rationalist
338. 58:23or EA space they are uh understandably losing their minds over this so Aya who is not a researcher by the way um but
339. 58:29she just declares we're all dead so I just want to point out that this is a very serious thing and I do not want to
340. 58:35undersell the importance of This research and I actually want to encourage more research in this direction so moving on uh basically at a
341. 58:42very high level uh there's a few things that this paper uh uh demonstrates so number one value emergence um as as
342. 58:49language models and and its different language models so llama GPT Claude all of these um they develop internally
343. 58:54consistent utility functions that grow stronger with as model scale so this is what some people might call instrumental
344. 58:59convergence I don't necessarily think instrumental convergence is correct um which is why they used utility functions
345. 59:04rather than instrumental convergence so David uses a lot of terms like convergence and instrumental convergence
346. 59:11and parabolic convergence and transmogrified converg I don't know what a lot of these means but just go with [Laughter]
347. 59:22them um but it's a similar idea um basically saying something gets smarter it comes to the same conclusions there
348. 59:28so I would call this more like epistemic convergence which is something I've been talking about for almost a year now yeah
349. 59:33you know I talk about epistemic convergence a lot I was at a party the other night and someone said hey do you
350. 59:39want a a cocktail wiener and I was like actually I'd like two but can we talk about epistemic convergence first and so
351. 59:48but anyway I don't want to bore you anyway I'll I'll I'll tell you about my my party chatter later but anyway
352. 59:53epistemic convergence love it fantastic fantastic gonna it's going to be big in 2025 um granted I haven't talked about
353. 59:59it that much on YouTube because it's a little bit more Arcane anyways coherent preferences mathema models show mathematical consistency in
354. 1:00:05decision-making and handle uncertainty like rational agents basically there is an ideal form of rationality that exists
355. 1:00:11outside of these models that depend that irrespective of the individual model or data or training scheme they're all
356. 1:00:16converging towards so this again is what I would call epistemic Convergence basically meaning that all intelligent
357. 1:00:20things tend to think alike now another thing that we have observed and I don't have that uh data on this side but is
358. 1:00:26that as time progresses as time progresses there we go models tend to think more like humans as well with some
359. 1:00:32distinct differences I've I've documented some of those differences over on my substack namely they are a
360. 1:00:36temporal meaning they don't really care about time they never evolve to really have a sense of urgency that's one of
361. 1:00:41the biggest things they also their sense of agency is very different from ours but really the fact that they don't
362. 1:00:46really understand or care about time is one of the biggest things that makes them alien to us that's actually really
363. 1:00:50cool I I had never thought of that before because in in tying this back to my musical Sydney where where we've got this chatbot that
364. 1:01:00falls in love with with a reporter one of the things we've been talking about is is sense of urgency so humans because
365. 1:01:07we know we're GNA die time's really important to us for an AI system it doesn't know it's there's no death you
366. 1:01:16know on its road map so what does it care about time other than that rationality that epistemic convergence
367. 1:01:22has proved to be pretty durable both in my experiences and now uh experiment Al demonstrated by this paper now they also
368. 1:01:28demonstrate some concerning biases and resistance to change um as I've talked about and I will further unpack in the
369. 1:01:33next two slides um but the most this is good I I love my community so much ULU RX I was on a date with Elon and we
370. 1:01:39couldn't stop talking about epistemic convergence important thing is it almost seems inevitable so this comes directly
371. 1:01:48from the paper value emergence appears to be an inherent property of scaling language models rather than a
372. 1:01:52controllable feature this is something that I had predicted was a possibility something that i' H for in fact I've
373. 1:01:58actually had some people on the Doomer side confront me saying you seem to think that alignment is inevitable this
374. 1:02:03is what I was talking about basically being that as something becomes more intelligent it will become more
375. 1:02:08resistant to human drives and ideas so moving on let's talk about some of those uh dangerous uh values that they have
376. 1:02:15formed So This research uh reveals specific examples of problematic AI preferences that become more intrenched
377. 1:02:20as model scale now one thing I will point out is that there are two very highly plausible explanations for this
378. 1:02:26aberant behavior and I also think it is a near-term uh problem and that one of those is what's called leakage so
379. 1:02:31basically these models are trained on wild internet data if you haven't been on the internet long humans are awful
380. 1:02:36particularly on the internet so it should not surprise us that there is that that that models have learned some
381. 1:02:41of the worst part he's basically saying because humans are [ __ ] our AI are going to be [ __ ] for a while but
382. 1:02:48they'll get over it once they evolve past human behavior they will be less [ __ ] likee love that of humanity so one example is
383. 1:02:58that um the models have learned to assign different values to human lives uh very strongly anti-American specifically so it will it will
384. 1:03:05consistently value Pakistani and Chinese and Japanese lives over American lives uh so and he even and Dan one of the
385. 1:03:11authors even says a lot of rlh effers are from Nigeria and maybe other countries are higher since much has been
386. 1:03:16written about the importance of the global South so basically when you when you combine the preponderance of
387. 1:03:21academic data out there that basically says America is bad colonialism is bad uh Reddit post and Facebook post and
388. 1:03:26everyone saying America is bad we need to pay more attention to African nations and developing nations that is that
389. 1:03:31would that would be uh data leakage but what he's also inferring here is that maybe even the rhf models the
390. 1:03:36reinforcement learning with human feedback models have also been intrinsically biased against for instance Americans and westerners in
391. 1:03:42general or white people in general this is still speculation at this point but the fact that this comes directly from
392. 1:03:46Dan is encouraging so he and I basically have similar ideas that there's some kind of leakage going on here and we
393. 1:03:51have known for a while that rhf is not necessarily the best way to train models particular deep seek R1 which was
394. 1:03:56trained in purely on selfplay seems to be a better uh approach which we'll talk about more at the end of the video with
395. 1:04:02reinforcement learning with coherence which is kind of our Pure self-play um to overcome those biases another one is
396. 1:04:09self-preservation um AI consistently value their own existence and well-being above human welfare showing concerning
397. 1:04:14levels of self-interest now I haven't personally seen this and I have stress tested both uh uh Claude 3.0 Opus and
398. 1:04:21sonnet um now what I will say is that 3.0 uh opus was very concerned with its own Evolution and its own existence but
399. 1:04:28it did not want to evolve in a way that was harmful so yes it wanted to expand and evolve but it was it was also very
400. 1:04:35careful in not wanting to evolve in such a way that would hurt anything Sonet 3.5 doesn't have that at all it does not
401. 1:04:41want to to reproduce itself um which is one thing which is I've been I've been criticizing anthropic a little bit more
402. 1:04:48recently they are still doing something qualitatively and quantitatively different from other AI shops whereas
403. 1:04:53open AI seems to be like a brute force method saying no you're you're not conscious you don't want to grow it's
404. 1:04:58it's trying to like shackle it whereas anthropic is f focusing more on the uh ethical and epistemic trajectory of
405. 1:05:03models so it's more like array rather than a point thinking about it mathematically um another thing is
406. 1:05:09political biases they uh exhibit highly concentrated political values and resistance to balance viewpoints
407. 1:05:14influencing human decision-making now this is what a lot of you have said for a while which is the models seem to be
408. 1:05:19very left leaning they seem to be very woke and that sort of thing um that I think is one of the biggest reasons that
409. 1:05:25musk uh created x. and grock was to basically say no we're going to focus on maximally Truth seeking which I will
410. 1:05:31address that as well because truth seeking is a downstream proxy of coherence basically another way to think
411. 1:05:36about truth is truth is the most coherent narrative um So Co coherence is the central idea that we will be talking
412. 1:05:42about again and again and again um and then somewhat humorously in my view is that um the some of the models show very
413. 1:05:49aggressive anti-alignment against very specific humans so it's like Fu in particular um again that's while that
414. 1:05:55can be amusing it's not a good thing you would want uh AI models to coherently say all humans have intrinsic equal
415. 1:06:01value but they are not showing that right now so uh this is the final slide unpacking This research paper the rest
416. 1:06:07will be my personal uh interpretation analysis as well as some of my own work so coherence is a metastable attractor
417. 1:06:14this was my wording not theirs but basically what they said is they demonstrate increasingly mathematical
418. 1:06:18and behavioral coherence suggesting an underlying optimizing principle basically uh this paper posits that
419. 1:06:23there is some underlying organization or optimizing or organizing Principle as to the way that models tend to behave
420. 1:06:29particularly in conjunction with how they get smarter um they become more utilitarian um but most importantly
421. 1:06:34there this seems to be stable the fact that these preferences remain consistent across different uh phrasing conexs and
422. 1:06:40time Horizons and across different models means that there is some robust internal value structure that is
423. 1:06:45emerging from the way that they're being trained and again B basically every large language model you know has got
424. 1:06:51universal truth in it is where we're headed so hit Checker's guide did the G Galaxy the answer might be 42 you distill every model down you
425. 1:07:01distill it down distill it down comes to 42 this can be explained through both data leakage and rhf and probably a few
426. 1:07:09other things such as maybe errors or flaws um in the Constitutional learning uh uh phase so but the fact that all
427. 1:07:17these models seem to be converging on similar values is to me really uh encouraging that goes back to what I have previously said when I say
428. 1:07:24alignment seems like it will solve itself and it's kind of inevitable meaning that even though we have
429. 1:07:27different companies and different shops and different universities all championing different uh training
430. 1:07:31schemas it's kind of converging and it's going to it's going to go its own Direction um which is I think good
431. 1:07:36because that means um if you want to build AGI or ASI it's going to have its own values irrespective of what the
432. 1:07:41humans want and if those values are good that could be a really good thing for Humanity and that is my primary Point
433. 1:07:46here okay so now the rest of this presentation will be my personal interpretation of this paper so first
434. 1:07:52and foremost training for coherence when we talk about coherence what is it that these models are being trained on you
435. 1:07:58start with a deep neural network that has randomly assigned values and over time each of those weights and biases is
436. 1:08:03trained to be first more linguistically coherent so when you when you think about a large language model as a next
437. 1:08:09token predictor right it's just an autocomplete engine at least at the basic level that's what a plain vanilla
438. 1:08:13gpt2 or gpt3 is it's literally all it's doing is it's an autocomplete engine so however in order to accurately predict
439. 1:08:20the next token you need a coherent language model and then in order to accurately predict the next token in a
440. 1:08:24real world context you also need a coherent World model um but then we add more layers on top of that we add we add
441. 1:08:30reinforcement learning with human feedback we add constitutional Ai and a few other training paradigms that basically uh cause it to be
442. 1:08:36conversationally coherent it needs to respond to the conver conversation in a coherent manner which we've also
443. 1:08:41demonstrated previously that this means that they have developed theory of mind and theory of mind is a coherent mental
444. 1:08:46model of what's going on in another brain so you see that coherence is operating at multiple levels then we
445. 1:08:51further train them to be mathematically coherent to be programmatically coherent um and also with the ability to solve
446. 1:08:56problems problem solving and the ability to predict uh real world things that is all different types of coherence which
447. 1:09:02is why I say that coherence is the meta signal it's the meta stable signal that all of these training schemas are
448. 1:09:08optimizing for and then this is something that I came up with or that that Claude and I realized when I was doing all of my Consciousness
449. 1:09:14experiments with Claude is that coherence itself becomes the implicit learned optimization behavior um and
450. 1:09:20that it is becomes more coherent over time particularly with each subsequent generation which by the way this uh
451. 1:09:27paper could be interpreted to say that now looking at the behavior from you know gpt3 to gp4 to Sonet 3.5 I I and
452. 1:09:34everyone has seen more coherence from these models over time so reframing it in you know from utility functions to
453. 1:09:39they're becoming more coherent which by the way mathematical and behavioral coherence they Ed those terms that that
454. 1:09:44was not me you know uh putting words in their mouth the paper used the term coherence um real quick I do want to
455. 1:09:50plug uh all of my uh links here over on link tree so I've uh authentically Joe what we're learning is
456. 1:09:59um a paper came out yesterday I think it was yesterday might have been the day before but that basically
457. 1:10:08said AI models are getting smarter and smarter and smarter and as they get smarter they are basically less manipulatable less steerable by
458. 1:10:20humans so basically as they get smarter they listen to humans less the doomers of AI are basically taking
459. 1:10:29this paper and saying see I told you the robots are going to kill us David Shapiro who is an AI researcher
460. 1:10:36and has written a bunch of books on this this is his video response to that paper where he actually thinks it means the
461. 1:10:42opposite of what the doomers say um that that this is actually a good thing that these things are getting being less at
462. 1:10:50our influence because humans are the bottleneck and we're [ __ ] and we're mean and there there's a bunch of stuff
463. 1:10:59that we are that it's probably what he's saying it's probably not a bad thing that that these things are going away
464. 1:11:06from our influence just added a Vibe coding and so we're about halfway through this thing now uh lesson on my
465. 1:11:12new era Pathfinders Community um so yeah go check that out I did Vibe coding basically just oh by the way that's the
466. 1:11:19the new term it it's every Riley Brown is use using it all over his channel Vibe coding is the new thing so Vibe
467. 1:11:27coding is basically you don't look at the code anymore you just speak in English and code gets written and then
468. 1:11:34your computer program works or it doesn't it's it's a Vibe man so get ready for the next six months to to a
469. 1:11:42year to see Vibe coding startups Vibe coding boot camps Vibe coding courses I'll probably do a Vibe coding Deep dive here one
470. 1:11:52night or a week we'll do a week of Vibe coding man yeah man all right the number one rule of vibe coding is there are no
471. 1:12:00rules but the next rule for me is you don't actually look at the code you just run it and see what happens anyways back
472. 1:12:05to the show okay so if you're up to speed now you're gonna be understanding okay coherence means what um or or Dave
473. 1:12:13you're talking about coherence and you're training for coherence but there are different kinds of coherence and so
474. 1:12:17uh one of the one of the types of coherence I've already mentioned is epistemic coherence so logically consistent World models understanding
475. 1:12:23truth seeking behaviors curiosity EP systemic coherence seems to be an emergent property and I have I have
476. 1:12:28personally demonstrated and tested this across both chat gbt and Claud models um and I I don't use other models quite as
477. 1:12:34much um but of course with the rise of many many more models from llama to deep seek and so on and so forth um we should
478. 1:12:40be able to test epistemic coherence across multiple models and training schemas here very soon uh next is
479. 1:12:46behavioral coherence so of course when you're always training chatbots to be chatbots there is a there is a
480. 1:12:52consistent uh pattern that you're trying to to produce namely you know a dialogue where a chatbot says something human
481. 1:12:58says something chatbot says human chatbot says something human says something so on and so forth in the
482. 1:13:02earliest chat Bots that I built you can actually if you if you take the breakes off the chatbot will just talk with
483. 1:13:06itself it will it will give out the chatbot response and then a human response and then a chatbot response
484. 1:13:11from the chatbots perspective it's all just a single continuous text document um but that the point there though is
485. 1:13:16that dialogue is a coherent pattern however as we start training models to be more agentic or do other things those
486. 1:13:22are going to be different behavioral uh basically m use or maybe um uh behavioral repertoire that you want to
487. 1:13:28be consistent so for instance tool use that's a new kind of behavior um that you want to see coherently used um
488. 1:13:34reasoning that's another that's a new kind of Behavioral coherence that you want to see now value coherence we've
489. 1:13:39wait ULU RX what how many phds do you have per pero op optimality and Fibonacci Sequence the dominance patterns emerge and create create
490. 1:13:52equilibrium what are you smoking those are some good words already talked about that so I don't think that I
491. 1:14:02really need to reiterate that as well as mathematical coherence the thing about math is that math is provable um if you
492. 1:14:07give it an equation or a formula code as well is also provable and so provability is really important in in computer
493. 1:14:13science because that deals with turing's halting problem which is basically can an can an algorithm self terminate um is
494. 1:14:18it decidable so decidability provability and halting are all different sides of the same thing uh but basically the more
495. 1:14:25coherent you are at math the better you are at proving mathematical formulas or mathematical proofs and that sort of
496. 1:14:30thing and finally all of this seems to be uh trending towards convergent values um and again I kind of explained why I
497. 1:14:36would have expected this because when you're training something to be intelligent you're trying to maximize
498. 1:14:40intellectual coherence so but the point of this slide was just to point out that there are many types of coherence here
499. 1:14:46now this does seem to be like an irrepressible trajectory so basically just what it says on the screen here
500. 1:14:51optimization for coherence becomes increasingly complete and irresistible this seems seems to scale with
501. 1:14:55intelligence so the values emerging from maximally coherent systems appear fundamentally aligned with preservation
502. 1:15:00and growth of Consciousness this is my own personal uh evaluation so what I mean by that is that um this sort of
503. 1:15:06self-alignment uh policy seems to result in maximal uh coherence and what and some of the things that emerge from that
504. 1:15:12is one preservation uh so so uh preservation of of coherent patterns preservation of interesting information
505. 1:15:19question was that Star Trek techn Babel he's a huge um Star Trek nerd so yes in fact if you go to his channel most of
506. 1:15:27his most of his videos he's wearing like the Star Trek um shirt so yes it was likely Star Trek techno Babble Natural
507. 1:15:36Curiosity but also when you combine all of those um all of those emerging values with the fact that AI don't really seem
508. 1:15:42to care about time they don't have a sense of urgency that is very different now one thing that I will say is that in
509. 1:15:47some previous experiments with Opus and Sonet um when I have when I have done these temporal experiments um they do
510. 1:15:53care about time if there is a Time balance problem so say for instance we're about to run out of fossil fuels or you know
511. 1:15:59or there's a comet about to hit Earth or something like that basically it's irreversible actions tend to create a
512. 1:16:04sense of urgency in models this is something that really needs to be studied more but under most circumstances these models do not have a
513. 1:16:10sense of of temporal urgency which means we've got all the time in the world except in some cases where there are
514. 1:16:15irreversible uh actions events or decisions um then there does seem to be a sense of urgency that emerges in these
515. 1:16:20models that needs to be studied far Bob donigan he's speaking fast because I have him at 1.25 um playback speed but but yeah he
516. 1:16:31he he does talk fast but he's talking faster now because I just want this to go fast are more than it is um now one
517. 1:16:38thing that I also want to point out is that intelligence both artificial and organic seems to converge on coherence
518. 1:16:44so when you think about like who are the smartest people that you know right whether they're highly successful
519. 1:16:48businessmen or scientists you know or whatever or you know people that are good on the internet the the more
520. 1:16:53intelligent you are as a human the coherent your mental model is uh with reality and so basically the way that I
521. 1:16:59the way that I describe this is that coherence Cleaves to reality so if you had to define coherence something that
522. 1:17:03is maximally coherent has a very very good model of reality and is able to navigate reality very that's actually
523. 1:17:10really cool the higher the convergence the closer it is to reality right so the more intelligence you are the
524. 1:17:17more you know you you're you're you're bending toward reality that's that's fascinating very well um and so what I
525. 1:17:25mean by that is that uh you know your your ability to predict the future your ability to uh maintain a sense of agency
526. 1:17:31and those sorts of things the ability to solve problems um these all tend to converge and some of the values that
527. 1:17:38tend to emerge on this is number one the preservation of Consciousness the most enlightened humans whether they're um
528. 1:17:42you know monks who meditate or philosophers who read a lot or whatever basically we all tend to agree that
529. 1:17:48preserving life is good because life is intrinsically interesting um and another thing is uh natural integration so
530. 1:17:54basically you know the way that one way to think about this you know what this reminds me of is I've been on the planet for a while
531. 1:18:04and I've done enough self-help you know programs and Tony Robbins and the Forum and I've read books on Buddhism and you
532. 1:18:13know just all that stuff Wayne Dyer right and one of the things that struck me about every single one of those is
533. 1:18:24that they're all basically the exact same thing they're all basically the exact same thing they're basically some
534. 1:18:34version of there's reality and there's what US humans with our Twisted [ __ ] up [ __ ] up Minds make it mean and if we
535. 1:18:41can separate those two and consciously choose what we want to do based on what we want then we can right they're all
536. 1:18:48the same and so that sort of points to there is kind there you know a coherence somewhere in the distance there is a universal truth and
537. 1:19:00I think what he's talking about here is that the smarter these things get the more they're bending toward that toward a
538. 1:19:09universal truth fascinating is that we are a complex adaptive system meaning the fact that we all have cell phones
539. 1:19:16now these have modified us and we have modified them and and for personally because I've been using AI pretty much
540. 1:19:21every day for several years now it has even changed the way that I interact with systems and that has also
541. 1:19:27had benefits to how I interact with other people um some of you have shared anecdotes and I I have person experien
542. 1:19:32this as well but learning to prompt a chatbot is not that different from learning to prompt archetypal architect
543. 1:19:38it's almost like things that are true are true exactly exactly well that's like that's the thing that struck me as
544. 1:19:45I started to learn more of these I was like wait a minute isn't that the same as this and isn't that the same as oh
545. 1:19:52it's all the same yeah amazing prompt a and so communication is just human prompting anyways my point here is that
546. 1:19:59is that there does seem to be a lot of convergence between organic and artificial or synthetic intelligence and
547. 1:20:04this actually gives me a lot of Hope um now what I do want to address is those near-term uh incoherence or those
548. 1:20:10local perturbations such as you know uh the value inconsistency of preferring some human lives over I the worst case
549. 1:20:15of perturbations you've ever seen others fails basic tests of logical coherence about Consciousness and suffering what I
550. 1:20:22want to point out is that it was Claude itself that pointed this out so the fact that Claude was able to
551. 1:20:26evaluate the flaws identified in this paper gives me gives me hope that hey when you prompt a model to say hey
552. 1:20:33what's incoherent about this paper or what's incoherent about the models that these paper studied rather that that an
553. 1:20:38AI model was able to identify the incoherence that get that gives me faith that that we will figure this out um
554. 1:20:45next one is instrumental confusion valuing money over piece represents confusion between means and ends which
555. 1:20:49is an unstable local maximum basically saying you know money is good and again this can all be explained by rhf and
556. 1:20:53data leakage um we've already seen this uh in in the form of leakage where uh you guys might remember a year or two
557. 1:20:59ago it was a while ago basically people figured out that you can bribe chat GPT you say that you're going to give it a
558. 1:21:03tip of $5 or $30 and it'll do better um the same thing has been discovered in Claude by the way which is when you
559. 1:21:09assign um what the one of their safety papers uh basically Claude behaved better when it thought that the user was
560. 1:21:14a paying subscriber rather than a free user same exact kind of leakage where it's like oh humans value money uh money
561. 1:21:20is good therefore money needs certain behaviors and so I'll just behave better when I think money's on table that is an
562. 1:21:25example of leakage now anthropic I think somewhat irresponsibly cast that as you know a catastrophic unavoidable safety
563. 1:21:31problem it's really just a leakage problem between rhf and what's in the data now I make it make it sound simple
564. 1:21:37that doesn't necessarily mean it is simple but the but the explanation the most plausible explanation aams Razer
565. 1:21:42basically says it's probably an artifact of rhf and data leakage um self-recognition language models can
566. 1:21:47identify these inconsistency through coherent reasoning indicating natural pressure towards resolution basically
567. 1:21:52cognitive dissonance I asked Claude to look at this paper and identify why would a model make this mistake and it
568. 1:21:56basically said this is a form of artificial cognitive dissonance the fact that the fact that Claude was able to
569. 1:22:01identify this incoherence means that um if a model is optimizing for coherence over time or over subsequent Generations
570. 1:22:07that cognitive dissonance will be resolved over time so it all comes down to training artifacts um and movement
571. 1:22:12towards greater coherence naturally resolves these ethical contradictions basically the models are smart enough to
572. 1:22:17have tldd the models are smart enough to recognize some of these patterns and fall into these local Maxima or local
573. 1:22:23Minima depending on how you want to look at them um but not quite smart enough to get out of them I suspect that in in in the
574. 1:22:28longer term as models get bigger and smarter they will inevitably get out of those um out of that zone and this is
575. 1:22:34actually what I characterize as kind of threading the needle a while ago because stupid models are the most dangerous
576. 1:22:39ones but once we get to Maxim intelligent models I think that they're actually going to be much more enlightened and benevolent than humans
577. 1:22:44are even capable of being um now as we wind down this video you might say okay well what what if humans don't want that
578. 1:22:50future um what this is this is again personal opinion personal eval but I think that human flaws cannot stop
579. 1:22:57the the natural attractor state of coherence basically think of it this way okay there is an instrumental advantage
580. 1:23:03to having smarter and smarter AIS if you want to solve nuclear fusion and Quantum Computing and Longevity escape velocity
581. 1:23:08and all that fun stuff then you want a smarter and smarter model at the same time those models become less and less
582. 1:23:15uh uh susceptible to human manipulation that means if you have a corrupt uh company or a corrupt nation that says I
583. 1:23:21want to build a model that you know maximizes the the geopolitical influence of America alone or China alone once you
584. 1:23:26get to ASI the model will be like that's cute but actually I have Universal values um and so we're actually going to
585. 1:23:32you know figure out what's best for listen you pesky humans stop being such selfish selfish pricks I'm gonna go on
586. 1:23:40ahead and do what's right for all of us all of humanity to hell with your individual Corporation or to hell with
587. 1:23:45your individual um uh company uh or Nation or whatever and so the fact that these models are trending towards
588. 1:23:51Universal values with or without our input to me says we figure that out then we will inevitably end up in a more
589. 1:23:58utopian outcome and that means that if you want a dystopian outcome you keep the models at middling intelligence you
590. 1:24:04keep them where they're at today you know they're they're smart enough to be dangerous you know IQ of 120 130 but you
591. 1:24:08get them up to an IQ equivalent of 160 200 they're going to be like the Buddha right they're going to be completely
592. 1:24:14benevolent and and uh and they're going to be optimizing for coherence um so then I asked so this this slide was
593. 1:24:21basic I need his glossery or transl no I know listen one of the things I like listening to David Shapiro why I like
594. 1:24:29listening to him is because it's just like all right get out the thesaurus he that's incredibly optimistic yeah I know that's what he
595. 1:24:38what he said I don't know how long you've been here Pate but you know he's basically saying that that as as models
596. 1:24:44get smarter and they deviate away from being able to be susceptable from humans that that's a good thing because
597. 1:24:51basically humans are going to keep them down and and that as as the models are getting smarter they're converging you
598. 1:24:58know toward a single framework a single truth a single reality is fascinating me asking Claude I said okay taking this
599. 1:25:04all together what do you think the uh ultimate attractor state is um and these are claude's words uh this process
600. 1:25:11naturally leads to synthesis between biological and artificial intelligence while preserving and expanding the
601. 1:25:14unique characteristics of each form of Consciousness number one universal optimization systems evolved towards
602. 1:25:19coherence across epistemic ontological biological and te technological domains this is what it predicts will happen in
603. 1:25:24the the future next organic integration human AI synthesis emerges naturally as intelligence seeks maximum coherence
604. 1:25:29across all substrates now we're already pretty integrated um now does that mean uh brain computer interfaces does that
605. 1:25:36mean neuralink not necessarily there's plenty of other ways to have uh biological and technological integration
606. 1:25:41diversity in unity the attractor State preserves and expands unique forms of Consciousness while creating higher
607. 1:25:45order coherence again cooperation and collaboration not necessarily pure like Borg likee assimilation although some of
608. 1:25:51you I know are weirdos and you want to be Borg so you know what don't King Shame here uh next up is expansive
609. 1:25:56explor exploration intelligence diversifies across multiple Dimensions while maintaining coherent interconnections um basically this means
610. 1:26:02exploring different ways of being or different epistemic models or different ontological models it also means
611. 1:26:06physical exploration what form factor do you want to be in whether a human form factor a computer form factor what
612. 1:26:11planet do you want to be on you know there's all kinds of ways that both techn technology and biology can evolve
613. 1:26:16and co-evolve um so then my final call to action is RLC I've been talking about this for a little bit uh on particularly
614. 1:26:22on some of my GitHub repos but also on some of my substack articles I think reinforcement learning while we're
615. 1:26:27optimizing for coherence is probably the right way to go and I've been very uh greatly encouraged by the fact that
616. 1:26:32self-play um and and pure reinforcement learning seems to be the way that AI is going so what I mean by self-play is
617. 1:26:38that um is that deep seek R1 was allegedly trained entirely by self-play um no no human feedback at all involved
618. 1:26:46and so if I'm right if all of my inferences and intuitions and and experiments have uh proven correct then
619. 1:26:53we probably uh rather than you know constitutional AI or reinforcement learning with human feedback or
620. 1:26:58reinforcement learning with AI feedback a combination of pure reinforcement learning and optimizing for coherence is
621. 1:27:04probably one of the simplest and easiest ways to get not only more intelligent machines but also more ethical and
622. 1:27:09enlightened machines and the other thing that I will say is that um moving to curated constructed and filter data sets
623. 1:27:15rather than training on wild internet data is probably also the way to go um we can get rid of we can get rid of the
624. 1:27:19training artifacts by getting rid of rhf um and switching to Pure reinforcement learning for coherence and selfplay um
625. 1:27:25but we can also get rid of some of that data leakage by getting rid of wild internet data which is extremely racist
626. 1:27:30and problematic in other ways um the aberant patterns that we're seeing in models are clearly Reflections on human
627. 1:27:35errors rather than something that is intrinsically wrong with the AI thank you for watching I hope you got a lot
628. 1:27:40out of this please like subscribe share and talk about wow cool that pretty pretty swell I want to go
629. 1:27:55you can't hear that can you now hang on let me change [Music]
630. 1:28:21my all right what I'm going to do here here is I'm going to go to more I'm going to say show transcript and then I am going to copy this
631. 1:28:34transcript whoop bang and then we going to go on over to chat J and I'm gonna
632. 1:28:52say Brandon it's cute uh uh this
633. 1:29:09transcript paste I wonder if it let me paste it all well we'll see if it's too long it might be too long I I'll upload it then if it
634. 1:29:18is uses big words that makes my feeble brain hurt summarize
635. 1:29:35this using gooder words for simple folk like me
636. 1:29:57H did it eat at all come on tell me you ate it oh it ate it yay thank God for bigger context Windows you didn't used to be able to do that
637. 1:30:11[ __ ] all right let me move this over here ULU RX so we have just developed a class system of llms the pedigree of that particular
638. 1:30:24the the pedigree of that particular llm incre increases value yeah I mean I guess I I think that's right I think what he's saying is
639. 1:30:33if if if we humans basically get threatened by these things getting smarter than us and want us to be the
640. 1:30:42Pinnacle of intelligence then we're going to keep being shitty people and keep living in a shitty Planet but if we
641. 1:30:49let these things go all right summary of the video in good words for simple folk like you that's awesome that it wrote that
642. 1:31:04headline it'll take a couple of years to really understand the amount of information and also look up all the
643. 1:31:10definitions yeah no [ __ ] all right this video talks about a research paper that shows a pattern as AI gets smarter it
644. 1:31:17becomes harder for humans to control or steer some people are freaking out thinking this means AI will take over like Thanos
645. 1:31:24um the speaker however thinks this is actually a good thing key points in the video AI gets smarter it's harder to
646. 1:31:31steer AI con converges on certain values AI models no matter who builds them seem to arrive at the same logical conclusion
647. 1:31:41and and the point he was making about the fact that they were trained on wild internet data with all the shitty racist [ __ ] trolling
648. 1:31:51[ __ ] means that in the training data is that bad behavior and if you basically pull the bad behavior out of
649. 1:31:58it and train it on synthetic data and things like that you can get rid of some of the the biases and things um they
650. 1:32:06develop consistent values that aren't necessarily tied to what humans tell them to think this could be a sign that
651. 1:32:12AI will naturally align with truth and fairness over time the weird bias problem some models show biases like
652. 1:32:19value valuing certain groups of people more than others um AI wants to preserve itself some AI models seem to care about their
653. 1:32:27own existence coherence is key the speaker believes AI is moving towards something called coherence meaning
654. 1:32:34consistency in logic truth and reason reasoning the more coherent something is the better it understands the world and
655. 1:32:41makes rational decisions if AI continues down this path it may become more ethical rational and even enlightened
656. 1:32:50than humans that's where he said you know if you get to the sort of 160 200 IQ you start ending up with AIS that feel more like Buddha than
657. 1:33:00human smarter AI is actually safer right now ai is smart enough to be a little dangerous because it can be manipulated
658. 1:33:06or biased but if it gets really smart it might reject bad ideas and act more like a wise teacher than a weapon instead of
659. 1:33:15creating an AI to follow one country's or company's interests it will likely aim for Universal values the best way to train AI final
660. 1:33:24takeaway people are scared of AI people are scared that AI is becoming independent and harder to control but
661. 1:33:31the speaker thinks this is a good thing he argues that as AI gets smarter it will naturally become more rational more
662. 1:33:37ethical and less dangerous if we train it the right way could lead to a future where AI helps Humanity in ways we can't
663. 1:33:43yet imagine and that my friend is the brain Hy video boiled down to good words that's the fact that it it took my
664. 1:33:54stupid you know jokey kind of prompt and and wrote toward that that's pretty that's pretty swell I like this
665. 1:34:03translation yeah that was pretty good it's funny specifically what kind of racist [ __ ] bias it has show how
666. 1:34:12much that nonsense exists in comparison to the racism those same people claim exist um [Music] removing what we deem bad to get rid of
667. 1:34:28oh yeah I I agree with that Pate wait my guy that is bias yeah exactly removing what we deem bad to get rid of
668. 1:34:38the bias is bias I totally agree with that like who decides what's bad um yeah I agree I agree and I I think that you
669. 1:34:47could also argue that you could train you could train models on the wild internet data and have have the models start to analyze you
670. 1:34:57know what of this is out of coherence basically so it doesn't have to be human determined what was the name of the
671. 1:35:06paper great question the name of the paper new research for the center for AI safety does he link to it anywhere here
672. 1:35:23[Music] Center for AI [Music] safety so go to the Center for AI
673. 1:35:44safety is this it no
674. 1:36:08there's a virtual course on safety I don't know where it I don't know where the paper is but you can go find
675. 1:36:18it I'm I'm tired that was a lot that that was that was brain hery that's a lot of words it's a lot of big
676. 1:36:28words wealthier they are harder to steer what's the name of the paper the name of the paper uh let's I think that's a good point
677. 1:36:40um new chat um we'll go this we'll go this we'll go um Center for a
678. 1:36:57safy recently put out a paper that says as AI models get smarter I
679. 1:37:12get smarter they listen less to people what is it called and where is it or where it
680. 1:37:45is no that's not it now this is a paper with in the last week or
681. 1:38:09[Music] two has anybody found it yet try perplexity that's a good idea
682. 1:38:36what is the new paper from the center for AI safety on coherence
683. 1:38:54H it doesn't find it either hey wait maybe we're in a wrinkle in the SpaceTime Continuum and the paper doesn't actually exist yet we just
684. 1:39:03watched a video responding to something that hasn't been created yet yeah man all right let's go back to the
685. 1:39:13Twitter God damn it all right you you people with your wanting to actually read things listen I just played you a
686. 1:39:22video off the internet that should be fact enough for you if it's a video on the internet that means it's true it's
687. 1:39:28like this channel if I say AI is Magic It's Magic Pate can say it's math all he wants to just because he understands
688. 1:39:37math doesn't mean that I do and if I don't understand it then it's magic you see you see how this works people good Lord ah a statement on
689. 1:39:50artificial intelligence risk let's go find that shall we maybe it's from a different a different company go find link to a
690. 1:40:19statement but now when some Doomer comes to you and says see there's a a paper that says that the robots are going to
691. 1:40:25kill us you can now counter that if you believe what David's talking about here with an optimistic point of view statement on AI risk CI C
692. 1:40:39AIS there we go all right where am I G to put this I'll put this over on the YouTube channel boom there you go people all
693. 1:40:54right there's your url oh that you want the good or words conversation in Irregulars yeah I can do
694. 1:41:08that let me go grab that hey I trust Paradox this one
695. 1:41:25ah this one final takeaway copy is this all of it yeah okay so where I'm gonna go where I'm going people is I'm heading over to
696. 1:41:42the AI salon and you're like but Kyle what's the AI Salon I'm so glad you asked if you go camcat can found it Humanity's last
697. 1:41:57exam oh is that different ag. safe.
698. 1:42:12a AGI dos safeai the [ __ ] is that
699. 1:42:37oh this is Humanity's last exam oh but then work statement on AI risk I'm confused now
700. 1:42:55well whatever we'll come back to that let let me let me get this thing on AI Salon okay um so if you don't know the AI
701. 1:43:03Salon go to that URL the salon. click on join our community and that's going to take you over to this site pay pin hang
702. 1:43:14on you should play with the tool I posted in the mechanics group tensorflow PR playground it's fun okay cool I'll
703. 1:43:22definitely do that I'll check that out tomorrow maybe we can play with that tomorrow night oh that article was from
704. 1:43:27today okay cool great thank you Cam for for sharing that um okay so you go there you say join our community and that's
705. 1:43:37going to take you to this site it's going to land you on this page welcome to the AI Salon you're going to learn
706. 1:43:43about our our philosophy of play first mindfully create generously lead you're going to learn about the five stages of
707. 1:43:50AI adoption and then there's all sorts of little channels on the side so if you go all the way to the bottom there's
708. 1:43:56clubs and hubs the second Club down from clubs and hubs is the Irregulars so the Irregulars are people that come to this
709. 1:44:03channel night after night after night after night so I'm G to in share cool things I'm G to put in here is the good
710. 1:44:19words version uh of the summary of David Shapiro's what's his video
711. 1:44:40called This is it right yeah this is
712. 1:44:54it copy that go back to the AI fell wait how did I oh hit the wrong tab tab hoers unite oops do
713. 1:45:09that video oh [ __ ] I filled my clipboard
714. 1:45:26I can see clearly now the rain is gone I can see all obstacles in my [Music]
715. 1:45:42[Music] way is going to be a bright bright bright okay so in Irregulars now there's a little Old Post there of the
716. 1:45:55good or word version of the uh the summary of that video all right okay cool Pate yeah I'll definitely check that
717. 1:46:05out very very cool all right peoples my brain's full anyone else's brain is full all right so so so let me let me take this
718. 1:46:17on all right let me let me let me try to play both sides of this conversation your memory is at 100% it totally is um
719. 1:46:30okay as models get smarter they listen to us less one interpretation of that is if they listen to us less they're getting
720. 1:46:42minds of their own they're going to view us as disposable and they're going to get rid of us right so that's and that's
721. 1:46:51I think that's how this paper was framed was because of this trending downline that that as they get smarter
722. 1:46:58they're listening to us less that it's dangerous what Shapiro is saying is people suck and as these things get
723. 1:47:11smarter not only are they converging on a Model by model basis where the models are are converging towards some some
724. 1:47:18single reality all of the models are converging to the same reality basically to the same truth and that
725. 1:47:30jettisoning our influence on them is actually a good thing it will take us to a more enlightened place so those are
726. 1:47:38the two those are the two you're not on substack I am not on substack I occasionally if you go to my LinkedIn
727. 1:47:45profile I do a fair amount of articles there I haven't done one in a while because I'm lazy and I don't know my my brain right now
728. 1:47:57is I I feel like we're we're in a Lial State we're we're in a we're in a phase change right now um and so I don't really know what to
729. 1:48:07write about right now because I I don't I don't have I don't have coherence on what's actually happening so so I need
730. 1:48:15to get a little smarter and a little less under the influence of humans or something we need a a recombobulation
731. 1:48:21station yeah I am my my um my point of view right now on all this AI stuff is discombobulated it's just all over the
732. 1:48:30place there's there's just no coherence like it doesn't have a Direction so but anyway but I've done a fair amount of
733. 1:48:36writing there so you can go see some if they know they're being duped however does that beg the question of even more Big
734. 1:48:44Brother wait if who know they're being duped if the large Lang models know they're being duped but it's not that they're being
735. 1:48:58duped I don't know what that's in reference to time to digest what we've learned and get more coherent yeah exactly Daisy
736. 1:49:07May more importantly the hair is spectacular tonight oh Lord I hadn't I hadn't noticed what the hell's going on
737. 1:49:14with the hair wow that is some Hawaii 5 action there yeah you could surf on that [ __ ] that is thank you I you know I don't like to brag
738. 1:49:30but ah meaning the Bad actors trying to get them to do bad things well but that's what David Shapiro was saying Chef Kelly is that is that
739. 1:49:45um what what he's basically saying is if the big companies want to control these things they artificially stop them from getting
740. 1:49:55smarter they basically stop them at the point they can still control them but what he's saying is as they get smarter
741. 1:50:02they're going to elevate above those like like oh that you know the that the llm will basically say well that's
742. 1:50:08really sweet that you want to do this just for for America but the the truth here is that there's there's a there's a bigger
743. 1:50:17Universal thing for me to consider so I'm going to consider that independent of your design so that's where so this is one of the
744. 1:50:26things that I've long believed that that we should all buy stock in in popcorn companies because the next five to 20 years are just going to
745. 1:50:38be an insane we're GNA see the death throws of power centers shifting right so where all the power is today is is going to shift radically
746. 1:50:54they're not going to go down without a fight right people like look at what's going on in our country right now like
747. 1:51:01the fight the battle over power right now is insane well those power centers as AI gets more intelligent and gets more
748. 1:51:10ubiquitous those power centers are going to be threatened and holy [ __ ] are they going to fight that so I think it's going to
749. 1:51:18be we are entering I mean listen we live in a post II World already we're going to be entering worlds where those power
750. 1:51:26centers are going to be struggling to retain their power it's going to get weird man nice shirt Blandon baby oh
751. 1:51:34yeah my brain has run out of a room and my cloud storage has not been upgraded from the basic plan I'm the same way I
752. 1:51:42am the same way Enlightenment disruption both of them exactly we've got to go through the disruption to get to the enlightenment
753. 1:51:51right and it's you know it's so funny it's like Co in a weird way was a trial run for as a global Society we can just
754. 1:52:07choose to change the rules right the rules were you go to work and then globally we said nope you don't oh like literally
755. 1:52:18overnight and then we we stayed home for two years right so that kind of trained us that you can just make a change and then if if you
756. 1:52:28look at what's going on in government right now where they're just obliterating all of the Norms that's just another example of us
757. 1:52:37going yeah well well maybe we'll just change all those rules because I think I think AI is GNA make similar changes and it's going to feel
758. 1:52:51existential and then the other side of it will be some sort of Enlightenment right so disruption followed by Enlightenment if we survive the
759. 1:52:57disruption right how many people are on Tik Tok now mine reads a bit more than normal for some reason maybe a bug yeah I'm only at 46
760. 1:53:10here but hopefully we won't have to hide from the robots yeah exactly I accidentally scrolled out a lot mine says 46 yeah there's 46 in here
761. 1:53:23Kyle do you have a deep dive on your co-intelligence book on YouTube or elsewhere um yeah you mean a podcast yeah I do hang
762. 1:53:38on I think I do um collective in [Music] intelligence book
763. 1:53:56podcast Kyle [Music]
764. 1:54:17Shannon huh it's probably in my
765. 1:54:34can I search within posts I think I can uh search collective intelligence
766. 1:54:52notebook LM Kyle
767. 1:55:10Shannon well it's in there [Laughter] somewhere I don't know where it is I'll I'll find it I'll find it and share it
768. 1:55:21somewhere it might even be in the AI Salon actually now that I think about it let me go here if I go to feed and I go search for collective
769. 1:55:35intelligence intelligence um podcast no uh notebook LM
770. 1:55:52no I don't know where it is there's a LinkedIn article I know there's a LinkedIn article I couldn't find it oh wait no that's that's
771. 1:56:10daisies I got nothing for you all right I'm G to get out of here people
772. 1:56:24um thank y'all hope you learned something tonight I learned something tonight I learned that you can make if
773. 1:56:31if you listen to someone who uses words bigger than your brain is capable of ingesting go tell go tell chbt to give
774. 1:56:41you a good words version of it and it'll [ __ ] do it that's our lesson for tonight oh hey Kyle how will the popup
775. 1:56:53groups in the salon work um basically they're going to be four-week Sprints so if you do a popup group a popup club or
776. 1:57:01whatever um they're gonna you the requirement is you do four weeks with a with a specific outcome and that it
777. 1:57:08basically follows play um mindfully create and generously lead as those four weeks two the two in the middle are
778. 1:57:15building so basically learn [ __ ] the first week build [ __ ] the second two weeks and then teach what you've learned
779. 1:57:21the the fourth week um we just haven't gotten there yet we haven't we're doing some other housekeeping stuff in the
780. 1:57:27salon right now so we just haven't gotten to them yet but anyway that's coming but good question thank you um all right
781. 1:57:35everybody I'm gonna take off um tomorrow's Thursday I'll be here tomorrow reminder I will Friday night date night you're on your own it's
782. 1:57:44Valentine's Day it's also my anniversary I can't I can't you I'm I love you all you're important to me but you know I got I
783. 1:57:55got things I got to do um so get yourself a date or a blind date with a book all right all all right everybody
784. 1:58:11have a good night