4/16/2025 -ChatGPT's New Reasoning Models o3 and o4-mini-high: A Deep Dive

Transcript

1. 0:02You ready? Are you ready? Yes, I'm ready. Are you ready? Yes, I'm ready. I'm falling in love.
2. 0:19[Music] [Applause] [Music]
3. 0:33There's been something baby I've been trying to say. I don't know how past and a future now surrounding
4. 0:50[Music] me. surrender to an earth through can be [Music] found. There's been a little there's been a little trouble since you came to my
5. 1:06[Music] rescue. And if you like all of the rest, I would have quit you long ago, but I couldn't do that. [Music]
6. 1:22Oh, tell me now. Women and wine never went to make a man and make him go as hell. I know one night you wish me
7. 1:39well. But in spite of your trying, still going to have to find my way through.
8. 1:54morning. My dream remains hanging in the burnt fields of my [Music] memory. Time is a chorus of your [Music]
9. 2:11disdain. Left behind me. I'm searching solutions on the double. on his two ton
10. 2:28telephone. My friend says, "You ain't got no real troubles. There's too many choices that you left [Music]
11. 2:39undone." My friend says, "Put an action to those words." Put an action to those words. They're all tell me now
12. 2:55women never went too well. Champy sounds like he's been watching the news. You know what? I have not been watching the
13. 3:02news. So, I am in ignorant bliss. I I just assume it's all burning down. There's lightning on fire. So, here we go.
14. 3:16Right about the time they light it on fire, AGI shows up. Oh man. All right. I made a website using Claude and Chat GPT that users can learn
15. 3:30about Hang on. Christian music composers. Very cool. That's very cool. You know, it's it's really funny. We're gonna
16. 3:44[Music] There's going to be like there's just going to be some amazing things are going to happen and there's going to be a lot of
17. 3:58[ __ ] a lot of crap out there um and a lot of noise. But what's going to happen is people who are very passionate about a very specific
18. 4:09thing because they're going to be able to wield such powerful tools are going to be able to make really incredible
19. 4:17resources, applications, games, music, all that sort of stuff for their little specific niche. And I I think just like
20. 4:27I've talked for two years now that I you know we're about to enter the great renaissance. um kind of Renaissance 2.0. Um and
21. 4:39uh you know that that's a really good example of it, I think. You know, something you're passionate about, you put it out there,
22. 4:48bang, done. Love it. It's very cool. It's very cool, people. I like it. I like it a lot. at 88 day.
23. 5:06Um, who's that? Silverf Fox sharing the live with her friends. Thank you so much, Silver Fox. Um, tonight we're going to be digging
24. 5:16into chat jpai. You can make money with JP. You can [ __ ] solve math with chat GPT. Um, we we're at a we're at
25. 5:28a there's some crazy [ __ ] You know what's funny? I put out um I put out a video on Tik Tok I put out a couple of videos on
26. 5:38Tik Tok today. I was a little productive beaver. Um but the last one I did at the end of the day was um you know get the toe to a
27. 5:49nunnery. Get your ass to chat GPT and start playing. Start playing with these new tools. do it today. It's no longer
28. 5:55optional. I kind of [ __ ] up that part when I was getting in the car. I didn't quite remember what I was
29. 6:02saying, but um some of the comments are, "Wait, so it's 03 and 04, but that's better than 4.5?" It's like, oh [ __ ] Yeah, cuz the
30. 6:18naming conventions are so [ __ ] up that people are completely confused, right? They launched 4 40 4 then then 03 then
31. 6:284.5 then 4.1 to the API but they still announced 4.1 after 4.5 and then today 03 and 04. So if you're not paying
32. 6:40attention if you didn't watch my video explaining the naming conventions and memorize what I told you then you may be stuck in this
33. 6:49situation where you're like well I thought 4.5 was the best. And if you were off playing with 4.5 thinking it was the best and not
34. 6:57really understanding what a reasoning model was, these new reasoning models that came out today are going to shock
35. 7:08you. They're going to shock you at what they can do. [Music]
36. 7:22Hey Pate, I'm curious. I just saw that you came in. I'm curious if you played with 03. I think I understand, but no.
37. 7:30I'm curious if you played with 03 or 04 mini or mini high because you know the maths. I don't know the maths, but David Shapiro seems
38. 7:41to think they have solved math, which is I that feels hyperbolic to me, but 99.5 on one of the math benchmarks
39. 7:51is pretty pretty [ __ ] pretty [ __ ] close to 100. [Music]
40. 8:36Oh, you don't really do anything with chat GPT. Yeah. Oh, and then and then uh uh Google today said they're going to
41. 8:44they've got a new something coming out that's going to So, so what happened today? So, Gemini 2.5 Pro was at the top
42. 8:51of the the comparison leaderboard, whatever that leaderboard thing's called. Um and then 03 and 04 Mini just
43. 8:59topped that leaderboard. And then and then someone at Google said, "Yeah, we're coming out with a new one next
44. 9:04week. It's going to be better. So, the the com the the wars are hit heating up people. Math benchmarks are tricky
45. 9:14because you it's usually in the training data for a while. Yeah. Yeah, I know. But still going from like 70s, you
46. 9:23know, high7s to 99.5 on that one [Music] benchmark. But I understand your point. the the thing the thing that that's
47. 9:32really different. I don't know if you saw the announcement, you know, the stream today, Pate, but um probably not.
48. 9:40Um they were talking about one of the one of the things they had it solve it did 600 tool calls. So in the new model
49. 9:51it it will spin up um little Python uh windows and write code, execute it and then use that in part of the reasoning.
50. 10:00And they said it for one one particularly hairy problem. It it you know it invoked a tool more than 600 times which is pretty
51. 10:11staggering and then it got it right. So So we're going to go play with that today. And it's also it's they're also
52. 10:19multimodal. So they do the image generation stuff. I was playing around with some of that. They understand images. They will do things like
53. 10:30um split images up to analyze parts of them. Um it's pretty crazy. So we we're just going to go play. I I haven't played
54. 10:43enough to know what's really in there. So we're going to go do that. There's probably other things that we
55. 10:48should be looking at, but this this feels like a pretty significant situation. Situation [Music] situation. I know what I'm going to do.
56. 11:04Hang on a sec. [Music] And my the the other kind of indicator that this is a big deal is my my Twitter timeline like every single
57. 11:21post of anyone that I you know know or respect with you know points of view on AI like every single one is saying this is a big
58. 11:32deal. It's it's like without exception. Normally, normally it's like, you know, I think I saw one that was
59. 11:39like04 mini sucks because it's not good at creative writing. So, that very well may be the case. But, um, what am I looking for?
60. 11:56[Music] [Music]
61. 12:12Come on. Must be [Music] a
62. 12:27media. Here's the Okay, this is where we're going to go. I figure we'll start out with this. So, as you know, I'm a fan of this
63. 12:42guy named David Shapiro. I know I know Pate is not a super fan of his, but the the only two people that I know
64. 12:53well enough to like respect their opinions about math are David Shapiro and Payton. I don't I don't run in math
65. 13:03circles. Let's just say let's just put it this way. at Penn State, the theater building must have been, you know,
66. 13:11geographically opposite of wherever the math building was because I don't think I ever met a mathematician at Penn
67. 13:20State. Anytime I would say it would not be hyperbolic to say I know they're about to say something hyperbolic. It's
68. 13:26the one lesson I've learned in life is is when someone tells you what they're not, it is absolute fact that that's who they are.
69. 13:37Listen, I'm not going to be an [ __ ] but followed by [ __ ] Did Theater X on the X account engage with the dog Kyle? Nope. I got no
70. 13:49love from the uh Penn State theater social media people because they're los theater people. What do you expect them to be good at social
71. 13:58media? Um all right. What am I doing? Oh, I'm ch I'm sharing. Calm down. I know. Listen, I know that I am a little slower than
72. 14:09most. I understand that it just resized my window. Why does it do that? Good lord. Okay, black bar. We got our me. I got
73. 14:25it. Producer Brandon is particularly snappy tonight. I think he had a little extra cup of coffee in between. So, this
74. 14:33is this is this is producer Brandon's third live stream with me today. Twice as participants or as as audience
75. 14:41members and tonight as the producer. So, I think he's a little sick of me. He's jumping in there. All right, let's watch Yan
76. 14:56[Music] video. Yeah, those tend to perform way worse on unseen math problems. But let's see. 01. So, it's
77. 15:10the it's the AIM 2025 and the AIM 2024. And the the model that did really good was 04 mini with tools. And what
78. 15:21was fascinating today is they talked about they talked about these these models having access to tools. And I I guess how
79. 15:36they're how they're defining tools in this context is multiple Python libraries. So it's basically chat GBT
80. 15:44reasoning engine with Python and it's multimodal. And it was interesting Greg Brockman even said in this conversation he said
81. 15:55um he goes you know what's really remarkable? He goes, "This is still just um next token prediction, but what
82. 16:04they've done is they've put in this multi-step reasoning and they've given it access to to write Python and then
83. 16:11all of it is just next token prediction, but it's doing it in this in this, you know, thoughtful sequential way where
84. 16:19it's improving on itself within a within its thinking, I guess, or within its answering process." Um, and they've gotten it sort sort of
85. 16:30fast, efficient, and it's it's pretty cool to watch, too. Um, all right. So, let's let's get going. Yeah, that's my guess based on
86. 16:41how you described it. Python calls or subprocess calls. Yeah, it's a lot of Python calls. And what's cool, P, I I'll
87. 16:47show it here in a second. Um, you know how it used to be, um, if you if you had on a code interpreter, it
88. 16:56would start answering and then it would say whatever process analyzing and then you could open the window and you'd
89. 17:02watch the the Python write itself. Well, it's now showing all that work and and it's showing if it's doing image
90. 17:09analysis, it's showing that it's doing image analysis. Um, and so you're seeing it sort of think, process, get the
91. 17:17answer, think some more, process some more, think some more. It's pretty cool. So anyway, so let's let's see what Dave
92. 17:22has to say here. That OpenAI just made history. Now, to state that in as plain and clear and simple terms as possible,
93. 17:32I want to call your attention to just two numbers on their 03 and 04 performance, specifically 04 mini with
94. 17:41Python. Now this is 04 mini not 04 full. So on the AIM or AIM 2024 competition uh math benchmark they got
95. 17:5498.7% with 04 mini and Python tool use only. I want to just I just want to respond to Pate. Pate says sounds cool.
96. 18:01I think we're getting to the point where it's all just preferences at this point. I agree. I this is something that's been
97. 18:08kind of flying around my head for the past 3 weeks or so that in in essentially every instance of output whether it's video, music, large
98. 18:22language model, coding, math, um any tool that does any of the things that we're trying to get it to do at
99. 18:31this point are so good that you can effectively pick any one and then I think I think to your point the ones
100. 18:39we're going to choose are the ones that's like yeah that feels like me right if if you sort of imprinted on
101. 18:45Gemini you're going to do Gemini if you imprinted on Chad GPT or you like Claude's personality I I think that's
102. 18:51that's ultimately what the tools we're going to use and it and you know it's funny even though there's all of this
103. 18:58stuff happening and there's 12 different video tools and then this is going to be just like any other technical ical um birthing, right,
104. 19:09where you're going to have dozens and dozens and dozens of major players at the phase we're in right now. And so
105. 19:16keeping up with it is impossible, right, that we're experiencing. And then people will start to imprint as as
106. 19:24the tools start to stabilize and and start to, you know, get good enough to use for everything we're going to do.
107. 19:31Like the fact that Open AAI was the first one to give access to all of your previous chats, that's a reason to use
108. 19:40it. Even if you might like Claude's personality a little better, but you've got all this history in chat GBT, you're
109. 19:46like, well, I'll just do this one in chat GPT. And at some point, what'll happen is it'll just consolidate down.
110. 19:52All of the all of the sub players will either get acquired or they'll just they'll just peter out because their
111. 19:58usage will drop off. Um, now how long is that going to take? I don't know. That normally takes 10 years. Seven to 10 years. With
112. 20:08AI moving as fast as it is, that might hap happen in, you know, three to five. But but we'll see. Fascinating.
113. 20:17Um, I felt like when I made that website, any model can do HTML coding. Yeah, exactly. Yep. Exactly. Okay, let's
114. 20:25watch Dave late on AIM 2025. They got 99.5%. Now again this is with 04 mini with Python only. Now you might say okay
115. 20:35you know Sam Alman said we're going to saturate the benchmarks and you know they delivered on that. Now a benchmark is one thing but
116. 20:42what is a benchmark a proxy for? In this case it's a proxy for mathematical performance. Now I as a human being will
117. 20:50never even get to 01 level mathematics. Uh so there is that. But what I want to do is I want to show you the reactions
118. 20:57of some other people and then I'll kind of unpack my full reaction as to why these numbers are just astounding and
119. 21:05incredible. So the first reaction I want to show you is Aiden. So Aiden is an OpenAI employee. He's a really chill
120. 21:11guy. Um he and I talk uh not all the time, but like we exchange messages and like he's just a super chill dude. But
121. 21:18anyways, one of the things that he says is ignore literally all the benchmarks. The biggest 03 feature is tool use. Of
122. 21:24course, it's smart, but it does deep research quality in 30 seconds, debugs by googling docs and checking Stack
123. 21:30Overflow, writes whole Python scripts in its chain of thought for firmy estimates. It does a lot. So, one of the
124. 21:35things that they mentioned in the live stream announcement of 03 and 04 is that in some of the tests, it did 600 tool
125. 21:42use calls in a row. Now you might say, well a human could probably do it in fewer, but the fact that it was able to
126. 21:48keep track of a task that complex means that it is more and more agentic. So that's reaction number one. And that's
127. 21:56important, right? What he's saying. So the the the reasoning models, you give them your question and they talk amongst
128. 22:06themselves and you know do prompt prompt prompt prompt prompt internally and then answer. This is now doing prompt tool
129. 22:13use, prompt tool use, web search, tool use, prompt, right? So, it's using multiple things and it's keeping track
130. 22:20of all that. So, what that starts to look like is you give it an increasingly complex problem and then you basically
131. 22:28just sit back or just go to another tool and do something else. Go make some images in midjourney while it slogs
132. 22:33through all the stuff. Uh, reaction number two comes from Daria. So, Daria is big on Twitter. Um he's a researcher
133. 22:41of chronic fatigue and and immune systems. He's actually helped me with burnout. So Daria is a really good guy.
134. 22:46Um he also often gets early access to open AI stuff. Um Sarah, which which app chat GPT or PO? Not which one to not
135. 22:55sure which one to use or spend money on. Okay. The advantage of PO is that you have access to lots and lots and lots and lots of models.
136. 23:11Um, the disadvantage to Po is that, um, I think they may have their own model, but I'm not sure. I think they
137. 23:24do. Um, they're not as well funded. They're they're they're they're more like a convenience app than than a core frontier lab. Um, if you
138. 23:37want to be able to play with, you know, all of the latest models from whether it's Gemini models, llama models, open
139. 23:45AI models, anthropic models, if you want to use them all in the same interface, um, PO is an okay way to go. I
140. 23:53personally would go with chat GPT because what they what they launched today is staggeringly powerful and what
141. 24:02they launched on March 31st Image Gen is staggeringly powerful. So if you were going to spend 20 bucks a month for me
142. 24:11I'd go with OpenAI. Um so it looks like he had this this post on deck. Um but anyways he's he is a power user of these
143. 24:20things. He he actively uses these tools to help him with his research and I can verify that by using these reasoning
144. 24:28tools uh 03 mini 01 pro and those sorts of things I have gotten I'm not going to say that I'm above his level but I can
145. 24:35talk to him at near his level on things like burnout chronic fatigue and immune related issue. I was on a uh I was in a
146. 24:43Twitter spaces last week or over the weekend um and there was a a Yale um neurosurgeon who was in there and he was
147. 24:52talking about 01 pro, right? So that's not as good as 03 and 04 mini um and he was saying for him it you know it was
148. 25:03absolutely above his capability. He was he said he was with 01 Pro he was submitting a patent application a month.
149. 25:14Um and he said he you know he said we're we're we are in a you know we've passed the tipping point according to him. He
150. 25:22you know he he claimed it was AGI but it depends on the on which definition you go for. Sam Open AI is no longer compute constrained
151. 25:31after Microsoft lost its exclusive cloud provider status. Oh, that's fascinating. Yeah, fascinating issues, at least the ones that I've that
152. 25:40I've uh studied. And what I the that's not that's not really a flex. What I'm saying is these models are smart and
153. 25:47they can teach you very quickly and they're operating at the level where people like Daria, who is a worldleading
154. 25:54expert in this kind of stuff, is also impressed. And by the way, he was impressed with 01 and 03 Mini. Now
155. 26:00they've taken it to the next level. Yeah. So, those are two other reactions that I wanted to share before sharing
156. 26:05mine. And my reaction is here. So, I'll just read this to you real quick because it'll come out better. So, yeah, if you
157. 26:13go to his ch channel, he's got this this post is pinned on his channel. It's Dave Shapi. Dave Shappy. So, what I said is
158. 26:24AI has solved math. Open AAI did it with 04. not as close to solving math, not as competitive at math solved. This is far
159. 26:33bigger than anyone realizes. Let me explain why. First, you need to understand some historical context. Typically, with a IML, you know that
160. 26:41you're getting close to fully generalizing a problem space when you get to the 70% and 80% solution range,
161. 26:47which real quick, that's where we were with the previous generation. in eight months. I want to I want to emphasize
162. 26:53that these models came out in September 2024. In eight months, they closed the gap. So, just keep that in mind. So, uh
163. 27:03however, as we often see, reality is in the edge cases, meaning that the last mile jump from 80% to 99% is often far
164. 27:09harder. It usually takes much much longer to get from like from 0% to 70% when you're starting with a new AI
165. 27:18technology than it takes to get from 70% to 99%. They got to 99.5% in 8 months. But OpenAI did that not in years but in months. Remember 01
166. 27:29and 03 were announced in September of last year. It's been just over eight calendar months and they closed the gap.
167. 27:34Just from a research and development perspective, this is a remarkable velocity. But I'm not talking about
168. 27:39benchmarks. I'm talking about real world implications. This puts a world-class mathematician in every pocket on every
169. 27:45team. Do you know what math underpinn? This puts a worldclass mathematician in every pocket on every team. It's like, wow.
170. 27:56Including your little single startup, including, you know, the 10erson company you're a part of, right? Pretty much
171. 28:04everything. The first order consequences of a semi-agentic AI system that has conquered math is pretty obvious.
172. 28:10Anything that requires math, it can likely solve on its own or with very little d redirection. For for example, a
173. 28:16good friend of mine is in CFD, computational fluid dynamics, which is used for uh oceanography and meteorology
174. 28:22extensively. He's been using these reasoning models since they came out uh and they were helpful to him, but he's
175. 28:29they still needed an expert guiding hand. These might these new models might nuke that. Second order consequences
176. 28:36otherwise known as downstream impacts from this are difficult to predict but not difficult to overstate. Let's put a second order
177. 28:44consequence let's put let's put second order consequences in practical terms. This will accelerate AI research itself. AI research is among
178. 28:53other things math. It's also code. Guess what? These models crushed math and code. Beyond that they are
179. 29:00semi-aututonomous i.e partially agentic. They require less human direction, correction, and oversight. In practical
180. 29:06terms, it means they can use more tools without help, work on larger, longer problems without supervision, and are
181. 29:12less likely to make mistakes around user intent. Guess what else is math intensive? Biochemistry, robotics,
182. 29:19spaceflight, cryptography, nuclear physics, blockchain. So, this is right. This now starts to be the thing where
183. 29:29advancements in those um in those verticals start to accelerate blockchain. All of this has just been solved. Now take this uh now to make
184. 29:39this even more impressive these models did this with one tool Python. Not series of tools not mat lab not supercomputers Python. Now let me
185. 29:48underscore what this means in the long run. Your smartphone will be if you don't know what Python is it's a it's a
186. 29:53programming language. And so when these things are reasoning, if they need to do math, they will spin up what's called a container
187. 30:01um which is like a little, if you've ever heard of Docker, it's think of it like a little computer instance running
188. 30:07in a window inside the chat. It'll write Python code, execute it, get the answer from that, and then continue on with its
189. 30:14reasoning. Be a math genius before too long and a coding genius and a lingu and a linguistics genius. and and and third,
190. 30:21fourth, and fifth order consequences of this technology alone are impossible to overstate. And this technology will only
191. 30:27get better. You know that scene where Tony Stark is figuring out time travel in his kitchen? Yeah, that's the level
192. 30:32of AI math we're talking about in just one or two more generations. If warp drive is possible, these machines will
193. 30:38help us figure it out. So, I hope that this gives you some context as to what this means. Thanks for watching. All
194. 30:46right, enough Dave Shappy. All right, I gotta change my tabs or you can't see [ __ ] So, what we'll do here is I'm just
195. 30:55going to start playing. Now, what you need to know in case it's not patently obvious, I am not a mathematician. Sir, may I inquire? What
196. 31:10were your qualification? What pretail are your qualifications? Sir, talking about mathematics and superhuman approximation, I wonder what
197. 31:20what what your qualifications make such a statement, sir. Uh yeah, I got none. But we can do cool [ __ ] with pictures
198. 31:35and storytelling. I I did have a fun idea. Oh, actually, you know what? Let's just do it. We'll start with it. Um
199. 31:44Okay. So, so if you're new here, I'm at if you're Sarah, if you're still here, I'm at chat GPT. So, one of the things you get when
200. 31:52you get chat when you have chat GPT is you get all of the models. Now, people are making fun of chat GPT as they should because the naming
201. 32:03conventions here are [ __ ] insane. But here's all you really need to know know um f for 99% of what most people are working on
202. 32:18the first option is fine. It's fine. It's fine. It's good. It's really like it's really good. Um archetypal architect dropped his 50th book
203. 32:30tonight. That must have hurt. Did it hit your foot? Oh. Oh, you mean you you published your 50th book. That was it was it was publishing
204. 32:43comedy. Get it? Because you dropped it. Like a book like if it was a heavy book and it fell on your toe. That So
205. 32:52that's where I This is also kind of a comedy channel, like an entertainment. So, I like to entertain while I'm I
206. 32:59well, I can't really teach because I don't have qualifications, but and I'm not really funny, but anyway, I just because that was the
207. 33:06funny part, the the joke part of the dropping was the gravity. Well, if it hit your [Music]
208. 33:23tongue, the comedy gets so much better when explained. Exactly. Exactly. Those comedians that think that jokes are okay to just stand
209. 33:34on their own, I've never understood them. Anyway, all right. Fantastic. So in this list the first three models are not
210. 33:45reasoning models. So what they are is you give them a prompt they give you an answer. Reasoning models you give them a
211. 33:53prompt and they go huh the user asked me for this. Let me think about that. I wonder if I should go search the
212. 34:00internet. Let me go search the internet. It gets back answers. It's like, well, now based on what I've
213. 34:07learned, I had made this assumption, but now I do that assumption. So, they're they're literally kind of conversing
214. 34:13with themselves the way you talk in your crazy straw of a brain about stuff and work through problems. That's kind of what they're doing.
215. 34:22Um, so 03 is the is the um the the uh the evolution or the the next big model. Uh after 01, you're like
216. 34:38what happened to O2? It was a trademark issue. They just went from 01 to 03. Then 04 mini and 04 mini high are the
217. 34:47are the Okay, how do we do this? Okay, if this is 03, a box this size is 03 like in terms of like compute
218. 35:03required and size to keep it in memory and size to keep it on hard drives and amount of compute required to answer one
219. 35:11of your inane questions about I don't know fileo dough. That's that's the that's the the base model, the big model. And then they make
220. 35:21these mini models which have less parameters. This one's probably got a trillion parameters or 100red trillion,
221. 35:28doesn't [ __ ] matter. And this one's got some less billion, right? And so this is the mini model. So what we've
222. 35:34got is we've got 03 MacDaddy and then we have 04 mini and then 04 mini high just basically takes more time. It's this
223. 35:42model just thinking longer. Okay, so that's how they work. Um what's different about these models than previous reasoning models? Previous
224. 35:52reasoning models were not multimodal and I don't I don't think they had access to to the Python the the the tool use. Um so they were just
225. 36:05thinking along. So, so what the what these are is very very um optimized, fast, cheap, incredibly capable models
226. 36:17that can use tools and are multimodal. Multimodal just means it it's been trained on images as well as words and
227. 36:25then audio. So, it can do voice. So, it's it's a it's a very sophisticated model. Okay. And so what we're going to
228. 36:32be playing with tonight is 03 and 04 mini high. All right. So let's see. Let's go with 03. And then I think what I'm going to
229. 36:46do Well, let's see. No, I'll do I'll do um a thing that I did before which was kind of fun. Hang on. Is this it? No.
230. 37:05[Music] which yeah, this is it. Okay. So, I'm going to copy this image. So, let me show you this
231. 37:19image. See that image? I went to Google and I said um I searched for complicated image and that came up. It's a
232. 37:34doodle, right? Like some dude or dudet got stoned and you know this was the [Laughter] weekend. All right. So then I said, what
233. 37:48did I do it in? I think I did it in in uh Oh, I did it in 03. Okay. So, so let's do a new We're going to do new
234. 38:01chat. We're going to do 03. And then I'm going to say I want you to catalog
235. 38:17any item you can identify in this drawing. And I want you to
236. 38:32create a hex color palette.
237. 38:48um that features the 10 most prevalent colors in the
238. 39:03drawing. Okay. So, so in order for it to analyze the drawing, it's going to have to zoom into certain pieces of it and look at it and
239. 39:14analyze it. It can understand drawings because it's natively it's multimodal. It can understand images. And then I
240. 39:21spelled pallet wrong. Is that two L's in pallet? I guess so. No. How do you spell pallet? Oh, one one L1 T palate. Fine.
241. 39:37Um, prevalent. Jesus. Okay. Um, and then identifying like it's got to do an analysis of the
242. 39:52whole image, figure out the percentage of all the colors in all the pixels and sort of condense them down into the top 10. And
243. 40:02that's going to take some Python. So, we should get some some good action here where we can watch it do it. Bring out
244. 40:08the V vin diagram. Yeah, exactly. Okay, here we go. Let's see what this thing does. And I'll expand the thinking when it starts doing it.
245. 40:24[Music] Thinking GBT, what's the matter? Everybody using you? The user provided an image. For the image
246. 40:39analysis, I'll use Python. Look. So there it is. So So in line of it doing the conversation, it's now spun up Python. So it's imported some
247. 40:50libraries. Um it's here's the image path. It's analyzing it. It's doing some arrays, some image ni shape something or
248. 41:00other shape of pixels. It's using a k means of 10 clusters. So I guess k means is how they're going to do the analysis of the colors.
249. 41:09It seems like there might be an issue with the ninit auto setting in an older vision version of scikit-learn. Um, let's see. Let's
250. 41:21proceed. Wait, this should work smoothly without causing any errors. It's I guess it's still thinking, but I don't know.
251. 41:28Maybe it might have got stuck. So, we'll see. It's creating its own app to get the answer. Yeah, it's it's creating
252. 41:37multiple apps. So it's actually what it's actually doing is it's using Python to write a tool. So it's basically
253. 41:43saying, okay, the user wanted some colors. I know from because I'm smart that we could use this K whatever it
254. 41:50called it, the K K means clustering. It knew it could use K means clustering. And so what it does is says I've got to
255. 41:57write a K means clustering tool. And so it'll write that in Python. Now it does look like it's stuck. Let me stop it here and let me
256. 42:05just say um try again. We'll see see what it does. It it that that might have been a premature abort abort but you know
257. 42:17whatever. Yeah, it's creating its own app to get the answer. Yes, apps skarn cluster import k means import pandas as pd.
258. 42:30So, so it's doing some stuff. Just nuts. It's thinking. It's doing whatever
259. 42:45it's doing. [Music] And if you wanted to learn code, you go
260. 43:00figure out all the things it's doing in its little tools, which you know when you watch it when you watch it write code, you're
261. 43:08like, I don't want to learn that [ __ ] Um, does anyone know why AI movie why AI is moving so extremely fast?
262. 43:21Um part of it part of it is because um okay so the so a couple of things almost all of the
263. 43:37AI the the consumer AI large language models and the tools that we're using today are leveraging the same core
264. 43:45technology. It's a thing called the transformer that Google invented in 2017. One of the attributes of how the transformer works is that the
265. 43:59more data you throw at it and the more compute you throw at it, the more thinking time you get give it, the
266. 44:06better it gets. Sam Alman was the one that sort of got this idea first and biggest and said, "Hey, if I can, you
267. 44:20know, raise enough money, we can throw we can go get all the data and and throw all the compute at it, right?" And so he
268. 44:28cut a deal with Microsoft um to be able to use their big data centers and you know whatever the initial investment was turned into a
269. 44:38billion dollar investment and then that turned into a $10 billion investment. They recently just closed a $40 billion
270. 44:45investment on a on a valuation of I don't know $320 billion. So these relatively um low employee count companies are
271. 44:58raising billions of dollars. They're spending most of that money on GPUs on the things that are that are making these
272. 45:08models. Couple other things that that are that are why there's so much money here. So why the acceleration is
273. 45:15happening is because there's so much money. Why is there so much money here? AI's been around for ages, right? AI's
274. 45:24been around for 30, 40, 50, 70 years, you know, depending on on your definition of it, neural networks and
275. 45:31things like that, for a long, long time. Um, AI historically has been very specific to a specific kind of data and
276. 45:40a specific kind of use case, right? So, if you want to do map analysis, you train it on maps with, you know,
277. 45:46different kinds of AI. The transformer kind of unlocks this generalpurpose capability if you use it right. Sam Alman figures this out and in
278. 45:572020 comes out with GPT3. So they did GPT1, GPT2. When GPT3 came out, this was the first thing that started to act like
279. 46:07what we're used to seeing in chat bots today. And then November 30th, 2022, they take that basic core technology GPT3. They've
280. 46:18got the next version which is GPT3.5. And what they did is rather than having an engineering interface which is
281. 46:25designed for engineers to use the tool, they launched chat GPT which was the first consumer interface for AI and and machine learning.
282. 46:39And there's there's if you don't know this, the history of this is absolutely [ __ ] fascinating. The
283. 46:48instructions for how to create chat GPT were in the documentation of GPT3. So in the documentation, anyone could have built using
284. 47:03GPT3, anyone could have built chat GPT. So, OpenAI cuts this deal with Microsoft in in October of 2022. Sam says, "Hey, we're going to
285. 47:16announce a billion dollar investment from Microsoft at the end of October or at the end of November. Tech team, can you guys whip
286. 47:24up that chat GPT thing that's in our documentation? It would be cool to have something to show." And so they did. So like in a
287. 47:34month they whipped up the chat GPT interface because the instructions were in the in the documentation. They just had to execute
288. 47:45it and then they launch it and within five days they had a million users. They went out for drinks the night before.
289. 47:50They they they said they would have been happy with like you know couple of thousand people using it and maybe 5,000
290. 47:57comments on whatever on X or whatever. A million people used it within five days. A hundred million people used it within six
291. 48:09weeks. Six weeks from zero to 100 million users. Let me put that in perspective. The early days of the worldwide web, I started one of the
292. 48:18early uh digital agencies in 1994, company called agency.com. When we started agency.com, like barely anyone was using worldwide
293. 48:27web. Like not not everyone had internet all the time. You still had to your computer had to dial up the internet. It
294. 48:32was it was just jankier than crap. The worldwide web to get from when it started to 100 million users was six
295. 48:42years. Chat GBT was six weeks. It was the fastest adoption of technology in history. And it could code and it could
296. 48:51write and it could do all this stuff. And so what happened was because of the runaway success of that, everyone turned
297. 48:58from anything else they were investing in and they just came came to AI. So So basically starting in 2023, all of the
298. 49:08money, all of the engineering talent, all of the research labs, all of the schools just all went the same
299. 49:15direction. We got to do transformers. We've got to create frontier labs. Meta said, "We've got to get in on this game.
300. 49:21we're going to open source ours. Anthropic spun off. So, so you've just got everyone running in the exact same
301. 49:27direction because of the runaway success of chat GPT. So, that's the that's the seinal piece of of technology that
302. 49:35kicked everything off. And quite frankly, it was just a simple interface like interface matters. You could do the exact same
303. 49:44thing you could do with chat GPT on the developers playground. But if you were on the developers playground, you had to choose
304. 49:51a model. There were all these different models and you had to choose temperature and you had to understand what a system
305. 49:56prompt was versus a chat prompt and like it was just like it was confusing. You had to be an engineer to get it. And so
306. 50:06chat GPT just made it super simple. Tik Tok question. Major ethical cons concerns concerning AI can't tell you what time it is.
307. 50:18Who's going to say it's broken? I don't know what that means. Labor. AI can't tell you what time it is. I just made a post which I regret
308. 50:31not using AI for. That's a good [Laughter] one. Um, if you can't trust it with the basic question, how do you know? I
309. 50:46labor. I don't know. Wait, it means whether you work on a project for 5 days or 5 years. It's not tracking your project. I still don't
310. 50:56understand what the question is. Yeah, it like I it sounds like I do. Are you using um AI labor because what you're saying was true in
311. 51:112022. Like chat GPT is connected to the internet now. So it knows what time it is. It can write its own Python code.
312. 51:23So oh the clocks is are you talking about the clock thing? Well there there's a clock thing which we should
313. 51:29talk about. Network error retry. Um here let let me let me do a new let me let me show you something. Um chat
314. 51:39GPT. So, so you are on to a point that it is maddening, which is this. Make me a clock, a picture of a cuckoo clock.
315. 51:49We'll do cuckoo clock. Cuckoo clock. Um at uh with with the hands reading uh
316. 52:077:30 7:35. No, because that they would be overlapping. 7:25. So they'll be on the 7:25. They'd be on the bottom of the clock like this.
317. 52:17Right. Okay. Reading. Let's make this image.
318. 52:37talking about how it doesn't know when you worked on something this week or a year ago. Um, well that I mean that could
319. 52:45just be a software issue. So chat GPT did just get access to all of your chat history. I don't know like it knows when
320. 52:55it had a chat with you. I don't know if it's got access to that metadata but like that feature is a week
321. 53:04old. So, so part of what your part of what your question is is, if I can paraphrase your question, basically what
322. 53:14you're saying is, if this [ __ ] [ __ ] ain't perfect, why would I ever use it? Um, because it's one of the most
323. 53:21profoundly powerful technologies in the history of humanity. And the opportunity whenever there's a new
324. 53:28technology, the opportunity is when you get in early. If you want to wait for it to get perfect, you can, but but then
325. 53:37everyone will be using it and the opportunity will it'll be much harder to capitalize on the opportunity.
326. 53:44Um, so so be very very careful not to use the fact that it's not perfect as an excuse to sit on the sidelines like understand. Yes, Brandon.
327. 54:00Kyle. Um, yeah, I'm trying I've been having some back and forth with Labor in the Tik Tok comments. Um, so we we still
328. 54:07haven't gotten it. Um, I'm taking a different approach. I think what, um, Labor is saying, Labor, please correct
329. 54:14us if we're wrong still. I think what we're saying here, um, is that the overall operating system isn't stable.
330. 54:22You know, when you're working in coding and logic, you expect very consistent output. And Chad GBT will occasionally
331. 54:30say things like, "I can't make pictures." And you're like, "No, you can just try again." Oh, yeah. I completely
332. 54:37forgot to make pictures. There's a lot of inconsistency. Like sometimes it just You've never talked about Blotato on
333. 54:45this program. Oh, wait. We have That's right. I forgot you sponsored it last week. Yeah. Yeah. Yeah. Okay. All right.
334. 54:52So, really good question then if if that's the if that's the if that's the question. Okay. So, here's what you're
335. 54:59coming up against, Labor, which is um something we're we're all going to come up against. Oh, this says image created.
336. 55:07Oh, they're they're um they're try again. Um their their servers are hosed. Every everybody's using it. Their servers are hosed. Okay.
337. 55:19Um Okay. For all of my lifetime, for all of your lifetime, for all of all of our lifetimes, computers behaved a very specific
338. 55:30way. You would write logical code that had logic, which is why they called it logical code. Well, they just
339. 55:38call it code, but it's got logic in it, so it's logical. And you would put in rules, right? If this, then that loop here,
340. 55:47right? you you would explicitly code in logic so that when a user used it, if the user typed in Bob, it would go into
341. 55:56its code and say if user types in Bob, then you know, make up a last name or whatever, whatever it was, whatever was
342. 56:04programmed here, the output was consistent and predictable based on the logic, right? Okay. Generative AI changes all of that.
343. 56:19For the first time in our lifetimes, computers do not behave the way they've always behaved because of the G in GPT stands for
344. 56:30generative. It's generative. It's not just saying, I know Python code. I'm going to use Python code like I'm going
345. 56:38to copy a block of code. You ask for an Asteroids game. I'm going to go look for Asteroids code. I'm going to copy the
346. 56:45code. and I'm going to come over here and paste it. No, that's how that's how it would have worked in the olden timey
347. 56:51days 3 years ago. Now, in the in the new days, it's actually generating it's actually writing the Python code. So, you ask for whatever, it
348. 57:03writes the code one token at a time or maybe blocks of tokens, but whatever. But it's writing it. It's generating it
349. 57:12each time. And if you ask for it a second time, it may write it different. If if you've ever used any of these vibe coding
350. 57:20tools, it makes you an app and then you're like, "Oh, could you change the background color from blue to white?"
351. 57:26And then it completely breaks the app because it went in and just wrote the code different. Now, part of that's just
352. 57:32bad software design. That's something they could fix. Uh, new Tik Tok pin. Yes. Also, it discredits someone's life
353. 57:39work if they're spending years on a project. I I don't I don't know that it discredits it. Labor. I think
354. 57:52that well, it doesn't discredit it. The person that's been working on a project for a lifetime may feel discredited. They may
355. 58:04feel like, "Wait a minute. What the [ __ ] have have I been doing for 30 years? If if I take pride in my coding and I take
356. 58:11pride in making sure it's clean and clear and all that sort of stuff and then some [ __ ] punk like the dude at
357. 58:17the AI learning lab can come along with a theater degree and write code and claim that he's a coder now. Well,
358. 58:24that's an affront to my value as a human, right? Every single every single knowledge worker on the planet is about to face
359. 58:35the crisis that you're describing. And here's the only thing I know like this. Here's my philosophy on
360. 58:43this. There's you can absolutely judge it, but the reason this channel exists is here's what I know. I know that AI isn't going away.
361. 58:57And I think one of the things that we're talking about here tonight with 03 and the fact that it just made this sort of
362. 59:02quantum leap in quality, um, it's getting better really fast. And so then you only have two choices. It's actually quite binary,
363. 59:15which it sounds like you like code, you like predictability. Here's the binary choice. If AI is not going away and it's
364. 59:23getting exponentially better, you have two choices. You can embrace it or you can ignore it. And if you ignore it, you're going
365. 59:33to be the victim of it. It ju you just are. And if you embrace it, you may not like that how you write code is changing. You may choose to
366. 59:48write some of your code the way you've always written it and then use AI for other things to make other efficiencies
367. 59:54in your work. But what you also might discover is that maybe you got really good at one particular programming
368. 1:00:01language and you always you wanted to move to some more modern platform but because of your career you you were kind
369. 1:00:06of locked into this into this platform. Now you could be masterful at 10 different programming languages and you
370. 1:00:13could take your expertise as a developer, as an engineer and apply that to multiple different programming languages. Or
371. 1:00:24maybe you always wanted to write a novel, but you were always coding and but now you can use AI to write or make
372. 1:00:33images. One of the one of the guys in uh in the AI salon, a guy named Roger Scott, he was a software engineer for 30
373. 1:00:40years, 35 years. Like real keragin, like he's just like he he's just very matterof fact. Um and he discovered Midjourney. And you
374. 1:00:52know how he discovered Midjourney? We were in the Discord early on in the salon. We were on Discord and all these artsy fartsy types were in
375. 1:01:03there making midjourney images and Roger kept asking questions like, "Well, what specific prompt did you give it to get
376. 1:01:11that to to get that look and feel?" Because he would take the prompt that that that person did and he would get
377. 1:01:18something different and he was really frustrated by that, right? He's like, "Why is it different? Why is it
378. 1:01:23different? Like, why is it doing that?" He was trying to figure it out. And so people within the salon said,
379. 1:01:29"Well, just go play. Just go over to MidJourney and start playing." And you know what Roger discovered? He [ __ ]
380. 1:01:34loved making art. And because he had programming skills, because he had logic, he approached midjourney
381. 1:01:42prompting like coding and he got really [ __ ] good at it. And he ended up with some of his artwork in an art show.
382. 1:01:51So on the other side of the existential angst that I hear in your question is there's the possibility that
383. 1:02:00it can free you to do things you you know in a much more amplified way than you could have historically but yeah it
384. 1:02:08people who've been doing something for 30 years they are absolutely going to be confronted by this. So, um, yeah, and
385. 1:02:16and just I was following along in some of the back and forth comments and yeah, you know, I think there's definitely
386. 1:02:20something to be said for the the value of the human in the loop. I mean, this stuff was trained on all of humanity and
387. 1:02:27you're you're really working with humanity on this, but to that point, you're working with it on a platform
388. 1:02:34that two and a half years ago was a research institution that fumbled their way into being this juggernaut of a
389. 1:02:43company. And so when you know they are definitely playing catch-up. I always try to give them a little bit of grace
390. 1:02:49in the fact that they don't have all of their ducks in a row in terms of having uptime reports and you know fail safes
391. 1:02:56for when their systems go down because the research company you know cosplaying as a uh actual company. Um before you
392. 1:03:06hit enter on the this um I was a couple of folks in the comments have noted that maybe 03 doesn't have image capability.
393. 1:03:16It does. It does well or or right now because it keeps but if you change the model you might have better uh luck with
394. 1:03:24generating images at least for the purposes of this demo. Yeah. Yeah. Yeah. Um try again. Yeah. I I think 03 looks like it's
395. 1:03:35buried. Try again. Yeah. And of course, we we encourage labor and everybody to join us at the salon.ai. Yeah. Pop the pop the URL up
396. 1:03:46there if you haven't. So So if if you're not a part of the the salon, go to the salon.ai. Um click on the button that
397. 1:03:55says join our community and that'll pop you over to a Mighty Network site where the where the community is. And there's
398. 1:04:01like a really simple sevenstage um uh whatever you call it uh welcome you know welcome area and and you know step two
399. 1:04:11in that process is introduce yourself. So please please introduce yourself uh and just tell us where you are and and
400. 1:04:18like you know if you're frustrated with this like I totally get the frustration. Um it it looks like the image generation
401. 1:04:24here is not is not working out. I might have um an image here of a clock. Let me see. Clock
402. 1:04:39image. C clock. Cuckoo. Clock is generating artwork. Wait 9 minutes. Clock gears. Steampunky readiness. Beach
403. 1:04:57ball. No, I don't have one here. Yeah, it looks like their servers are hosed. Let me let me start a new chat
404. 1:05:05because maybe I just ended up in some sort of weirdass loop. Uh, make a picture of a clock with hands
405. 1:05:24indicating 7:30. Dennis, I see no issue with using AI as we all build on top of past research. It just gives us knowledge faster and wider
406. 1:05:38access to it. Yep, I would agree with that. I was going to say we have the the fire we have the firewolf. Why the hell
407. 1:05:45not? Oh, the dire do. That's funny. YouTube spotlight. Um, yeah. I I'll tell you one of my um like you know I talked about my
408. 1:05:59philosophy on this site is AI is not going away and so you you know then you have two choices right deal with it or not and
409. 1:06:07listen I have if you choose not to at least try AI don't think there's a good excuse for that if you choose to try AI and then
410. 1:06:20after really giving Even at a good gipper go, you decide, nope, I hate it. I never want to use it. Um, I have respect for that.
411. 1:06:30Um, I lost my train of thought. I don't know where I was going. I was thinking about my clock clock hands. Oh, whatever. Oh. Oh, wait. It
412. 1:06:42was Oh. Oh. Oh, yeah. So when I first started learning about AI, like I started, one of the reasons I started this channel from a selfish
413. 1:06:51standpoint is I wanted to learn everything I could about AI. And so I thought if I if I learn out loud, if I
414. 1:06:58learn stuff and then as I'm learning it, I share it, I'll learn it better. And for about the first year when I
415. 1:07:05really got my head around how the transformer model worked and and how embeddings worked and how the latent
416. 1:07:11space worked and how token prediction worked, it's absolutely incredible. Like mathematically it is stunningly elegant
417. 1:07:20and it's actually weirdly simple, right? You you you figure out the semantic groupings of tokens and you put them in thousand
418. 1:07:32dimensional mathematical space and then when someone writes a prompt you use the same basic basic math to identify a
419. 1:07:40probability region in thousand dimensional mathematical space. It's it's beautiful. It's elegant but when
420. 1:07:47you understand it it's a calculator. Transformers are a calculator and yet your experience of it is this weirdly human poetry writing
421. 1:08:02programming mathy image generating thing and I couldn't reconcile those two. I couldn't figure out what why is
422. 1:08:11it that this thing is got this humanity to it. And someone on Twitter they they they wrote this very simple tweet. I'll
423. 1:08:21never forget it because it was it was this revvelatory moment for me. They said artificial intelligence is the collective intelligence of
424. 1:08:30humanity or AI is the collective intelligence of humanity was the tweet. And what what finally dawned on me which
425. 1:08:41is patently obvious like like all good revelations they're obvious once you have them. The reason it feels so human
426. 1:08:48is what it was trained on. The P in GPT stands for pre-trained. So what what OpenAI did first well and then what every
427. 1:08:58subsequent frontier model has done since is they've gone out and they've gotten as much data as they they could and yes
428. 1:09:06you can absolutely argue that that's probably unethical and possibly illegal. Yes, all of that. What I can tell you is
429. 1:09:13if they had asked for permission, we wouldn't have these tools today. So they went out and they did the San Francisco
430. 1:09:19thing and they said, "Fuck it. We're just going to go get get all the data. We're going to throw it at all these
431. 1:09:24GPUs, but what they actually did was they took the collective output of humanity and they compressed it into
432. 1:09:32this beautifully elegant math ball that we all get to tap into. And what it is in a very literal sense
433. 1:09:42is is you put in a prompt and it's like you're collaborating with everyone that's come before us and it's it's quite elegant. It's quite
434. 1:09:55amazing. Um we as humans still have agency. We can choose when to use these tools. We can choose what to use them for. A lot
435. 1:10:06of creative people that I know, they've got hermetically sealed parts of their work that they're like, "I'm never going to use AI for
436. 1:10:15this." Awesome. Perfect. A good buddy of mine, a creative director, will never use another artist's work in his prompt
437. 1:10:27because that's an ethical line that he's choosing to draw. He's like, "I want to take advantage of these remarkable
438. 1:10:34tools, but I won't invoke an artist's name. I'll figure out how to describe their work, but I won't invoke their
439. 1:10:40name because I don't want to do that." Like, I have total respect for that. Um, but that's what the that's what this is. This
440. 1:10:51is, you know, kind of imagine it like this. Imagine taking all of the artifacts and all of the knowledge of every museum, every
441. 1:11:02research library, every university, every university library and giving everyone who uses this tool instant access to all of that
442. 1:11:14knowledge. That's what AI is. So, so you know when I hear individuals that are alive in in this time window going, "Oh, they stole my work."
443. 1:11:25Well, they trained it on sort of everything and your work likely got swept up in that. Yes. Um, but to identify when when this
444. 1:11:37outputs something, to identify which piece of which artist's work is in there, the the the the closest equivalent I can come is if you're
445. 1:11:47making a cake and you take one piece of one grain of sugar and you throw it in the cake batter and you stir it up and you put it
446. 1:11:56in the oven and you bake a cake and then someone says, "Well, I want my grain of sugar back. It is not in that form
447. 1:12:05anymore, right? It's it's it's molecularly completely different, which when you embed data into a large
448. 1:12:14language model, the original works cease to exist. They get turned into these little fragments like little grains of sugar and blasted into
449. 1:12:22thousand-dimensional mathematical space. The original works cease to exist. The only thing that remains is the semantic
450. 1:12:28relationship between the tokens. Anyway, um Becky Rue has a good prompt for Grock. Create the six for Grock. Oh, the six flag the six flags
451. 1:12:41guy AI image who's really Jeff Bezos. Oh, with a new ride called Blue Origin. Oh, okay. Yeah, because Okay, because
452. 1:12:49Rock will do things. Yeah. Uh it looks like we it looks like uh yield uh chat GPT is kind of useless tonight. So I
453. 1:12:57guess we'll go do other things. Um let's go over to we'll go over to Twitter. So anyway, does that answer
454. 1:13:04everyone's question, Kyle? Irregulars ASAP. Okay, good. Oh, what's this? That looks cool. Am I in a regulars? I posted this to another
455. 1:13:16channel, but it should be in here for AI brand search. What is this? Oh, that's cool. Huh. All right.
456. 1:13:30[Music] Um, nice. They say duct tape has limited usage. Wait, I don't need a plan. I need caffeine. Caffeine duct tape and a mild
457. 1:13:41disregard for the consequences. That's really good. Nice one, Dennis. Too old for Tik Tok trends, but too
458. 1:13:50young for bingo night. Too crumpy for both. That's me. Pretty much me. I I relate to that one story behind the
459. 1:13:59picture. Amid the beautiful chaos. Beautiful. Love it. My dog is a human. Nice. Oh man, that was using chat GPT
460. 1:14:13this morning. That's cool. Let's see. Valerie, I believe our way of life will exponentially change. The hope is for
461. 1:14:20the better. Here here's my here here's my opium for the for the radical change that's coming. Um, and I I do think it's
462. 1:14:28going to be quite radical. Um, I don't think it's going to happen as fast as the the tech bros think it is because
463. 1:14:35just humans humans are humans take a while and they resist change. Um, but I think it's going to be profound and and
464. 1:14:44and profound in ways we can't anticipate. So, if you think about when the steam engine came out, we we were
465. 1:14:51largely an agrarian society. Everyone worked on a farm and everyone worked with horses and everyone worked with
466. 1:14:58their hands and everyone worked with their muscles. And the steam engine came along and it essentially replaced human
467. 1:15:12strength. Now imagine being a strapping back then you you probably wouldn't be strapping until you're 50, right? You
468. 1:15:20probably only lived till you're 45, but you're a strapping 35y old man, which is like old in those days, but
469. 1:15:27you're still still good for your age, right? You're a strapping 35y old man. Along comes this thing called the the steam
470. 1:15:34engine. And we don't need you anymore. We don't need the We only need two strapping men and we've got 10. So eight strapping men and their
471. 1:15:45families are displaced, right? And if you sort of take in the very real fear of today, AI is going to
472. 1:15:55take your job and it's going to replace a lot of people. We only need two. those other eight, we just don't need them
473. 1:16:02anymore. Yes, that's going to happen. What happened in steam engine days? Those people that lived on farms moved to
474. 1:16:11cities and the cities got bigger and as the cities got bigger an entirely new kind of dynamic was created and service
475. 1:16:19economies were created and manufacturing and you know all sorts of things happened as a result of that shift that
476. 1:16:26you you just literally couldn't even imagine well what would I do without a horse I don't know live in a place where
477. 1:16:35all the stores are within walking distance and you don't need one. Oh, wait. Is that possible? That's that's the kind of
478. 1:16:43shift we're at the beginning of. So, the only thing it feels like right now is [ __ ] existential fear, right?
479. 1:16:58But what if for the first time since computers were invented, they did
480. 1:17:12more than make our work more efficient. Right? If you look at the history of computing, it is largely an extension of
481. 1:17:21the industrial age where machines replaced our muscles and made us more efficient at doing physical stuff and then computers came along and
482. 1:17:31made us more efficient at doing, you know, knowledge work and and you know, information processing stuff. But it was
483. 1:17:38still us doing the work and the computer was there making it more efficient. Which is why when people first start
484. 1:17:46using AI, one of the first things that they do is they're like, "Oh, I could automate all the [ __ ] I do and make it
485. 1:17:53more efficient." That's what computers have always done. But they're missing the the relevance of the G in GPT
486. 1:18:07generative. Yes, you can make it to make your work more efficient. You could also use it to completely reimagine how that
487. 1:18:15work gets done that we've never had before. So imagine, and I feel decently confident this is where we're headed, imagine that
488. 1:18:25AI systems are going to be able to do the bulk of the work, the bulk of the work that crushes people's souls.
489. 1:18:39filling out forms, sitting on hold for 4 hours, trying to get your insurance paid for, um, doing data entry because your
490. 1:18:48company still uses fax systems and someone has to type that [ __ ] in. Imagine if all of the drudgery was done by the machines and our only
491. 1:19:02job was to ask the question, how do I want to impact the world? what kind of difference do I want to make? And then because of AI, you have
492. 1:19:13the resources to be able to execute that. That starts to look like a pretty [ __ ] enlightened future where where anyone who wants to
493. 1:19:29do something for the world, for their neighbors, for their family can do it at any level. I've got chronically ill kids with Lyme
494. 1:19:40disease. No one's putting a dime into Lyme research because the insurance companies won't pay for it. Well, we're probably within five years
495. 1:19:52of someone going, "Fuck it. I'll cure it myself." Right? So, all right. People worried about jobs at the
496. 1:20:05bank with ATMs in the 80s. Yeah. It went on to create thousands of new jobs. Every single time. Every [ __ ] single
497. 1:20:14time. Every time there's a new technology, people resist it. Every time. Every time. Every single time. And
498. 1:20:24the more profound the technology, the stronger the resistance. Danielle Mike Myers. Oh, wait. You can't see that.
499. 1:20:35Hang on. Can you pop the screen up there, Brandon? That's pretty funny. Yeah, [Laughter]
500. 1:20:50baby. That's [Laughter] awesome. Oh. [Music] Gorgeous. Gorgeous. Gorgeous. What's
501. 1:21:05that? I've been vibe coding mural application for muralist for the metaquest here. Oh, that's cool. Made some progress. That's
502. 1:21:17neat. I love making these crowd renaissance images. I made this with Google's whisk. That's really cool. I took a photo. I've got a photo
503. 1:21:26that I took in the shuttle train at at Denver airport where it's a photograph of people in the other car, but the glass
504. 1:21:35reflection shows the people in my car and everyone's just looking in a different direction and it very much looks like a Renaissance kind of
505. 1:21:44composition. It's really cool. Oh, Steo did the Austin Powers thing. That's good. That's really good. Oh man. Not in your children's
506. 1:22:00grandchildren's lifetime. I don't know what that's in reference to. Alex, I'm a truck driver. I would love for the industry to go completely
507. 1:22:11driverless. It'll never happen. Um, well, okay. Let me give you another. I don't know if this one's
508. 1:22:29um this is this is me trying to put a put a silver lining on on some of the some of the changes that are
509. 1:22:39happening right now in our government. Um, in order for there to be radical change, there needs to be radical
510. 1:22:52change. So, I would I would agree with you that if everything stays the way it is right now, unions won't allow, you know,
511. 1:23:03driverless trucking, right? Um, we saw the we saw the the dock workers do do their strike. Um, if what's happening
512. 1:23:13right now in government ends up obliterating the the infrastructure of how businesses run. Um there there is a
513. 1:23:26very real likelihood it will be recombined over the next I don't know 10 or 20 years um in radical new ways and
514. 1:23:35and because this technology is accelerating and and because you know capitalism is capitalism um people are
515. 1:23:43going to look for ways to do things more more and more efficiently. Um and and if if everything changes, right, if if just
516. 1:23:51everything breaks apart, um then then it can be much more radically reassembled, right? Um if that doesn't happen, then I
517. 1:24:00agree it's going to take some amount of time for that change to happen. But but but you know, saying that there will
518. 1:24:07never be driverless trucks is kind of like saying, you know, steamowered tractors will will never be able to farm
519. 1:24:16farms as good as horses. Well, for a generation maybe, right? And then at some point it's like, well, yeah, the tractors are actually
520. 1:24:26better at the farming part or at the large. You going to Tik Tok going to throttle you for opium? Yeah, [Laughter]
521. 1:24:41exactly. Change is inevitable, but change will also be resisted. What do you mean except from a vending
522. 1:24:58machine? Oh, change is inevitable. Yeah, you'll never get your change back from a vending machine. Musk
523. 1:25:06says, "Nobody will have work in 10 years and 90% of cars will be autonomous in 5 to 10." Um the the like something I do take
524. 1:25:17umbrage with is Musk didn't say 90% of people won't have work. For most people work will be optional. Those are two very different
525. 1:25:32things. Human beings want to connect and want to contribute. Like we want to have have purpose. Some people don't
526. 1:25:42have purpose and when you lose your purpose, you tend to die, right? You tend to just fade away. There's a there's a very like like
527. 1:25:52I find this incredibly strange that one of the fears of AI is that people will lose their purpose. Now, they may if they're if they remain
528. 1:26:01completely attached to how we used to do it, they will absolutely lose their purpose. But what? So, so if you've got a 30-year
529. 1:26:09career as a coder or as an illustrator or as a as a art director and then AI comes along and it changes how you do that. It doesn't
530. 1:26:21change what you do, but it changes how you do it. Again, you have two choices. You can hang on to the way you've done it before. You can
531. 1:26:31quit and do nothing or you can find a new way to express yourself. And what you have the opportunity to do is say, well, what was
532. 1:26:41the reason that I got into art in the first place? Do I still want to do that here? Maybe I got into art because mommy
533. 1:26:50was an artist and she told me to be an artist and I went into it, but I always resented it and I always wanted to be a lawyer.
534. 1:26:57Well, now go be a lawyer, right? Now go be a mathematician. Yeah, but I can't go to Stanford. And it doesn't
535. 1:27:04m The new chat GPT got 99.5 on a kick-ass math benchmark. It's probably going to be better at math than you
536. 1:27:15would ever be if you went and did four years at a college. So, everyone's going to have that opportunity to do that. the the
537. 1:27:23sort of trope that if people get displaced then they're purposeless and they're jobless. I just don't buy it. I mean the
538. 1:27:36tools the tools are going to go from being able to do [ __ ] for us to being able to do to to kind of spin up virtual
539. 1:27:45employees that work on our behalf 247. So, a way you could think about this is if with your little laptop sitting on
540. 1:28:00the beach, you could have a company of a hundred employees that were all PhD level quality just working for you. What would you
541. 1:28:16do? Like that's the question we get to ask. And that's like why I've got such optimism for this [ __ ] is because what a [ __ ]
542. 1:28:27remarkable time to be alive in history. Are you [ __ ] kidding me? Really? We get to do that? Yes. But if you just sit on the
543. 1:28:41sidelines with your arms crossed like harump. Well, that's not how I used to code or that's not how I used to art
544. 1:28:48direct. The [ __ ] the jet is taken off the carrier. You're going to be on a big old rusty boat in the middle of the [ __ ]
545. 1:29:00ocean that runs out of gas at some point. Do your kids Wait, do your kids the future believe like you do? Um, well, my kids are chronically
546. 1:29:12ill, so their relationship with the future is very different than mine. I like I have from a very young age,
547. 1:29:21there's something about the way my brain is wired, you know, maybe it's the ADD. I don't know what it is. There's
548. 1:29:27something about the way that my brain is wired that I don't need much information in to be able to to pattern match and
549. 1:29:34and see see what's coming. So, I've always just had a really good instinct about what's coming. And so, for
550. 1:29:42me, one of the one of the things that gives me the the most joy in the world is like imagining what it's going to be
551. 1:29:50like 5 years down the road and experiencing that in the present. And so, that allows me to go, oo, if that's going to be the future,
552. 1:29:59here's what I need to do right now to get there. Um, they don't have that. They're like, "Am I going to be in pain
553. 1:30:08tomorrow if I go out with my friends tonight?" Like, their their their worldview is very, very limited. Um, and also they've got a dad
554. 1:30:18who's like, "AI is the best. You guys should love AI." Well, what do you do as a kid? [ __ ] you, dad. Whatever dad wants, I want the
555. 1:30:27opposite of that. Every new software I've helped it implement has come with negativity. Yep. I used AI to learn. Wait, what was
556. 1:30:43that? Someone just had a post up there. What was it? I used I used AI to learn digital drawing and I draw by hand. It's
557. 1:30:52the most amazing tool for learning. Yeah. Like there's a there's a great example, right? If you're an artist and you look at AI
558. 1:31:04and you see someone go, "Give me a picture of whatever and it spits out a picture that's kind of like the picture you
559. 1:31:11draw." Your immediate instinct is that's evil. [ __ ] that. It's taking my job. Another way to look at it is what if you
560. 1:31:20used it as an ideation tool? What if you could spit out lots and lots of images to inspire you and then you draw them because you
561. 1:31:31enjoy drawing? Like like we as human beings, we get to choose where we put our time. But just because a tool can do
562. 1:31:39something as good or better than you can, it doesn't mean it doesn't invalidate that you enjoy drawing. What it might do is
563. 1:31:48devalue the thing you used to be able to make money on. And that's what's going to [ __ ] a lot of people up because then you've got to
564. 1:31:58force, right? You're being forced out of the thing you've you love to make a living by. And that we're all going to
565. 1:32:05have to confront. Tik Tok question. Kyle, could you explain what Musk means when he says people will have the option to work? Yeah.
566. 1:32:15Um so one of the things Sam Alman um Sam Alman has talked about so so there's this concept of universal basic
567. 1:32:26income the ba the basic theory is this the AI is going to get so good and so autonomous that assuming it doesn't kill
568. 1:32:35us all which you know that's still possible but it seems you know increasingly not likely but whatever we
569. 1:32:42may get there. But assuming it doesn't kill us all, in in which case then we don't have anything to worry about if it does.
570. 1:32:53Um the computers are going to do the bulk of the work. And you're like, well, not drive trucks or not uh dig ditches
571. 1:33:00or not. Well, yeah, with robots, right? If Musk has his way, there will be at least a robot for every human on Earth. And he said probably
572. 1:33:10two. So physical work and knowledge work will just be done and it will create so much prosperity and this is a thing I have a
573. 1:33:23tough time with. It will create so much prosperity that the capitalists can't keep it all that they will actually have
574. 1:33:30to give some back right because otherwise people would revolt and you know eat their babies. I probably can't
575. 1:33:36say that on Tik Tok either. So if if I disappear from Tik Tok, you'll know why. Um so if the computers are going to do
576. 1:33:46all the work, one of the concepts Sam Alman put out there was um the idea of universal basic compute. There's a concept of universal basic
577. 1:33:56income, which is if you have enough prosperity, you take some of that prosperity, you give it back to the
578. 1:34:01people as a salary and you get paid to do nothing. Right? Sam Alman says, "What we should do is rather than give people money, we should
579. 1:34:11give them compute cycles. We we should give them compute time, which on the surface sounds just [ __ ] stupid and weird, right? But let's
580. 1:34:21say that plays out and every person on a monthly basis is given enough universal basic compute that they could run an AI
581. 1:34:32that's the equivalent of a 100 employees. So that means that every person could create a company with up to a hundred
582. 1:34:42employees running that company or executing the vision of that company. Well, some people might not want to be an entrepreneur and they
583. 1:34:51might want to might not want to do that. Okay, so maybe they sell all of their universal compute to someone who wants a thousand
584. 1:35:00employees, right? And so then they get money and they can do nothing. So, the optional part is you could choose to just do
585. 1:35:09nothing. I think that's appealing for people that are overburdened with work and overwhelmed with work. Oh, I'd love
586. 1:35:16to just do [ __ ] nothing until you're doing [ __ ] nothing for a month or two and then you're like, "Oh, this sucks."
587. 1:35:23And so the work is optional part is then then we get to say so if we all have enough resources to say I I can choose to do
588. 1:35:32nothing at the point at which that becomes boring for you what do you choose to do knowing that you've got this machine
589. 1:35:42that can execute anything you put your mind to so that that's where he said that's that's what he's talking about when he says work is
590. 1:35:50optional that there will be you know so much prosperity that there'll be more than enough for everyone. It's it
591. 1:35:58effectively all goods and services become infinitely inexpensive. So everyone can have everything they want. And then you start
592. 1:36:08to ask the question, well what has meaning? Well, that's our job as humans. Go make meaning. Tik Tok question. Why are they
593. 1:36:22dictating the way the world works? Who are they? I don't know what that means. Why are they Sam
594. 1:36:37Alman? They're not dictating the way the world works. Well, Musk is. He's doing that for power, right? You
595. 1:36:50know, why do why do rich people do what they do? Why do rich people try to gain and expand their power? I don't know.
596. 1:37:00Because that's that's what rich people do. Um, I would argue that that Sam Alman is not dictating the way the world works. I
597. 1:37:12think what Sam Alman is doing is has has recognized that that this is a profoundly important that AGI artificial general
598. 1:37:24intelligence which people are arguing that the model that they launched today is really the first model that that is
599. 1:37:32that where where it's as good or better than most humans at most jobs. I would probably argue that it's not that yet,
600. 1:37:39but we're pretty [ __ ] close. Um, I think what these companies are trying to do is compete to be the two or three
601. 1:37:54companies that survive the shakeout that's coming in the next five years, which every time there's a new technology, you start out with 20
602. 1:38:01companies and it whittleles down to three logos. There's always three. Like when it was browsers, it was Netscape and AOL
603. 1:38:10and something else, right? It was always three brow. It's always three logos and it always distills down to three logos.
604. 1:38:16So So they're all fighting to be one of those three logos. um what what Open AI and quite frankly what Google's doing right now, what Meta
605. 1:38:30is doing right now, what Anthropic's doing right now is they're trying to get this technology to the point that it's
606. 1:38:38good enough that it can sort of live into this vision that I just described. It's not good enough right now. It's all
607. 1:38:44jankier than [ __ ] So, technology will always evolve. It always has. Um, if they succeed in making the technology that
608. 1:38:57good, the existence of the technology is what's going to change the world. It's not them saying you have to use AI.
609. 1:39:06It's them saying, "I want to create a a a piece of technology that is so compelling and powerful that people change their
610. 1:39:16behavior to do things this new way." Like, could we choose to not use AI and just keep [ __ ] sitting on hold
611. 1:39:27for four or five hours waiting for a customer service representative to not be able to help us and then hang up on us? We could. Do you want
612. 1:39:38to? I would like to have that not be in my life anymore. Wouldn't it be nice that the the IRS creates these tax codes
613. 1:39:50that are so complicated they're designed for only the rich people to be able to leverage the loopholes because they have
614. 1:39:56the best accountants in the world. Well, what if we all had the best accountants in the world? [ __ ] them. And now taxes become infinitely
615. 1:40:05simple, then they'll change how taxes work, right? So, so we could hang on to the way we do things now. I just I like
616. 1:40:13I, you know, I I don't see this as as maniacal as as a lot of people think it is or or perceive it is. I think this is
617. 1:40:25people who truly believe, you know, listen, I was there in the early days of the worldwide web. And what I can tell
618. 1:40:34you is if you were there in '94, '95, 96 in New York City, in San Francisco, in in sort of the major hubs where the web
619. 1:40:43developed, there was only a handful of people that were doing anything in that. And like what we knew was that this
620. 1:40:49technology was going to change the world. And if you look at some of the 60 Minutes um episodes from the era
621. 1:41:00around 98 99 2000 right before the bubble burst when it was at its maximum hype and everyone was throwing all this
622. 1:41:09money at the worldwide web and businesses were getting funded that had no right being funded. Everyone was
623. 1:41:15like, "Oh, this is just a bunch of [ __ ] suckass punks trying to change the way tra change change trying to they
624. 1:41:22believe they can change the world and they're just a bunch of punks trying to [ __ ] up the way we do things.
625. 1:41:30Um, it got very perverted, like the perception of it got very perverted once all the money came in." And so what
626. 1:41:38we've got with AI is we've got a profoundly powerful tool that like the amount of money that's come into AI
627. 1:41:48development makes like it's almost like the web wasn't even funded. Like as hyped up as as as the the.com bubble
628. 1:41:56was, that is [ __ ] nothing compared to what's happening in AI right now. So when you take that much money and
629. 1:42:05then you put it into a tool that the more money you throw at it, the better it gets, you're going to get this crazy
630. 1:42:10ass acceleration. And if it gets as good as it looks like it's getting, it's going to change everything. So I I don't
631. 1:42:17know. I mean, I I like I I just I don't buy the, you know, there's there's two or three people that can just change everything. Um I
632. 1:42:27think current administration is trying to do that. Um, it'll normalize at some point. I think we grew up with sci-fi
633. 1:42:36movies that portrayed cool technology that seemed unattainable. We're now beginning to see come true. I know I talk about we're living in a
634. 1:42:45postci world. Like, we grow up with movies that that a lot of the stuff we're playing with today. That was my very first my
635. 1:42:55very first reaction when I used chat GPT was I'm in a sci-fi movie. And the second thing that popped into my head was the Turing test just
636. 1:43:11just got passed, which it probably hadn't, but at least in my head, like I'm interacting with this thing that's just [ __ ]
637. 1:43:18talking to me. It's not programmed with logic. Like I I don't know if you paid attention to this. People have been
638. 1:43:27talking about AI chat bots for about 20 years cuz AI's been around and chat systems have been around
639. 1:43:36and chat systems that have AI underneath them have been around. But the the way they worked was a logic tree
640. 1:43:47that human beings went in and said, "If the user says this, then answer this." And then at some point it might use AI
641. 1:43:55to go figure out, you know, based on this user, which movie should I recommend to them, right? Like AI was used as as this sort of enabling
642. 1:44:05technology underneath shitty logic trees. So for 20 years, chat bots and AI chat bots were these [ __ ] horrifically disappointing things.
643. 1:44:19And then Chat GPT comes out and all of a sudden it's like holy [ __ ] This is This is like the movies. They They did it. Like that that
644. 1:44:32was like my immediate response was they did it. This is that this is that thing. This is what they've been promising.
645. 1:44:43You know, the week after Chat GPD started, I started the AI salon and I started this Tik Tok channel. I think I
646. 1:44:50started them, you know, like the day or two after Chat GPT launched cuz cuz I've been waiting for this [ __ ] for my whole
647. 1:44:59life and like it's here and we've got a lot of people like sitting on the sidelines like I don't like it. It makes it makes
648. 1:45:10typos. Okay. Do do you know it can write code for you? Do you know there's a thing called vibe coding now where you can just sort
649. 1:45:19of sit there like Rick Rubin with your big [ __ ] beard and your big long hair just going I think I'll make an app that
650. 1:45:26you know talks about the growth of fingernails today. Okay, it'll make that for you. That's a reality today. People are bitching about it.
651. 1:45:37Well, you know, I I don't know. I don't get it. But I do. Change is scary. I got my bot on Discord. Then I was able to
652. 1:45:46give it conversational voice with 11 Labs. Oh, that's super cool. And this is just the beginning. Once you add
653. 1:45:52advanced robotics and quantum Yeah. quantum computing, it gets insane. It gets insane. We we as a species have never ever
654. 1:46:10ever lived in a time where change is exponential. So, so you know how that works. Do you do you know how that
655. 1:46:21works? So, linear change is what we've always experienced. You have a thing here, you improve it and it's, you know, it's this much
656. 1:46:35better and then the next improvement is this much better and it's linear and it's predictable. Exponential change starts
657. 1:46:44out really flat. It seems like nothing's happening at all. David Shapiro talks about this. He goes, it's nothing,
658. 1:46:50nothing, nothing, and then all at once. We're in this part of this curve, the curve that goes Right. 50 years of neural networks and
659. 1:47:03somewhere in here was big blue beat Gary Casper off and then we did the, you know, there, you know, there there were
660. 1:47:09sort of specific AI milestones along the way, but it's basically flat. It has nothing to do with society. And then GPT3 comes out and
661. 1:47:20it's kind of like it's pretty good. And then chat GPT comes out and that's the first thing where we've got a
662. 1:47:26generalurpose tool. This is really important. General purpose tool that can do everything from recipes
663. 1:47:33to kids books to physics to mathematics to programming to business plans to marketing plans to analysis, data analysis. And and so
664. 1:47:49November 30th, 2022, there's a little blip in the flatness, right? And then we're just slowly doing this.
665. 1:47:59And what happens is you don't just go from like one to two, two to three, three to four. You go from like 1 to two
666. 1:48:08to 4 to 8 to 16 to 60 or to 32 to 64, right? The changes are accelerating and they're going to
667. 1:48:22accelerate so fast that it'll be impossible to even comprehend them. And I think that's starting like that's why this feels so [ __ ]
668. 1:48:35weird is we've never experienced this before. If you like checked in on chat GPT like in 2023 and you're like, "Oh
669. 1:48:42yeah, I'll do chat GPT." It's whatever. It's like Google. Yeah, I get it. And then you went away for two years and now
670. 1:48:48you're coming back because maybe you saw that image gen thing. People are making those studio Giblly pictures and they're making the the
671. 1:48:55the blister packed dolls of themselves. Oh, how are they doing that? Let me go check that out. And then you go into
672. 1:49:02chat GPT and you're like, holy [ __ ] what is all this stuff? And then if you look at what they launched today, even though it doesn't
673. 1:49:13work now because everyone's overusing it. Um um where it's it starts thinking for itself. How do you process that? This is
674. 1:49:23one of this is one of the things that that I just it's why I started this channel. I want as many people as
675. 1:49:28possible to try this [ __ ] so that when the the exponential starts to take off, we're at least aware that it's taking off. Most people
676. 1:49:39don't most people haven't even heard of chat GBT. They don't know what it does. They're like, "Hey, it's that AI stuff
677. 1:49:44that's that's for the geeks." They don't get that it's not for the geeks. It's for all of us. You don't have to be a
678. 1:49:51mathematician to use chat GPT. You can How long do you think they will have to throttle this back on us? When When will they turn it loose?
679. 1:50:04Um, listen, I think that I think the frontier models are are in a kind of they're in a bit of a bind, right? If
680. 1:50:19they just turned it loose in all of its power and it really could sort of make itself autonomous and cut us out of the
681. 1:50:28loop and shut down our power grid because it doesn't like the way we do stuff and it makes its own power off
682. 1:50:35somewhere else. Like that could be bad, right? So I think they've got to put some guard rails in there. But then we
683. 1:50:41as humans were like we don't want any guard rails. We're Americans. We want it to do everything. So, I don't I I I I think
684. 1:50:51because these tools are new and because computers are acting in ways they've never acted before, I think having some controls right now is
685. 1:51:02fine. Um, is it really frustrating if you want to make like a horror film and it's like you can't use the word bloody, you have
686. 1:51:12to use a guy covered in red paint. Yeah, that's a bit of a pain in the ass, but at the same time, like, you
687. 1:51:21know, you got to balance the safety while we're figuring out what these tools even are. Start comments and
688. 1:51:26StreamYard rapid fire. Okay, cool. And then time check. Yeah. Okay. So, let's do some Okay. And then, holy
689. 1:51:35[ __ ] I don't know what that was in reference to, but I like it. Um, okay. Valerie, we are at the point in the very
690. 1:51:44near future where if it's possible, it will be reality. You know, I was really struck. Brandon the other night talked
691. 1:51:50about his four-year-old child um who made a song inso um and now says, "Daddy, I want to hear the song I made."
692. 1:52:02Um that kid is growing up in a world where you can ask for what you want and it will it will be manifested, right?
693. 1:52:15You can you can have an idea in your head and bang it's in the world. We're get you know what's you know what also is really remarkable
694. 1:52:26right now we are in we are the last generation that will know what it's like to not have AI.
695. 1:52:42Brandon's four-year-old kid won't know that world. I mean, think about it. When Brandon's fouryear-old is 12, eight years from
696. 1:52:58now, eight years from now, like we've been we've been doing this, you know, if you count chat GPT is the start. Two and
697. 1:53:05a half years. two and a half years we've gone from chat GPT to all this crazy image gen stuff, video stuff, 3D world
698. 1:53:14building stuff, math, programming, talking to dolphins in in in [ __ ] three years. Imagine eight years from
699. 1:53:26now what his 12-year-old boy is going to be able to ask for and get. Oh, you know what would be cool? Hey, Dad. You know what would be cool?
700. 1:53:37What if we started up like a micro micromanufacturing plant where um we took we took garbage from from the store
701. 1:53:46and we converted that into plastic and then we turned that plastic into these cool sneakers that I have and then we sell them to NBA
702. 1:53:56players. And then Brandon's like, "Yeah, cool. Yeah, that sounds good. All right, cool. I'm going to go start that." heads
703. 1:54:04off to his room, doesn't type. Typing is for old people, just talks. Hey, uh, I want to start a new business today.
704. 1:54:11Okay, what's it called? I don't know. Give me 30 names. And then a month later, he's selling sneakers to the NBA. I I like I don't
705. 1:54:25know. Kid's going to own his own robot for his 16th birthday. You damn right he will. All right, Brian.
706. 1:54:32I've been waiting for AI since Eliza in the 80s. I know. Honest to God, like every time Google talked about
707. 1:54:38TensorFlow or the the AI thing. It was talked about on our podcast today about 2018, the the Google whatever event
708. 1:54:47where Sundar Pchai showed the thing where the the AI called the restaurant and we're like, we want that and like we
709. 1:54:53but we don't have it. It's been decades where we've been promised AI chat chat bots and cool AI [ __ ] and AI is going to
710. 1:55:01change the world. Well, now we're here. Now we're here and people are bitching about it. Yeah, but it doesn't spell right all the time.
711. 1:55:18Well, still waiting for the GPT store. I think we'll be waiting that for a while. True, Tom. It's been said that even if
712. 1:55:24they changed nothing with AI going forward, it would take us decades to figure out the things it could do.
713. 1:55:31I just with with GPT40 with the base model, the non-reasoning model, if they stopped it today, they said, "Okay, no more advancements in AI.
714. 1:55:42Everyone's just going to use chat GBT40. We're we're going to we're g we're going to liquidate and and erase all of the
715. 1:55:51code for all the image generation tools. Midjourney's gone. Song things are g everyone everything's gone. Just chat
716. 1:55:57GPT 40 basic chat model. Yeah, I think we could explore it for 10 years and not hit the bottom of it. Absolutely. Oh man, I typed Eliza into
717. 1:56:09my Commodore 64. Nice. Okay. My first imp impression when I use it was, "Oh my gosh, uh, I'm ahead of my colleagues." Yep. Yeah. Yeah.
718. 1:56:24Exactly. Exactly. it. I I find it really depressing that so many people are not
719. 1:56:39only like if people are ignorant of AI, I don't hold that against them. It's the people that are willingly, it's beyond staying ignorant.
720. 1:56:54They're willingly resisting it. It just makes it It's really depressing to me because I kind of feel like if they
721. 1:57:06took that passion and put it into exploring AI and what it makes possible, imagine the [ __ ] difference they could make in the
722. 1:57:15world. And what's going to happen is the AI is going to accelerate away from all of us. And those people are
723. 1:57:24going to be like, "Oh, I get it now. Wait, can I how do I Now, it's also possible that the that acceleration is it gets so easy to use
724. 1:57:36AI that everyone just uses it. Anyone who resisted is like, oh, [ __ ] it. That's definitely a better way to do the [Laughter]
725. 1:57:45world." Oh, man. Okay. All right, people. Some people need to take more time to process. Absolutely. Everyone's a prompt genius all of a
726. 1:57:55sudden. Exactly. Uh transformational information available through AI. Willing resistance is just human nature.
727. 1:58:03I know everybody resisted. I know every single technology has been resisted. It just makes me sad because this one this
728. 1:58:11is like nothing I've ever experienced in my life. And like I want everyone to experience it. I on this channel the the
729. 1:58:17folks that are here all the time know this. Like my hope is that every person I touch experiences their first Kevin
730. 1:58:25Mallister moment with AI which is that was from Home Alone. That moment you can't unhave that moment with AI. And you also can't explain
731. 1:58:40it. You can tell someone, "Oh, it's so amazing." But you just sound like a [ __ ] cult member. Please don't touch the
732. 1:58:57viewers. All right, I'm going to get out of here. So, what's today? Today's Wednesday. Um, I've got a whiskey night
733. 1:59:05tomorrow night. Um, a whiskey club, so I'll probably be late tomorrow and I might be a little tipsy. Isn't he drunk now? He seemed a little
734. 1:59:18drunk. What? What did he say? What was he talking about? I need a [Laughter] drink. So, it'll be a little late
735. 1:59:32tomorrow. And then we got Friday night date night. Uh, I'm going to try tomorrow. I think I've got some cycles,
736. 1:59:38but I got some other [ __ ] to do. But I'm going to try to uh at least get a couple of other use cases going for 01 and 03. And hopefully
737. 1:59:49they'll have their their servers reconfigured or rebalanced so so that the [ __ ] actually runs. Um anyway, that was kind of fun tonight.
738. 2:00:00I mean, I like going on rants. That was a good rant. Good ranty rant. Hope you had fun. Hope you learned something. Um
739. 2:00:06join the AI salon. Go to the salon.ai. AI and click on join our community. That's the one right there. Go to that
740. 2:00:15URL. Click on join our community. Um yeah, that's all I got right now. Just keep coming back to this. Keep
741. 2:00:26coming back. Here's the whole point of this channel, by the way. If you're new here and and you're just like, is it
742. 2:00:31always like that? What is this like? Is it Does he ever It's the learning lab. Does he ever do the teaching? Does he
743. 2:00:37have a a chalkboard? What? How does the Is it a whiteboard? It's just chaos. Every night it's chat add. There's no plan. There's just show
744. 2:00:49up and then what we do is just talk about AI. Why do that? Because be in the conversation. be in the conversation with other curious
745. 2:01:02people with people asking really good like the questions tonight were really good. They were really good. Especially the hard ones like wait
746. 2:01:12a minute isn't this [ __ ] up people's careers? Yes. Isn't that bad? Yes. Is it also a massive opportunity
747. 2:01:29also? Yes. Right. And this stuff is changing so fast that if you're not in the conversation, it it will accelerate away from
748. 2:01:41you. And that's not good. So, just keep coming back. Okay? So, do me a favor. If you're on YouTube, please like, you
749. 2:01:51know, do the do the like thing over there and subscribe to that channel. We're trying to grow a subscriber base
750. 2:01:55there. Okay. Um, you can subscribe there. There are two paid subscriptions. If you want to support the channel, you
751. 2:02:02can do that on YouTube. You can also do that on Tik Tok. Don't know how long Tik Tok's going to be around. So, you know,
752. 2:02:07do it there. And if it goes away, then that comes off your monthly bill. I'm new. I've selected 04 mini. Can you give
753. 2:02:14me a fun prompt to try, please? Uh, love being in the conversation. Thank you, Jeff. Um, a fun prompt to try for 01 mini is, um, let's
754. 2:02:29see, find me the latest data on population growth globally and write a report with pretty graphs in
755. 2:02:43it that talks about the top 10 trends. and the the things behind them. Something like that. Just do something or or take something that you
756. 2:02:58know and make a ridiculously complicated or ridiculously ambitious request of it. Like just give it something that in your
757. 2:03:08mind you know this thing shouldn't be able to do and then just see if it can do it. If nothing else, it'll be
758. 2:03:15fascinating to watch how it how it tries to get there. Um, yeah, the one that I wanted to show you tonight where it analyzed the image and
759. 2:03:26pulled all the the did a catalog of the [ __ ] it found in it, I did it this afternoon. It was really cool. So, we'll
760. 2:03:32do that tomorrow night. All right. I'm going to watch Celebrity Jeopardy. Good night. Good night, Harshwede. Perfect.
761. 2:03:38And that's the perfect the perfect followup to an AI learning lab. Celebrity Jeopardy. Awesome. The cuckoo clock was a great
762. 2:03:50example. Well, it was, except it didn't work. Son of a [ __ ] Did it Did I ever get it? No. Let me see. Try We'll try it
763. 2:04:00one more time. Try again and then we'll get out of here. Oh man, that's the example. Yeah, that's the example. Come on, man.
764. 2:04:14Oh, by the way, labor, if since you're still here, the reason it can't tell time, it looks like this is going to
765. 2:04:20work now. The reason it can't tell time is, you may or may not know this, but every clock ad, every watch ad in the
766. 2:04:33history of watch ads, the hands are always at 10 and two. They're always at 10 and two. Why? I don't know. It's probably some
767. 2:04:43psychological geometric whatever that it's pleasing to humans. So, the hands are always at 10 and two. And so, you'll
768. 2:04:50notice as it's generating this clock, the hands are at 10 and two. You cannot you cannot get it to not do
769. 2:04:59this. Now, what it did do here, which is interesting, is it put the second hand I told it to do 725. So, it put the second
770. 2:05:07hand at 25 25 seconds, but it didn't get the seven. That's because in the training data, there's no clocks with hands in different
771. 2:05:17positions. It's just one of those things. Now, could someone train a model? Could they explicitly train a
772. 2:05:25model with a bunch of different pictures of clocks and watches and things with hands at different positions? Yep, they
773. 2:05:30could. And that would fix it. And so, there's all sorts of things like that. Wine glasses are another one. You can't
774. 2:05:36you can't get it to produce an image of a wine glass that's overflowing because pictures of wine glasses, they're always two/ird
775. 2:05:45full. But then some things it can do brilliantly, right? Like like some of the psychology stuff it does because
776. 2:05:52like all of the writings of all of psychology are have just been out on the internet for decades and so it knows all
777. 2:06:02of that stuff and it's brilliant at that. So that's what it is. All right. I've got a Becky Rue image and a regulars.
778. 2:06:09One final image. We'll go one final tour through a regulars. Becky Rue. Nice. Is that Jeff Bezos with my
779. 2:06:22glasses? That's pretty funny. Nice dodo bird. Solid. Direwolves. Oh yeah. They they just un unextincted the direwolf,
780. 2:06:31so might as well bring back the dodo bird. All right, everyone. I'm gonna get out of here. Have a good night. Enjoy
781. 2:06:38Celebrity Jeopardy. Thanks everybody. Thanks for the really good questions, too.