8/10/2026 - AI’s Future: Personal Agents, Local Models, and the Fight Over Who Controls Them

Transcript

1. 0:03[music] [music] [music]
2. 0:21[music] [music] [sighs and gasps] Good evening good people. Happy Monday. Happy happy Monday. Uh welcome to the AI
3. 0:39learning lab. Um let me write something here. Um, host.
4. 0:55Okay. Um, [cough and clears throat] welcome to the AI learning lab. Welcome to the AI Salon. Uh, if you're new here,
5. 1:05um, the AI salon is an awesome community that you should, uh, take advantage of. One of the things that I talk about a
6. 1:14lot and I've talked about since the beginning we founded the AI salon the week that Chat GPT came out because I
7. 1:21saw Hey Danielle I saw Chat GPT come out and I lost my ever loving mind. I was like this is going to change
8. 1:30everything. This is going to change the world. And one of the things that I knew was that when you have something that's transformative,
9. 1:39really transformative, nobody knows anything. Everyone's trying to figure it out. And the only way you
10. 1:45can really figure it out and be in the loop is to be in a community of curious people also trying to figure it out. Um,
11. 1:52I know less today than I did back then. I mean, I know a lot, but it's changing so fast that I don't have a flipping clue.
12. 2:01Um, okay. Beautiful. Fantastic. Thank you. Thank you. Hey, Mel. What's shaking? What's going down? Um, so it's Monday. Monday I will just
13. 2:12generally chat about stuff. If you have questions or things you want me to cover, put that over there on the side. Um,
14. 2:21there's there's a fair amount of news items that are worth worth going over that I'll talk about. Um, couple of
15. 2:28housekeeping things. If you want to chat during the live down in this corner right here, if you click on that button,
16. 2:35it'll pop the chat out and you can chat and I can see it and Brandon can see it and the other people can see it. Um, and
17. 2:42we can grab those comments and put them on screen, things like that. So, if you want to chat, do that. Um, you can also
18. 2:48open up a second tab where you've got the this live stream in one tab and the AI salon in the other other tab. If
19. 2:55you're new to the salon, I'd encourage you to just go check things check things out over there. And uh and yeah, that's about it. Um
20. 3:06where do I want to begin tonight? Well, one thing I want to do is I want to go look at the letter that that Zuckerberg
21. 3:22put out today. um he's he's not been super good at um predicting the future. I think I think there's some other people way better at
22. 3:36that than him. Um but Meta is is a major force right in the technology world. Let's see. Aloha from from Kawaii just
23. 3:46put something in community chat. Oh, never mind. I can see that I can't share an image here. All right. I'll go check
24. 3:52out community chat. Um, welcome Linda. Thank you. Welcome everybody. Um, but Meta today put out a new um model
25. 4:06um that you can run on a small a small computer. Um, it's apparently good at agentic work. So, the the theme of these agentic
26. 4:17um they call them harnesses. I want to come up with a better word for that. These pieces of software that run agents
27. 4:24where you can you can run local models, you can run cloud-based models, you can, you know, just tell it to go do stuff
28. 4:31and it'll go do stuff on your behalf. Those things are out there. They can use tools and um and Meta um you know
29. 4:40launched a new open source open weight. They're well it's open source and they're launching the they're releasing
30. 4:47the open weights in the next day or two. Um, basically what that means, the open weights as I think it was Gareth, Gareth
31. 4:55or digital gods kind of described open weights as like when they were making the model good, they were sort of
32. 5:02turning all these dials and the open weights are basically saying here's how we turned all those dials and you as a
33. 5:09developer can go in and fine-tune your own settings and make it custom to what you want, things like that. So, I don't
34. 5:17know enough about that to really know what's going on, but it's a big deal to release the open weights because you're
35. 5:22basically saying, "Here, have at it. Here's everything we know, and you can go do what you want with it." Um, so
36. 5:28that's pretty powerful. Um, also today, Bernie Sanders said to Dario Amade, Sam, Sam Alman, and Mark Zuckerberg, slow down kids.
37. 5:42Let's let's put a pause on this AI development thing and if you don't do it, the Senate will. [laughter]
38. 5:53So, so, so yeah. So, China's barreling right along building fusion reactors and and open sourcing everything they're
39. 6:05they're doing to flood our markets with with free intelligence. Um [laughter] and uh our government's like, "Let's
40. 6:14what why don't we slow down and think about it?" Um it's not necessarily a bad thing to slow down and think about it,
41. 6:22except that we'll be the only ones doing that. Um the the EU has has kind of taken that route and and it's put them
42. 6:30very much behind in the in the conversation. Um anyway, so so that's that's some stuff going on. Um, here's a
43. 6:38story that I'll talk about that I think is is interesting. So, in Australia, I think this was over the weekend,
44. 6:48um, a guy who had like I think it was Open Claw, some sort of agentic system, said to his agent, "Hey, can you book
45. 6:59me, you know, some time to go play tennis at the tennis club or whatever at the gym?" and his agent went to the gym to book
46. 7:10him time and that class was all filled up. And so, and so what the agent did was it said, "Huh, I wonder if there's
47. 7:19any security vulnerabilities in the booking software." Oops. Yeah, there are. Why don't I just hack the booking
48. 7:26system and I'll go in and I'll find someone else's name and I'll remove it from the list and I'll put my dude's
49. 7:32name in there. So, so, so this agent hacked a system so that his human um could get a slot at the gym or the
50. 7:41tennis courts or wherever it was. It's kind of humorous on the surface. However, imagine now that we all have No way. Oh my god.
51. 7:56I know. Isn't that bonkers? So, imagine we're now all starting to have our own agents, right? And I'm like sending mine
52. 8:04out. Go book this vacation. And you're sending yours out. Go book this vacation. And and mine goes to book a
53. 8:10vacation. And it's not available, but it sees that you just booked one. So I can just cancel yours. So we're going to
54. 8:16have agent wars where the agents are going to be battling one another and battling systems to try to take care of
55. 8:23us, which is it's kind of sweet. Uh but like what the hell? [laughter] So, oh yeah, fantastic Bob. Um, if you all
56. 8:40have any questions, feel free to pop them in there. Okay, this is a note from producer
57. 8:56Brandon. the issues that you're having in chat whereby it sucks and like you can't have the the stream open and the chat open
58. 9:06and it it makes the thing really skinny and all those other problems. We have reported all of those to Streamyard.
59. 9:11Streamyard has gotten back to us and said they're working on it. Okay. So, I think producer Brandon took screenshots
60. 9:18and things like that. So like we we we said the software that you have created that apparently is you know you're charging money for
61. 9:29including us doesn't work. Could you go on ahead saying come on in on Saturday and make your [ __ ] software work. So
62. 9:39So we're aware that the chat experience is is not ideal [laughter] but we appreciate your patience. Um, if
63. 9:48anyone has any questions, pop them in the comments there, uh, and I will address them. And then I think what we're going to do,
64. 9:58I'm going to do this. I'm do this. I'm gonna share my screen. Um, okay. Let's go to X.
65. 10:13So, I've got a bunch of stories that Brandon and I gathered throughout the day. So, let me go back and look at a few of these. August 7th.
66. 10:28All right. This is old.
67. 10:50Ah, this is one I wanted to talk about. So, let's listen to Satia Nadella. >> Yeah, this post is from >> For example, I'll give you uh at
68. 11:02LinkedIn, For example, I'll give you uh at LinkedIn uh we used to have product managers. we had designers uh we had front-end
69. 11:13engineers and then we had sis backend engineers and so on. So what we did is we sort of took those first four roles
70. 11:19and combined in fact increased scope and said let's they're all full stack builders. So I like that because that's
71. 11:27a structural change that allows for us to increase the change both the work and the workflow between these functions
72. 11:35>> and I would assume the velocity because you don't have four people communicating and that throughput of ideas it's just
73. 11:40one person and five coding >> exactly and there's a new workflow so what at the same time as you can imagine
74. 11:45if to build an AI product today there's a complete new workflow right it starts with eval right so basically there's this eval to science
75. 11:55to infrastructure and so eval are done by these full stack builders and what have you and product managers in form
76. 12:02the infrastructure is built by the systems engineers in the back end because they support the science that
77. 12:07supports the product. So in some sense there's a new loop uh and you have to structurally change and so a lot of what
78. 12:14is happening inside of tech is that change uh which is I think going to be pretty massive uh >> okay >> for example [clears throat] so
79. 12:26let me see there we go um so I want to talk about this we talk a lot about let's band together and vibe code.
80. 12:38Wait, where was that? A new Mighty Networks alternative. We could we could certainly do that. [laughter] Um, Mel, I would love to know your
81. 12:47thoughts on a AI forward phones. I'm overdue to get a new one and well, it's an interesting jungle out there. Um, [clears throat]
82. 12:59let me come back to that, Mel. That's a good question, but let me come back to it. I want to talk about this Sati
83. 13:05Nadella thing. So they're taking um product front end backend and what was the fourth one? product management. Hang on. It was
84. 13:20Oh, designers, front end, back end. Designers, front end, back end. Oh, product managers, designers, front end, back end. Okay, those four
85. 13:35roles at LinkedIn, they've rolled them into a single role. Um,
86. 13:50so at the heart of the great repurpose that that you know I've been thinking about for a year, it got coined on the
87. 14:00live about four or five months ago. Um, I saw Silver Fox in the pre-show. I hope she's here tonight.
88. 14:14What that came out of was the realization that not only is AI going to disrupt jobs and like some people are
89. 14:23going to lose their jobs, even if you keep your job, your do job is about to fundamentally change. So, if you're just a designer and you
90. 14:32love design and the only thing you like is design and you're working at a company like LinkedIn that all of a sudden says if
91. 14:40you're a designer, a front-end, backend, or product manager, you're now all full stack design. You're you're all full stack developers.
92. 14:52Wait, wait, what? So, if I'm if I'm a designer with a capital D, I love design. Right. I know actors with a capital A.
93. 15:02You know those people. They deeply identify with that that role. That role's now been collapsed into three other roles
94. 15:11and your responsibilities have been expanded dramatically. Not everyone's going to survive that, right? So even if you keep your job and
95. 15:22your company makes that kind of change, what are you supposed to do with that? What if you don't care about code?
96. 15:34What if you don't care about product management and gathering requirements from the stakeholders? And
97. 15:44I don't know, you're going to have to go find a place that's just looking for a designer. And as things accelerate,
98. 15:52how company A consolidates their roles and con company B consolidates their roles is going to be different.
99. 15:59And some people are going to be like, we want to go all in on AI. And others are going to be like, we're just going to
100. 16:04use AI for efficiency. And others are going to be like, this is about agents, baby. And we're going to we're going to
101. 16:09find every process in our company. We're going to identify that process, and we're going to design agents to
102. 16:15replicate that process. And we need agent builders. And there might not be agent builders working in that company, but maybe
103. 16:23you're a receptionist and you've been nerding out on claude code and you're like, I think I could make an agent. Oh,
104. 16:34come on. Come on over here. Can you make that process? Okay. Hey, Claude Code, could you take this process and do this?
105. 16:43Here's access to our Google Drive. Now all of a sudden you go from receptionist to being head of agentic programming for your company.
106. 16:56So what this says to me is we we've talked about this a lot. Understanding who you are, what you value, what you enjoy, what you're good at
107. 17:12is super important. being curious and playing with AI, understanding what it makes possible, understanding that if I'm a designer, I
108. 17:22could also maybe do some code. That's not something you would naturally go to necessarily, but if you've got that
109. 17:32curiosity and you're willing to kind of circle around what you do well and what you care about, you're going to be more valuable to your
110. 17:40organization as things start to get weird. things are going to start to get weird. Even within a company that lays no one
111. 17:49off, [clears throat] those people that are clinging to the past and clinging to the old way of doing things, at some point
112. 18:03the company's going to say, "Uh, I think we got to do things a little different around here." Anyone thinking different?
113. 18:16And it's not going to be the most talented people that are thinking different. It's not going to be the most valuable
114. 18:24people necessarily that are thinking different. It's probably going to be the weirdos, like weirdos that show up to the AI
115. 18:32learning lab. We're not weird. We're irregular. Okay. But you're not normal, right? It's going to be those people in
116. 18:45organizations where when someone says, "Does anyone know what's happening with this stuff?" Uh, I kind of do.
117. 18:57So, that's what's go And then you ask for a raise. You absolutely [ __ ] ask for a raise. Here's the here's the other
118. 19:02deal, you people in here. It it it's so amazing to me because I still feel and I know a lot of you still
119. 19:12feel like we're so far behind and we really don't know [ __ ] and yet we do. and
120. 19:28your ability to just think critically about if they're trying to solve this problem. Well, I could do this with AI or I could
121. 19:40do that with AI or if I take what we traditionally do and I do this piece of it with AI and then do the other thing
122. 19:46that we do that could probably cut three steps out of that process. Anyone who's thinking like that is going to be
123. 19:52incredibly valuable for a business and anyone who's not at some point is just not going to be as valuable. So like all the the fact that
124. 20:02you're here trying to glean whatever you can from my rants and rambles is I think some version of a key to the
125. 20:13future. Be curious. Be adaptable. Learn. Try to figure this [ __ ] out. Try to figure out what it means. Try to figure
126. 20:20out how you apply it. Try to figure out how you collaborate with other people. You know, one of the things that we've
127. 20:26talked about, we created the the AI ambassadors program. So, we've got people that uh or the AI salon ambassadors program.
128. 20:35We've got people in the salon that really care about it and really care about this idea of empowering people through the use of AI. And
129. 20:49there's a there's an instinct to say, "Hey, we just got to go, rahrh. AI is awesome. Come use AI." except
130. 21:02we're entering conversations where most of the people in those conversations don't know what AI is, are afraid of it,
131. 21:10andor are pissed off at it. And so part of the generous leadership that we talk about in the AI salon, part of what leadership looks like
132. 21:28is being empathetic to the fact that someone may be afraid, that someone may be angry, they might be ignorant, they
133. 21:36might have bad information, they might have the right information, but they're using that information as a way to avoid
134. 21:44having to deal with AI. And so if you can be one of those human beings either within your company or
135. 21:52within the world, within your circles of influence to not just be this polyana kind of AI is the best, but to really say I I
136. 22:04actually deeply understand the issues here. A lot of what you're saying absolutely right. The thing you're saying about water use
137. 22:14in data centers is awesome. I assume that means you're not golfing or eating almonds or eating a burger because those all use way more water
138. 22:24than data centers, but that's neither here nor there. [laughter] I saw I saw a stat someone did the math on a quarter pound beef patty
139. 22:37um uses as much water as 200,000 chat GPT prompts. [laughter] So the the number the the facts are kind
140. 22:49of ridiculous, right? You know, a lot of this is sort of, you know, trumped up China amplified fear-mongering,
141. 22:59but some of the issues are real. Yes, it's going to take jobs. It's also going to create jobs, right? So,
142. 23:12so that's coming. And you know, I talk about the seven economies, this sort of uneven behavior of companies that are adopting
143. 23:22AI at different speeds. You might be working in a company where they're not doing any of this kind of consolidation
144. 23:29that that Sati Nadala talked about right there. You might be working in a company that's just like, we're just steaming
145. 23:35ahead. We're in economy, too. Maybe we'll use a little bit of AI for efficiency, but we do what we do. We're just steaming ahead.
146. 23:45And you might be safe within that company for a while. But there's going to be a point at which one of these
147. 23:50other companies that's doing innovative adoption of AI or aggressive adoption of AI is all of a sudden going to start to
148. 24:01create products or services in your market that start to lose that company business. And so if you're in the unfortunate
149. 24:16position, I think of being in a company that's slow rolling AI and you just keep your vertical job, your your your specialist
150. 24:26job with no curiosity about going beyond that and at some point that company gets competed against. You're going to end up
151. 24:36in a job market that's now only valuing people that understand what AI is and how to use agents and how to create
152. 24:42digital brains and whatever is the latest crap of the day. Like what was it? It was a year ago it was prompt
153. 24:51engineering and then it was it was uh talking to agents and now it's about graphs which are basically multiple agents
154. 25:01functioning at the same time. It's just a way to burn more tokens I think. So all the tactics of what we have to
155. 25:09learn that's going to keep changing but the mindset that's going to remain the same. All right, let's go back to yield X.
156. 25:26Wait, Kyle, before we go too much further, we have a unanswered question from Mel in the queue. >> Oh, cool. Oh yeah,
157. 25:39I [clears throat] would love to know your thoughts on AI forward phones. I'm overdue to get a new one and well, it's
158. 25:45an interesting jungle out there. Um, Pixel 11 and Pixel Watch 5 came out on Wednesday. Um, Brandon is talking about being committed
159. 26:00to to being a green bubble for the rest of his life. He loves being a green bubble. If you want to be a blue bubble,
160. 26:09then you'll listen to me. [laughter] Listen, being a green bubble is a secondass citizen in the world of texting. [laughter]
161. 26:20Okay? And by the way, if you don't know what I'm talking about, here's a here's a a chat with me and Brandon. See the green bubbles?
162. 26:32>> [laughter] >> Here's a chat with my friend Andrew. See the blue bubbles? Fellow iPhone guy. Bluebubble. That's
163. 26:43me. I'm looking at the new iPhone and and Mac laptop. So, here's here's generally nonrcs chat says what. [laughter]
164. 26:55Me, too. I need a new phone. Okay. Okay. Here's the deal. Um the small the small language model the S slms
165. 27:11are are kind of here, right? If you if you do get an Android phone, the the Gemma whatever 4 billion parameter model,
166. 27:21there's there's some there's some small model that's going to end up be be usable on your phone. Um, yeah, this is really interesting.
167. 27:35There's a couple like, okay, I'm from the future where only green bubbles are awesome. Blue Boo, Turkey Hill Press. Oh, Cindy [ __ ] The the
168. 27:48depth of disappointment from a futurist being being green bubble positive. It's Listen, this is we we are we are we will absolutely
169. 28:01shame the green bubbles on this channel. I'm I'm having tic tacs. [laughter] You green bubble people with it's just
170. 28:10as good as an iPhone. No, it's not. An iPhone suck. [laughter] Okay. Hardware decisions. >> [clears throat]
171. 28:30>> I'm thinking of my phone right now, not as an inference machine. And I suppose probably within 3 years,
172. 28:43we can certainly think of our phone as an inference machine, right? It'll have a big enough processor and enough RAM
173. 28:52and and the models will be distilled enough from you know the 15 trillion parameter models will be distilled
174. 28:58enough that we'll have really powerful local large language models on our phone. So I almost feel like you don't
175. 29:06need to worry about phones right now. What color were your Tic Tacs to though? They were green. Fair point. Fair point.
176. 29:17It's a lot of copium from over there in the Android corner. A little iPhone. I think I only have an iPhone 14. I'm a loser.
177. 29:29Um, but so right now the real power of this thing is when I flip it into I think it's called dispatch mode for
178. 29:44claw code and it's called remote for codeex. I can I can put my phone in remote and talk to it. Put it in voice mode. So, I
179. 29:54go remote and then it's going to bring up all of my um my chats and then I can put it into voice mode.
180. 30:12Hey Quinn, how's it going? >> I'm here. What's up? >> Um and I can be managing my codeex or my claude code that's on my
181. 30:27my laptop, right? Or my desktop that's that's running it. So, I don't think you need to go do go too crazy with with the stuff on the
182. 30:38phone. Now, I think that um Samsung and Google are probably ahead when it comes to that stuff. I haven't used it, so I
183. 30:49don't know if it's any good. I wouldn't I wouldn't put as much money in the in the in the mobile device for the for the
184. 30:58high-end [ __ ] as I would in the in the computer side. On the computer side, you basically have two choices. You can say, "Well, screw it.
185. 31:12I just want to buy cheap, so I'm going to buy like a Mac Mini. They used to be $6.99. Now I think they're $7.99 or
186. 31:18$8.99. like everything's gone up by like 25%. Um, you can just buy like a 16 gigabyte M1 or like an M4
187. 31:32um, Mac Mini or you can buy like a MacBook Pro with an M4 or M5 processor with 16 gigabytes. Um, and what you're basically choosing to do
188. 31:45there is you're you're basically saying, I'm going to choose to do most of my inference in the cloud.
189. 31:53So, I'm going to do most of my inference on cloud.ai or chat GPT or Gemini, wherever it is. I'm going to do it out
190. 32:00there. And I'm just going to use my computer for like the the orchestration and the file storage. The other way to go is say, "Hey,
191. 32:15um, we are entering an era where the demand for AI is going to dramatically outstrip the supply and producing tokens is going
192. 32:25to get really expensive. And if you were to buy my the the MacBook that I bought in I think it was May,
193. 32:34April or May was like 4,100 bucks. That same machine right now is like 6,500 for the exact same machine. 25% increase.
194. 32:49That trend is going to keep marching because the demand for inference is for tokens is going to be higher and higher. So,
195. 32:58the way I've always thought about buying a Mac is you buy the the most blasted out version you can reasonably afford,
196. 33:10right? So, I got a 64 gig machine and I was like just this close of spending like the extra 700 bucks to get the 128
197. 33:20and I didn't do it and I'm pissed off. Hey Lori, welcome Silver Fox. Welcome back. I'm glad you're here. Um
198. 33:35I I have a I have an interesting thought on that, Charles. I'll I'll talk about in a second. Um if you want to go the route of
199. 33:44running local models which again I think it is smart on many many levels to understand these agentic harnesses to
200. 33:52understand local inference to understand local models. This is from a guy I have ignored open source for my whole
201. 34:01life. I don't care about it. Open source just feels like it's for the tinkerers and it's always substandard for the commercial [ __ ]
202. 34:13The local openw weight models that you can run on a [ __ ] laptop are insane. Like what the scientists have done over the past
203. 34:29since 2017 since Google invented the transformer. What these scientists and engineers have done is they've effectively taken all of
204. 34:40human knowledge and squished it into a file you can put on your laptop. And so what it means is that you can take that laptop
205. 34:55that's got one of these models on it and you can be completely disconnected from the internet and still have it go do all
206. 35:01sorts of work for you and still have it answer all the questions that you ask chat GBT and help you write and help you
207. 35:07organize things and help you like all that stuff and it's going to get better and better and better and better. So I would
208. 35:16if you can afford it like 32 gigs is kind of the minimum. I went with 64. 128 seems to be the size of of unified memory.
209. 35:30So this is on the reason people are buying Macs over PCs is on a Mac the amount of RAM that you have is available available to both the
210. 35:42CPU and the GPU. So it means you can allocate a lot of that RAM to your AI inference. If you're on a PC you've got the CPU and then you
211. 35:53have VRAM on the GPU which is they're separate buses or whatever. They're separate lanes. And so you can only do
212. 36:01your inference on the GPU. You can't do it with the CPU. And you can tell me I'm full of [ __ ] but I'm pretty sure that's the case.
213. 36:11Why am I coming in as Turkey Hill? Probably because I don't know. Maybe because that's your Vimeo. I I don't know. Turkey Hill.
214. 36:23She's Big Bad Turkey. Turkey Hill Press. if you don't know it. Um, so I would buy a badass machine. And if you're a business owner
215. 36:37and you can afford even more, like you could get a Mac Studio, I don't even think you can buy them right now. I
216. 36:45think they're I think they're not available. A Mac Studio with 512 gigs. Um, I would do that. You could also you
217. 36:53can also wait, right? the the new line of Macs is going to have they're going to be designed for AI. Um but anyway,
218. 37:02let's see. Yay for Apple. Um need to go offline. We'll catch the replay. We'll bring feedback about chat GBT work to
219. 37:08office hours on Friday. Cool, Linda. Um take off and thanks for thanks for stopping in. Um, okay. What about these OOTV single-use
220. 37:20devices, Tiny and Auto, which are basically pre preloaded rebranded Mac minis? They're purportedly shipping for the holiday season, private
221. 37:31and secure. Yeah, they're they're going to be fine. Um, I don't know how much they are. What? Okay. So, so you want my real take on them,
222. 37:45Brandon? They're trying to turn a non-consumer experience into a consumer experience and it's going to be
223. 37:55a [ __ ] disaster because if you don't actually understand how OpenC claw works and how you allocate models and how you download
224. 38:06models and things like that, when the thing craps the bed and just all of a sudden your agents stop doing
225. 38:13what you told them to do, people are going to be like, "Hello, tech support. It's they're going to be an absolute
226. 38:19disaster. AI is not ready for consumers yet and and certainly agent harnesses are not ready for consumers yet. So what
227. 38:28those companies like auto are doing is trying to consumeriize a product that is inherently for tinkerers. It's just not good. Um I
228. 38:40don't think anyway Charles had a question up here I wanted to get back to. Can't imagine being pro blue bubble in
229. 38:47the AI age when Apple is dragging its feet like it has. Okay, I've for years on this channel have yelled at Apple for dragging their feet
230. 38:57with AI. I have actually come around. I still think they missed a major opportunity. I think Apple should
231. 39:04absolutely be one of the frontier labs. I can't believe that they're not. But to their credit, they understand who their customer is.
232. 39:16And what's what what Apple's going to do is this. They they're I think in September Tim Cook steps down and the
233. 39:24new CEO used to run the hardware division of Apple. The M7 chips that they're developing right now are going
234. 39:31to have a maximum unified memory of 1.5 terabytes. So, you're going to be able to buy a Macintosh that can run full-on
235. 39:47major models, like frontier level models running on your local Mac. Now that that 1.5 TBTE MAC is probably going to cost 50 grand, but
236. 40:01it [laughter] can it could, you know, basically serve out tokens to, you know, a large part of a company, if not the whole company.
237. 40:11Um, and and what's going to happen is so Apple's going to hit the hardware first and then at some point when the models
238. 40:22are good enough and their hardware is good enough, they'll just turn it on and it'll just be built in to all of their
239. 40:30apps. It'll be built in at the hardware level. It'll be accelerated at the hardware level. So, I think Apple's play
240. 40:37is we're not going for the early adopters. They're not a tinkerer kind of company. They never have been. What
241. 40:44they're going to do is they're going to wait until the technology catches up and they're going to they're going to launch
242. 40:50what auto is trying to launch right now with with their refurbed Mac minis. Apple's just going to deliver, but it's
243. 40:56probably three years out. So, so, so yes, I agree. And I shouldn't be defending Apple because their quality lately sucks. It's horrible.
244. 41:09Horrible, horrible, horrible. Do they lay away? Okay, any other any other questions, thoughts?
245. 41:26Okay, now got some exciting news that I want to talk to you about. a week from tomorrow, August 18th, Tuesday night. So, the AI
246. 41:43learning lab goes Monday, Wednesday, Friday, we have the AI salon presents on the first Tuesday of the month and then on
247. 41:52the third Tuesday of the months month, we have back to basics. Last month, I did kind of an overall an AI overview.
248. 42:01What we're going to start doing moving forward is these back to basics courses. I'm gonna I'm going to pick a specific
249. 42:06topic, a specific use case, and then we're going to go deep into that. So, one month it might be image generation,
250. 42:13one month it might be um video, one month it might be vibe coding, things like that. So, um, we're going to pick a
251. 42:22topic and then that way when we string these sessions together, we can we can have this overview session that we
252. 42:29started last month and then we can kind of have deep dives on specific use cases. We'll start at the very beginning
253. 42:35in those deep dives, but we'll spend the whole 90 minutes going deep on a particular topic. We're going to need to check your blood
254. 42:44pressure soon. I the it's just the amount of stupidity in the world right now around this AI stuff is it's just maddening because
255. 42:59we're at this remarkable moment in history. It's we're we're at a remarkable moment in history. We are lucky enough lucky enough to be alive
256. 43:11when all this [ __ ] is happening. diseases are going to be cured. Science is going to be advanced. Like just all
257. 43:18this stuff is going to happen. We get to be here when this happens. And we're like, I don't like it. Change is hard. Let's stop it.
258. 43:28[laughter] I am I'm gonna need I'm gonna need a I'm gonna need a dookie. [laughter] All right.
259. 43:44Yeah. Did they pass laws to outlaw the internet? Well, they certainly tried, right? If you're downloading music from
260. 43:52the internet, you're a thief. That was the Grammys in like 1997. The head of the recording academy got up on
261. 44:00stage and he pointed to the two young people in the audience. They're stealing. They're on Napster. They're
262. 44:08they're stealing all your music. It was humiliating for him. Um, oh yeah, [clears throat] this was an interesting one.
263. 44:19I wrote about a China, a China simulation of a billion people. They simulated a billion people. Think of like NPCs in a video game, but
264. 44:31the people are all doing things. They're now simulating humans at scale. Crazy. >> That's 8 billion people. Wow.
265. 44:38>> Oh, this one was 8 billion. Harvard and MIT researchers simulate the entire planet with 8.3 billion AI
266. 44:47personas. It's big. In a de development that feels straight out of science fiction, Welcome to the new world. The
267. 44:55collaborative team led by researchers at Harvard and MIT unveiled Matrix. Like Matrix but with an AI in it. Get
268. 45:06it? Like the great renaissance with the AI like I do, huh? Clever. A population scale AI simulation infrastructure
269. 45:16designed to model the behavior of virtually every person on Earth. It centers on persona 8B, a data set
270. 45:22containing 8.3 billion unique digital profiles. Each persona is defined across, 1290 categorical dimensions that
271. 45:31capture everything from demographic background to psychological traits to spending habits to technical literacy,
272. 45:37behavioral quirks, and lifestyle preferences. It's amazing. It's amazing. It's just crazy. Bonkers. Bonkers. Bonkers. Bonkers. All
273. 45:49right, let's get back in here. All right, that's that one. Okay. Yeah, this was the one that the man asked Open Claw to book a gym class and
274. 46:01it hacked the system. That's a pretty good one. Oh, we both had that one. Okay, this is the the Okay, so so we're
275. 46:09going to read two things. There's a letter from Bernie Sanders, Mr. Altman, Mr. Omade, and Mr. Zuckerberg.
276. 46:21Stop building machines that humans cannot be con humans cannot control. Let me be very clear. If you do not take
277. 46:28action now, my colleagues and I in the US Senate will no Bernie impersonation. Dear Mr. Altman, Mr. Amade, and Mr. Zuckerberg, [laughter]
278. 46:44almost every day there's a new story about how your companies are losing control of the AI. You know who's planting those stories?
279. 46:52Open AI and Anthropic Bernie [sighs] with the potential cataclysmic results.
280. 47:08[laughter] This week we learned frighteningly that AI has been used for the first time to create new viruses. Okay. Okay. Can I just Can we
281. 47:22see now I'm getting pissed off? This week we learned frighteningly that AI has been used for the first time ever
282. 47:30to create new viruses. As you know, this type of development in the wrong hands could lead to new bioweapons. Yeah.
283. 47:36Yeah, it could. And every single other technology in the wrong hands can be used for bad birdie. Do you know what
284. 47:43they produced? Do you know what those viruses are? They're bacteria phages which don't affect humans. You know what they affect? Bacteria.
285. 47:53They go kill them. Anyway, just the drama here. And it was a it was a US model that made the vir was it a US model? I honestly
286. 48:05don't know. Last month, the world found out that Open AAI lost control of an AI model. The result the the model hacked
287. 48:11into the other computers computers. that was res that was reported by OpenAI um hacked into the other another company's computers. A clear
288. 48:23violation of federal law, not a clear violation of federal law. A human being hacking into a computer is a is a clear
289. 48:35violation of a federal law. But an agent that does it I is is there established case law that that is a violation of
290. 48:44federal law? I don't think so. After conducting internal reviews, they reported their model similarly escaped their control.
291. 48:53Uh, one of the targeted companies called the AI hack an unprecedented event that deserves unprecedented response.
292. 49:00Yasha Benjio, the most cited living scientist in the world, said these in incidents should serve as a wake-up
293. 49:06call. I agree. So, do the top companies at the companies you lead, the very people building this technology. Oh, so
294. 49:12do the so do the top companies. Yeah. So, a bunch of people at OpenAI and Anthropic, a bunch of the workers wrote
295. 49:19a letter that said, um, we don't think we should be developing the things we're being paid to develop. So, yes, they
296. 49:30warned there's a real risk of cap uh capability development rapidly accelerating beyond our control. And
297. 49:36yet, at the moment, we have seen human loss of control and cred. Okay, that's absurd. Uh, the impossible. Okay, let's
298. 49:44see. Let's see what he actually says here. Hang on. Um, here's the second one. Okay, the moment is here. AI
299. 49:52capabilities have reached a critical threshold. There is a reason why the head of the CIA said that AI models are
300. 49:59akin to dangerous nuclear weapons, almost like a doomsday device. Mr. Alman, Mr. Amade, Mr. Zuckerberg, in
301. 50:07the interest of humanity, stand by your words. Pause AI development. It's not too late to avoid disaster. Stop
302. 50:15building machines that humans cannot control. Let me be very clear. If you do not take appropriate action now, my
303. 50:20colleagues and I in the US Senate will this is so so this is the the
304. 50:35intended consequence of Dario Amade and Sam Alman fearmmongering their own models so that they can get the US government to stop
305. 50:47China from competing with them. That's what this is all about. And what's going to end up happening is they're going to they're
306. 51:00Bernie Sanders, so he's he's probably not going to be taken that seriously, but 80 83% of people as of six months ago
307. 51:15have a negative view of AI. So there's massive amount of fear out there. So what's going to happen is they're fear mong fear-mongering.
308. 51:25The American people don't want it. The politicians are just going to go wherever the power is. So if there's
309. 51:33power to slow it down, they're going to push all these companies to slow down. So we're going to artificially slow down
310. 51:38America. And what's going to happen is all the other countries are not going to slow down. and our ability to battle the
311. 51:46risks of our own models. What happened in the in the hugging face attack this? Okay, see now I'm pissed off. Hang on. I
312. 51:57got I gota I got to see my own face here. What happened in the hugging face attack was the open AI model
313. 52:05attacked Hugging Face. Hugging face saw that they were being hacked. Hugging face went to Anthropic's fable model.
314. 52:16You know, the really dangerous one that we couldn't have mythos, but we had to have Fable because it was so dangerous.
315. 52:22We put all these guard rails on it to make it safe. Hugging Face tried to use Fable to stop the attack of Open AI and they couldn't
316. 52:32because the safety guard rails on Fable wouldn't let them do it. So, our safety guard rails made it significantly more dangerous for Hugging
317. 52:47Face to protect their own company. So, what did Hugging Face do? They went to Chinese models because they could use them
318. 53:00and they stopped the attack. The only thing we're going to do is if we slow down the American models is put ourselves at risk
319. 53:12from all the other models coming from the world. We need our models to be as good or better than theirs so we can protect ourselves.
320. 53:21This is absolute insanity and it's a really juicy narrative. And if open AI and anthropic make it scary enough, well, our government just
321. 53:33might shut down every other model, which if you lived through the 80s, you know how it went when they tried to outlaw encryption, right?
322. 53:43You can't outlaw software. I mean, you can. They probably will. [sighs] We have officially launched into Meltdown Monday. [laughter]
323. 54:00back in my day was escaping. We didn't know it. Oh my god. No hating on bacteria. There are plenty of really good and necessary bacteria.
324. 54:10Yeah, exact. Well, but but the the bacteria phasages that they've designed are designed to to attack the bad ones that make us sick.
325. 54:20Vickiy's in the house. All right, let's see what else we got here. They'll probably look back at these AI
326. 54:31development years, be like, they were crazy back then. Yeah, I don't know. Listen, I might just be a polyianish [ __ ]
327. 54:42right? Maybe these things are that dangerous. But what pisses me off is
328. 54:58of all people, Bernie [ __ ] Sanders, right? Let's stop giving money to the rich. Let's give it to all these other people need the money.
329. 55:12You have a technology here that if it does what it's purported to do, it will vault us into an era of hyper prosperity, right?
330. 55:29And he's trying to stop it. It's it's like
331. 55:43It makes me depressed. Makes me [ __ ] sad. No wonder they call them frontier models. Yeah, they're frontier models. It's the frontier. It's
332. 55:53the wild [ __ ] west. We have the most powerful technology in the history of humanity. We we are actively curing diseases
333. 56:07with technology that is that is the alphafold [ __ ] right now today is actively curing diseases actively.
334. 56:24Science and energy and mathematics problems are being solved actively
335. 56:38and it's just like it's like what the [ __ ] You know, it's it's funny. I remember I remember being in college
336. 56:54and it was the first time I'd ever kind of heard someone had an idea about like imagine having a news channel that only has good news on it.
337. 57:03And then just multiple times, probably once a decade throughout my life since college, there's been some attempt at that.
338. 57:20And the reason that good news channels don't succeed is that no one gives a [ __ ] [laughter] Nobody cares. There was that one
339. 57:36during CO that John Krinski did and it [laughter] it was like popular for like three months then it went away and it's like
340. 57:52it's like we're sitting here with this thing in our hands that is that's got the power to literally transform in massively positive ways
341. 58:06and it's happening globally. And so, yeah, bad people are going to get their hands on it and good people
342. 58:15are going to get their hands on it. And if we as the United States artificially put our hands around our own throat and squeeze
343. 58:29what's happening, man. [laughter] It doesn't make sense that Bernie and a AOC are anti-A's going to solve all the
344. 58:42problems they want to fix. I agree. I agree. But right now it's politically advantageous for them to go the big boogeyman.
345. 58:52Politics is about scaring us and dividing us. The big boogeyman are those shitty scientists. Anyway, all right.
346. 59:09Let's go read something more positive. Idiot. You'd better slow down or the Senate will make sure that you do. Really, Bernie?
347. 59:26Really? I'm I'm rage eating Tic Tacs right now. I'm probably going to choke out.
348. 59:41The government's going to make sure that your private companies can't do what you're set out to do, but your charter is.
349. 1:00:00What does baby Yoda say? [ __ ] off, Bernie. [laughter] Okay. rage eating tic tacs.
350. 1:00:15I am. It's hard to talk with Tic Tac dust in your mouth. [laughter] All right.
351. 1:00:37>> [laughter] >> Okay, this one. So, [clears throat] today Meta put out what's their what's that model called? The model's called
352. 1:00:51It's got kind of a cool name. Muse Glimmer. Introducing Muse Glimmer. An open weight. Oh, wait. Hang on. I want to do
353. 1:01:00something here. Hold. Hold. Please. Put this over here so I can see y'all. >> [clears throat and cough] >> Okay.
354. 1:01:25Okay. All right. Introducing Muse Glimmer, an openweight 30 billion parameter model optimized for local always on agent
355. 1:01:40workflows. So again, back to the question about what Mac you should buy. If you're going to buy a Mac for AI
356. 1:01:49inference, what you're effectively doing is you're dedicating that Mac to run your agent harness. Um, it's not going to be your daily
357. 1:01:59driver computer. It's going to be a computer that literally sits on your desk and then from your phone and from
358. 1:02:05other devices and from your old laptop you can you can access it, right? [clears throat] Um so this model is designed to live on
359. 1:02:14those machines and do that stuff. Muse Glimmer delivers strong performance on key agentic use cases and benchmarks
360. 1:02:21compared with leading models in its size category and designed to run entirely on consumer hardware like Mac or PCs with
361. 1:02:28performance GPUs with performant GPUs. In keeping with our long tradition of sharing fundamental AI research, we're
362. 1:02:36releasing the model weights under a permissive Apache 2.0 no license. Um, for a local agent to be practical,
363. 1:02:44generation latency must be low enough to maintain workflow continuity. To run Muse Glimmer on consumer hardware
364. 1:02:52without degrading quality, we use quantization to to strike the language model to under 20 gigabytes and a lightweight a lightweight D flash
365. 1:03:03drafter model to accelerate token generation. As a result, Muse Glimmer is fast enough for fluid conversation and
366. 1:03:11real-time agent interaction, all running entirely on your device. So, so it's running on a 5090, an M5 Max
367. 1:03:19with how much RAM are they saying here? An M4 Max, an M5 Max. Let's see. Doesn't say how much RAM they have. I assume 32 gigabytes.
368. 1:03:34I don't know. Muse Glimmer can complete multi-step agentic tasks end to end from a single natural language prompt.
369. 1:03:46Download it here. So, they're kind of back. And then Yield Zuckerberg
370. 1:04:00wrote this. And I I I'll read this until I get bored. [laughter] Um, the future is for everyone. The path
371. 1:04:08to a positive AI future. So, this is the same day that Bernie's saying, "Slow down, everybody." Zuckerberg's like, "Hey, here's an idea.
372. 1:04:19We are fortunate to live in an incredible moment in history. In the next few years, people will be able to
373. 1:04:23use super intelligence beyond human capacity to create and discover extraordinary new things, build new
374. 1:04:29businesses, express new ideas, learn new concepts, and advance our health and quality of life. As we get closer to
375. 1:04:36this moment, it is important to develop a philosophy for how we can best use super intelligence to ensure it improves
376. 1:04:43all of our lives. The defining question of our age are, who will have access to super intelligence? If Bernie has his
377. 1:04:50way, none of us. [clears throat] And what will we direct it towards? Will it be centralized and restricted to a few institutions? So, this
378. 1:05:00is Zuckerberg saying, um, we lost the lead to OpenAI and Anthropic. And so, I'm going to go on record as saying, if
379. 1:05:09they win, we all lose. Right? [laughter] So, that's that's him there. There's his little admission of corporate [clears throat]
380. 1:05:19shittiness. Um, or will it be a tool that empowers everyone? And this is this is where they're headed. I don't quite
381. 1:05:26know how they make money on this model, but we'll see. We propose a philosophy based on individual empowerment.
382. 1:05:34Um, and and again, this is actually quite consistent with what we've been talking about for the past six months or
383. 1:05:40so in the AI salon, right? understand data sovereignty, understand data provenence, um understand local
384. 1:05:48models, invest in hardware, you know, powerful enough to run local things. We propose a philosophy based on individual
385. 1:05:56empowerment as the source of prosperity. Invention as the primary purpose of wait oh individual empowerment is the source
386. 1:06:05of prosperity. Invention as the primary purpose of super intelligence and balance of power as the foundations of
387. 1:06:13of safety. So this balance of power right here that's what I was just talking about. If we choke off our frontier models from being in the
388. 1:06:22frontier, we give up our balance of power. It's astounding to me. It's like it the level of short-sightedness is just it's
389. 1:06:31making my eyes cross. Okay. All new technologies create opportunities and challenges. Super intelligence will be
390. 1:06:38among the most important technologies in history. So, its opportunities and challenges will likely be greater than
391. 1:06:44any we've seen in our lifetimes. We should take this very seriously. Still, it's surprising that the discourse from
392. 1:06:51many developing AI is so filled with doom. This is, "Hey, Dario Amade, shut the [ __ ] up." That's what that is. Oh, and by the way,
393. 1:07:01Sam Alman, could you shut the [ __ ] up, too, please? Thank you. I do not understand why anyone who believes that
394. 1:07:07AI will eliminate most jobs and much of human humanity's relevance would rush to build the future. Right? This is just a
395. 1:07:14a big old smackdown of those two companies. The notion that AI is so dangerous that the only safe path path
396. 1:07:21is an extreme concentration of power, which to be clear, that's exactly what they're asking for. Open AI and
397. 1:07:28anthropic, that's exactly what they're asking for. The notion that it's so dangerous that the only safe path is an
398. 1:07:34extreme concentration of power seems inherently problematic. I can't stand Mark Zuckerberg, but I actually agree
399. 1:07:41with him fairly deeply on this point. historically hoping that an absolute power will benevolently prov. It's also
400. 1:07:48ironic coming from him, right? But [laughter] whatever. Um I think I just launched an app. Did I just launch an app? Maybe not. Okay, good.
401. 1:07:58Historically, hoping that an absolute power will benevolent benevolently provide for humanity is sufficiently
402. 1:08:06enlightened if sufficiently enlightened has not led to a safer positive outcome. Humanity has witnessed many
403. 1:08:12transformative advances. Each time is the fear that people will be left behind. But each time humanity has come
404. 1:08:21out with more people sharing greater prosperity, health and freedom. That's true. We believe that this will be true with
405. 1:08:29AI as well. And the abundance of the future can be shared by everyone. We also believe that the values that got us
406. 1:08:36to this point like liberty, open inquiry, free enterprise, and equal opportunity are also the right values uh to to build a positive future.
407. 1:08:46Putting power in people's hands to pursue their own aspirations is how humanity got made has has made the most progress.
408. 1:08:54So, he's going all in on I'm going to provide tools for individuals to basically take back their own agency.
409. 1:09:00Novel ideas and major steps forward rarely originate from established institutions alone. They came from the brothers in a bicycle
410. 1:09:08shop who believed they could fly, the bookbinders apprentice with no schooling who figured out how to generate
411. 1:09:14electricity. And the kid in a garage who thought the personal computers could be for everyone, Steve Jobs. We believe
412. 1:09:21that this will continue to be true. As everyone gains more powerful tools, each person will become more capable of
413. 1:09:27shaping the future, not less. invention not automation will be the greatest contribution of super intelligence. This
414. 1:09:34is huge. So this is when I talk about the seven economies the automation economy is economy three. It's a dying economy. If all you're
415. 1:09:45doing is taking what you're currently doing and automating it, you're automating dead processes. Economies four and especially five, six
416. 1:09:55and seven are about invention, reinvention, completely imagining new ways of doing things, right? So I agree with him here
417. 1:10:03as well. I can't believe I agree with Zuckerberg this much. Invention, not automation, will be the greatest
418. 1:10:09contribution of super intelligence. Early AI could answer questions and do routine work. Right? Soon, it will increasingly help discover new
419. 1:10:19knowledge, ranging from discovering new jobs to cure a family member's diseases. I got a houseful of people ready for
420. 1:10:24this. To finding new ways to improve your business. While the number of questions a person can ask in a day is
421. 1:10:31limited, the number of valuable things super intelligence can invent to help achieve your goals is unlimited. Right?
422. 1:10:38As intelligence becomes abundant, the most important question will be how we direct it. Totally agree. Some argue that super intelligence
423. 1:10:48itself or a small set of experts who control it should decide what's best for humanity. We disagree, right? I I also
424. 1:10:56disagree. What's going on over here? Okay. Zuza Zuckerberg isn't one for slowing down, especially when he's on board his own boat.
425. 1:11:05Yes to invention. Um, we believe that delivering super intelligence to everyone is the way to answer the question.
426. 1:11:14This follows the tradition of putting the power of supercomputers and the internet in everyone's pocket and on our
427. 1:11:21desks. Rather than centralizing super intelligence, we should distribute it widely. This is what Apple's going after, right?
428. 1:11:30And give every person the ability to direct it. This has the potential to begin a new era of personal empowerment
429. 1:11:37where individuals can use this powerful, capable new capability to reach their full potential, pursue their interests
430. 1:11:44and improve their lives uh and the and the world more than ever before. So yeah, yes, that now this is actually radical and dangerous
431. 1:11:58if you're one of the people sitting in a in the center of power right now. You don't want everyone in the country or
432. 1:12:07the world to be able to do that because one of those people might come up with something that makes your thing not as valuable. Okay,
433. 1:12:18to make this more tangible, here are a few specific ways a meta is building personal super intelligence for all of our lives.
434. 1:12:28Everyone will have an exceptionally capable personal agent that understands you, your goals, and everything you care
435. 1:12:35about. This Mel is about to do an LOL where she's going to teach you how to do exactly this, right? You can create a personal agent
436. 1:12:46that you can use with any AI that understands you, your goals, and everything you care about. This is about
437. 1:12:51to be taught. Okay? Your agent will work 27 24/7 on your behalf to improve your relationships, health, career, finances,
438. 1:13:00home management, hobbies, and more. It will free up more time, blah blah blah. Okay. For example, my agent flags
439. 1:13:08interesting information and helps me prototype ideas. It helps me keep healthy by monitoring my sleep and then
440. 1:13:13watching as I train. My daughter loves to bake, so the agent plans personalized recipes for us to make together on the
441. 1:13:20weekend, orders the ingredients, and offer suggestions as we're baking. Okay, whatever. All right. Everyone will have
442. 1:13:25incredible tools for creation to express your ideas. My 8-year-old daughter can already code her ideas and produce
443. 1:13:31videos in an evening that would have taken me either months or been impossible previously. So, yes, creativity goes up.
444. 1:13:40Everyone will have powerful tools to create new businesses. Yes, you can do you can make money with this [ __ ]
445. 1:13:46Everyone will have a personalized tutor and coach with a PhD in every subject that's already essentially there. But,
446. 1:13:52you know, imagine that that now running locally on the hardware that you own and you're able to do that, right? That
447. 1:13:59everyone can have that. That's this is where I think that Apple is when they decide to turn on the the AI uh capability, it'll just be
448. 1:14:11there and it'll be it'll be really good. Um they're just not developing it, which is that's disappointing. 40 acres in a
449. 1:14:18mule. It's it's in the neighborhood, isn't it? Everyone will benefit from scientific advances and be able to contribute to
450. 1:14:25scientific progress. For example, Biohub has already shipped open source biological models related to virtual
451. 1:14:31cells and proteins that are helping scientists make new discoveries and design new drugs. When Priscilla and I
452. 1:14:38started this work, our goal was to help scientists cure or prevent all diseases this century. And now with advances in
453. 1:14:45AI, we expect this will be possible much sooner. Um, someone I heard today was said five years that all diseases will
454. 1:14:54be cured within five years. I think that's probably optimistic, but let's say 20. That's still pretty [ __ ] remarkable.
455. 1:15:02Since there's a long tale of health conditions and everyone's biology is different, people's direct involvement
456. 1:15:06will help unlock personalized therapies. And as this this is that dude from England or was it England or Italy?
457. 1:15:14Might have been Italy. His dog had a tumor. He biopsied the tumor, designed his own mRNA cancer treatment, and made his dog
458. 1:15:25cancer-free. Um, so that's going to happen. There's also going to be disasters. Don't get me wrong. Beyond
459. 1:15:32biology, AI will help people advance research in every field by applying the scientific method and working over weeks
460. 1:15:38or months to test and refine hypotheses. Scientific discovery will likely be one of the most valuable forms of invention
461. 1:15:45in society. Everyone will have free or affordable access to tools. Right? This is the thing where five years from now,
462. 1:15:52every one of our smartphones um w will be as as good as fable or better. Personal AIS are the future. The
463. 1:16:01important question is who will control the data the AI has access to? Exactly, Cam. That's super important. Um, the
464. 1:16:11time to get educated is now because your ability to understand where your data is and your ability to manage it and own it for
465. 1:16:22yourself is critical. Um, all right. Let's see. Then he talks about risks and basically doing white hat, black
466. 1:16:38hat, the good guys and the bad guys will duke it out basically, but everyone needs access. Job growth in the economy. People have
467. 1:16:46an infinite demand for new experiences and have always found new problems to tackle. There's a natural balance in the
468. 1:16:52economy between companies working to serve people's needs more efficiently through automation and individuals
469. 1:16:59gaining new skills to serve more advanced needs. When there's a healthy balance, overall productivity and innovation increase
470. 1:17:06while employment levels remain high. People fear that automation will outpace outpace individuals capability growth
471. 1:17:14leading to new job displacement followed by a difficult period where people have to learn new jobs. But there's no rule that AI must
472. 1:17:22increase automation faster than it increases. This is a really important line right here. There is no rule that AI must increase
473. 1:17:32automation faster that it increases individuals abilities or demand for new skills. So what what Zuckerberg's saying
474. 1:17:40here, so remember when I was talking about economy 3, if all you're doing is automating what we're currently doing
475. 1:17:48and you're just making what we currently do more efficient, that's the argument for AI will take all the jobs,
476. 1:17:56right? AI is going to take all the jobs of the existing companies doing [ __ ] the way we've always done it, more
477. 1:18:04efficiently. It'll take those jobs. But over here on this other side is AI is an inventor and is an amplifier of people's ideas.
478. 1:18:17So the people that get laid off over here because we made their job more efficient have access to the same [ __ ] tools
479. 1:18:26that got them laid off in the first place. This is the the the crux of the great repurpose is this. The entire narrative right now is AI is
480. 1:18:40going to take our jobs. It's just looking at one half of the equation. We all have access to the same tools
481. 1:18:49that are displacing us and we can invent new things and new ways of doing business and new competitors from the
482. 1:18:57shitty ass companies that didn't have the imagination to think that way. We can put them out of business. God damn it,
483. 1:19:10I'm rage eating Tic Tacs again. [laughter] Recent statistics suggest that it may be more likely that individuals capability growth could
484. 1:19:24match or outpace automation, in which case people will gain the ability to do many new things before their current jobs change. This would
485. 1:19:35lead to a healthy balance and potentially even job growth. Yes, let's figure out new ways to do [ __ ] or new things to do. Australian
486. 1:19:45guy with the dog. Okay. I thought it was Italian. Oh well. All right. What else? Building AI infrastructure. Oh, here's the data
487. 1:19:58centers. [laughter] Basically, there's a lot of jobs building the infrastructure for all this,
488. 1:20:13securing AI. In a free society, people will have the tools that can be used for good or harm, but law enforcement and military have
489. 1:20:23more weapons and intelligence gathering. With AI, the defenders will also have more compute to generate significantly
490. 1:20:31more intelligence from the same models. And in some cases, they'll have more advanced models as well. The strategies
491. 1:20:37to protect against cyber security, bioteterrorism, and other threats vary, but the general pattern is that society ensures the
492. 1:20:45defenders who keep bad actors in check must always have the greater balance of power and resources. Right? This is this
493. 1:20:52is the whole argument that Bernie's saying, "Slow it down." And he's basically just putting a chokeold on our
494. 1:20:58ability to defend ourselves, which is crazy. Kyle needs some jelly beans. Bllandon's flavored. I'll take it. [laughter]
495. 1:21:08All right, let's see. Protecting freedom against and preventing government tyranny. Open AAI and Anthropic are asking for
496. 1:21:16ensuring American leadership. Which nations lead the development and deployment of advanced AI will contribute to a new geopolitical balance
497. 1:21:24of power that will determine the prevalence of democratic values, prosperity, and security in the future.
498. 1:21:30To ensure this reaches a positive equil equilibrium, we must ensure that leading AI systems are the ones that encode our
499. 1:21:37values, grow our economy, and advance national security. The best way to achieve this is to distribute our
500. 1:21:43systems widely. Meta will contribute to this by building leading personal super intelligence agents and models,
501. 1:21:51including open- source models, and delivering them to billions of people around the world. They're kind of
502. 1:21:56seeding the the playing in the big game. Um, but I this kind of an interesting approach. Um, I I I don't know. I don't have any real
503. 1:22:07opinions on whether they'll do it or not or whether their models will be any good. alignment with people and existential
504. 1:22:15risk. Maintaining the control of super intelligence. I'll go I'll I'll go read this this whole thing at some point here.
505. 1:22:27The philosophy of personal empowerment, invention and balance of power is sever several policy implications. At Meta, we
506. 1:22:34will focus our efforts on delivering personal super intelligence to billions of people and small businesses. Oh, you know what?
507. 1:22:50All right. I just I just had a a flash of how they're going to make this valuable for themselves. They're going to give out powerful
508. 1:23:06personal models that that each of us on our individual laptops and phones are going to be able to run. What they're also going to do
509. 1:23:17is provide a really easy way for you to host them on their servers. And they're going to say, "We'll give
510. 1:23:25you your own bubble, your hermetically sealed bubble. We're not going to train anyone on that." But I guarantee you in that services
511. 1:23:34agreement, they're going to have access to your data. That's what they're going to do. So the good news
512. 1:23:45is that people that are smart enough can take these models. If their models end up being really, really good and they're
513. 1:23:50super intelligent for the masses, cool. The smart people will take those and put them on their own machines and have
514. 1:23:58their own data sovereignty. And then the other people be like, "I got free AI. I just stick it right there on the meta."
515. 1:24:09That's what they're going to do. Open source is positive and important force. Independence, governance, independent
516. 1:24:17governance and oversight is important. Given the highstakes nature of decisions around super intelligence, I do not
517. 1:24:23think it's my or Meta or the world's best interest for me or anyone else to be the sole decision maker on how super
518. 1:24:30intelligence gets deployed. Therefore, Meta is imple implementing a governance structure that gives our
519. 1:24:36independent board of directors the power to approve the safety criteria for releasing models and reviewing whether each model released
520. 1:24:46adheres to this criteria. While Meta is a founder controlled company, meaning Mark controls it, the CEOs of all
521. 1:24:54Frontier Labs currently have extensive authority over model releases. So, we encourage other to implement structures
522. 1:25:01similar to this as well. I think an indust industrywide version of this process would be helpful. Well, that's
523. 1:25:07the one that um Dario Amade wants to design that process. So, Mark's basically saying we're going to do one
524. 1:25:14and maybe we can collaborate rather than you just being a dick. All right. Government policy is necessary to ensure a positive future.
525. 1:25:22In mitigating the risks discussed above, I have outlined several policy recommendations. Overall, we should
526. 1:25:27encourage close co cooperation between Frontier Labs and the government. Conceptually, I'm okay with that. It is
527. 1:25:34important for national security that government has the technical resources to harden critical systems. I agree with
528. 1:25:40that. And make effective use of advanced models. Correct. But it is also important for national security that the
529. 1:25:49American AI industry continues to lead. Yes, we should accelerate the ability to build infrastructure, whether it's
530. 1:25:56energy, data centers, or silicon, and build trust and alignment with communities across the country. Building
531. 1:26:03trust and alignment with communities across the country is going to mean sharing in the [ __ ] profits.
532. 1:26:12We should be careful not to slow down American competitiveness. We should be skeptical of any proposals that lead to centralizing super
533. 1:26:19intelligence. We should be in favor of policies that promotes promote a healthy balance of power. All right. Um I can't talk anymore. Okay.
534. 1:26:31Good lord. Um, I have never really agreed with or jived with Mark Zuckerberg, but there you have it. I was just about to ask you why you're
535. 1:26:47trusting anything Mark Fecklas Zuckerberg says. [laughter] He's just trying to stay relevant. Uh, what's the next most valuable
536. 1:27:00commodity after your data after data? Well, so, so I think what happens is this. Human ideas and human taste
537. 1:27:16are the most valuable. I I see all this as rahrh propaganda. I I don't see it as that, Chef Kelly. I see it Well, I I absolutely see
538. 1:27:31What Zuckerberg's doing is he's calling out open AI and anthropic for doom and glooming as a strategy for regulatory capture so
539. 1:27:47that our government makes their competitors illegal. Like that's what's happening. So Zuckerberg's calling that
540. 1:27:53out. What Zuckerberg's also doing is he's saying, um, I can't compete with them on the frontier models
541. 1:28:04yet, maybe, if ever. So, what I'm going to do is I'm going to I'm going to step it down to small models and I'm going to
542. 1:28:11get really good at small models. So, you keep playing up there. I'm going to play down here and I'm going to distribute this far and wide
543. 1:28:21and I'll pull in all the individuals to use my technology because they can't run yours because yours is going to get
544. 1:28:27really expensive and you're going to continue to [ __ ] things up with the government. That's I I just see this as him
545. 1:28:33trying to stay relevant while they've taken a significant lead. Um what's what's the next valuable commodity after
546. 1:28:40data? So I think first is human beings and our ideas become the most valuable thing and then sitting right underneath
547. 1:28:47that is those human beings ability to take their data and their ideas and their projects and and have those two
548. 1:28:56things live together. So the human says I want to do this. Their context, their their data and their history
549. 1:29:06is is the thing that makes the AI actually pull off their idea the way only they can. So I think that becomes the primary
550. 1:29:15um piece is kind of like whatever AI you have access to plus your context plus you as the arbiter of taste and the generator of ideas.
551. 1:29:25That little triumvirate is how each of us get powerful in the future. All right, we got a minute left. That was saucy.
552. 1:29:37There was rage tic-tacking. [laughter] It was the fruit flavor. Citrus.
553. 1:29:52What did we learn here tonight? I don't know. Nothing. [laughter] Blue bubbles are better. I think that's
554. 1:30:01I think we learned blue bubbles are better. Those guys are flying too close to the sun. Hey, listen, man. This is
555. 1:30:10we're dealing with companies now that are going to have trillions of dollars of investment that in the current business model they
556. 1:30:23have no chance of repaying of of of returns. Right? The bet of why people are investing trillions of dollars into a handful of
557. 1:30:35companies is that if one of them wins, they kind of win the whole world, right? That's the bet. So, we're in kind of heady territory here.
558. 1:30:50Wasn't Meta buying Manis, but then China blocked it? What was the result? I don't remember. I don't remember. Anyway, it's 9:00. I'm
559. 1:31:01going to get out of here. Hope you had fun tonight. Um, I lost my voice. That's always fun. [laughter] I got a little cranky.
560. 1:31:11Hope you had fun. Talk to you later. Bye. Oh, wait.