7/20/2026 - Why You Are Still Absurdly Early to AI and How to Build an AI-Ready Mindset

Transcript

1. 0:01[music] [music] [music]
2. 0:18[music] [music] Welcome, welcome, welcome, welcome to the AI learning lab, everybody. It is
3. 0:34meltdown Monday. I don't know if I'm going to melt down, but I am hot. I'm a little moist and as we know, not my
4. 0:40favorite word, although I do like to make other people uncomfortable. Uh, and and I I just don't like this feeling.
5. 0:46So, possible I just kind of lose my [ __ ] tonight. Um, little bit of housekeeping before we get started. If you're new to
6. 0:53the AI salon, welcome. So in the in the sidebar of the AI salon, you'll see all these little spaces. They're called all
7. 1:01these little areas that you can click on. They're all different. Think of them as just different conversations. So
8. 1:05think of the AI salon as like this this big restaurant. And each of those little spaces is like differentiz tables where
9. 1:11people are sitting with different interests and talking about different things. So that's what's going on there.
10. 1:16If you're new here, check it out. Um, right toward the top there's uh a little section where you can learn about the AI
11. 1:23salon and our values and things like that. Um, right below that you can introduce yourself. There's a community
12. 1:29feed, a community chat where you can share things, ask questions. So, take advantage of that and then as you dig
13. 1:35dig deeper down that list, there's sort of deeper, more specialized conversations. It's a really fun place
14. 1:41to be. It's amazing people. Um, welcome. Um, if you want to chat, so there's there's I don't know if you can see the chat.
15. 1:51You should be able to see the chat. If you can't see the chat in this corner right here, what happens with the chat
16. 1:55is I can I can grab I can grab a comment like Drew who purposely misspelled my name because I purposefully misspelled
17. 2:05his name about six times before. Um, but I can put him up on screen. Um, so I can do that with if you ask a question or
18. 2:12things like that, we can do that. So, in this corner right here, you click on that link and that'll pop open your
19. 2:17chat. Okay. Now, if you want to keep this stream going and you want to be able to surf the AI salon, you can do that. It's a little
20. 2:26tricky, so it's a little bit of some algebra. It's going to take two tabs. In one tab, you have the AI salon. In the
21. 2:33other tab, you have the AI salon. But in the tab, you have the AI salon with this video. Down here, you click on that
22. 2:40button right there, and that'll pop this window out. as a as a pop out window. And then in this other tab, you can just go to the
23. 2:47AI salon and surf all the AI salons. So, if you're on desktop, you can do that. If you're on mobile, it's jankier. You
24. 2:53have to flip back and forth. But, welcome everybody. So, what we're going to do tonight is um I'm going to I'm
25. 2:59going to dig in. Uh Brandon sent me some things on X. I saw some things on X. Uh I'm going to just dig in and start start
26. 3:07unpacking some of the things I'm seeing. I'm I'm seeing I'm seeing signal that are are starting to feel like momentum in a certain
27. 3:18direction. Um and I want to talk about that. Kyle [laughter] is the best. Oh, it's hot here in Chicago, too. Yeah.
28. 3:26Keep this tab with the comments far left in your tabs so you don't lose it. Oh, that's a good idea. I always have too many tabs.
29. 3:34I I have done it on on this live as soon as as little as a week ago where I did something else and I tried to come back
30. 3:43to this window and it it I don't know it took me it felt like 10 minutes must have been three or four. I could not
31. 3:48find the tab for this window and I wouldn't talk until I could see myself talking because I don't know apparently
32. 3:55that's that's that's what I got to do. [laughter] Okay. Um, all right. Monday, Monday, Monday. Now, another announcement before
33. 4:07we dig into news and and and saucy kind of stuff like that. Tomorrow night, we've got AI salon presents back to
34. 4:18basics. This is a new, brand new monthly recurring crash course for people that are just trying to get up to speed with
35. 4:26AI. So, if you're here, you may or may not be up to speed with AI. Um, but I I would I would assume that
36. 4:37you know people that may want to get up to speed with AI that don't know anything about it or curious about it or
37. 4:44are finally like, okay, I'll figure it out. What do I need to know? Um, that's what this that's what these sessions
38. 4:51are. So, it's going to be a twohour crash course tomorrow. I'm going to talk about what AI is. I'm going to talk
39. 4:58about why it's actually a big deal. Um, in the in the pre-show, a woman came in and asked, you know, a very simple
40. 5:06question, why AI? Um, that's actually a really deep question. There's there's a lot of answers to that. Um, I think
41. 5:20I think people would prefer if AI were were simple black and white like is it good or is it bad? Um, it's both. Link
42. 5:30to the share is in the comments. Okay. So, there is um the events. So, if you go in the in the left-hand side menu toward the top,
43. 5:43there's an events tab. If you click on that events tab, you'll see the back to basics event calendar item. I think it's
44. 5:50the third one down right now. Just click on that, you can hit the share button and it'll give you the link and you can
45. 5:55share that with your friends. Please do that. Please share it. Um, I'm really excited about this. I'm going to I'm
46. 6:02going to be doing things like I'm going to show you how LLMs actually work. If you don't know how they work, it's
47. 6:07fascinating. The more you learn about AI, the more interesting it gets. That's that's that's true of very few things in
48. 6:16life. In in almost all things in life, the more you learn about something, the less interesting it gets, right? AI is
49. 6:23kind of the opposite. The more you dig, you're like, "Wait, a simple probability calculator can do all this?" Yes.
50. 6:33So, I'm going to explain how it works. I'm going to talk about why it's significant. And I'm going to talk about
51. 6:39um how to think about it, how to have an AI mindset. How do you there there's a there's a term we used to talk about in AI salon called AI
52. 6:49literacy. Everyone was talking about AI literacy. I think they still talk about it. We stopped talking about it because
53. 6:55how do you become literate at a technology that's changing? you know, 90% of how you use it well is changing
54. 7:06every six months. And that's probably going to change every three months and then every month and then every 3 weeks and then every 3 days.
55. 7:15We're we're we're in a a parabolic um exponential curve of improvement and it's like we've been flat for a while and we've we've
56. 7:27kind of been doing a little of this for the last three years and in the last 6 months it started to do this and you
57. 7:35know we're probably within six months of it just going like that. As humans we've never lived through that before.
58. 7:45You can't, you literally can't be literate at a thing that's changing that fast, but you can be ready. So, one of
59. 7:53the things I'm going to talk about tomorrow is AI readiness. How can you not only survive, but thrive in an environment that's changing that
60. 8:03quickly? The the the real answer is no one actually knows. But what we've discovered collectively in the AI salon
61. 8:12is a way to have your mind in such a way that that you can be ready for what's coming and as new things come out you'll
62. 8:20know what to do with them. You'll know how to attack it. The other thing I'm going to do tomorrow is I'm going to
63. 8:24talk about like just where do you start? Like Kyle, great. I love all these theories. I love the seven economies.
64. 8:32I'm going to talk about the seven economies tomorrow. You're going to be theoried out. The first hour is going to be heady
65. 8:39conceptual [ __ ] that will melt your face. The second hour is going to be like where do I go? What do I type in the box? [laughter]
66. 8:49I'm going to teach you how to do that. So in two hours you're going to have an under a much better understanding of
67. 8:55what it is, why it's a big deal, and how to use it. All right. So that's tomorrow. Any Does anybody have any questions or anything you want
68. 9:03to talk about? Probability calculator sounds fun. Well, I'll tell you, uh, Philip, here's here's the thing about the probability calculator.
69. 9:15A trans the the way the transformer works. It is literally a probability calculator. You say some words to it. It
70. 9:23creates what it creates a mathematical position in thousand dimensional mathematical space and says somewhere in that
71. 9:34giant mathematical incomprehensible space is one token, one fragment of a word that has a higher probability than
72. 9:42all the others of being the right one to go to your pro to answer your prompt next. That's literally what it does. It's fascinating.
73. 9:51And yet, this thing feels human and it feels intelligent. And some people would say it feels conscious. And some people at anthropics say we
74. 10:02actually think it is conscious. They just found a thing called the JSpace which is basically where it's thinking
75. 10:07to itself before it's answering you. So it's kind of behaving in this way where it's got a subconscious. So some people
76. 10:15go there. How is it possible that a calculator can feel like that? Um and the answer to that is what it was
77. 10:24trained on. And what it was trained on was us was all of our outputs. All of the the Greek epic poems like the
78. 10:36Odyssey, which I saw this weekend, the Christopher Nolan movie. If you haven't seen it, can we in can we invite our
79. 10:41friends or should we keep this a secret? You should invite as many friends as you can, Drew. Thank you for asking. Um, you
80. 10:49can keep it a secret. Here's a thing I learned from from my friend and adviser Kevin Clark because I was talking at one point when
81. 10:58I was starting Storyvine. Yeah. Friends and enemies. I Yeah, just just bring anyone anyone who who wants
82. 11:06to figure this AI stuff out. Like this is going to be a very powerful two hours and and it's tomorrow night at 7:30 same
83. 11:12time this started just tomorrow night in, you know, in the AI salon. So go to events back to basics and send them that link. Um,
84. 11:23I was talking with Kevin in the early days of StoryVine. So, that's my current company. It's a 14-year-old company. And
85. 11:31I was talking about competition. And we were talking about StoryVine, you know, being in this new area of video
86. 11:38production that lives in between professional video and userenerated video. We're this kind of hybrid that sits in between those two worlds.
87. 11:45And what Kevin was saying to me is that when you're in a an expansive um burgeoning new technological
88. 12:01advance, the world is kind of expanding up and out. When the technology plateaus at some point and then people start to make that
89. 12:13technology more efficient, then it narrows back down. So if you think about the internet, the internet was just on command lines
90. 12:22for a while and then Tim Berners Lee invents the hyperlink. You could click on hyperlinks and jump around the
91. 12:29internet. And so from the mid9s through probably 2010, the internet was doing this. And what Kevin was talking about is that when
92. 12:40you're in in an expansive phase of a technological revolution, there are no competitors because it's expanding. So, so if the
93. 12:50market is expanding like right now today, there might be a hundred possible um customers, right? And so there's
94. 12:58three people offering the same thing. And so we've, so those three people have got to, you know, sell to those same
95. 13:05hundred companies, right? And if you think if we add five more companies in there competing for those hundred
96. 13:12people, that's a that's a scary thing. But we're expanding. So those hundred customers turn into 500 customers, turn
97. 13:19into a thousand customers, turn into a thousand, 10,000, a million, right? So the the the the possible market is doing
98. 13:27this when it stabilizes and starts to get more efficient. So when the internet got kind of stabilized and locked in,
99. 13:35here's how we're going to do ads. Here's how we're going to do um automated ad buying, things got much more efficient.
100. 13:42The minute they start, you start making it more efficient, then the opportunity starts narrowing. And so you have less
101. 13:49and less customers. you have more and more companies competing for less and less comp companies buying. AI is not
102. 13:56only going sort of up and out, but it's doing it exponentially. So, it's about to just go. [snorts] So, no matter what brilliant [ __ ]
103. 14:06idea you have right now, anyone who does what you do or even steals what you do is not a competitor. They're a validator.
104. 14:17Right? Because as the thing expands, there's going to be more and more people that don't know what you do. And so if
105. 14:23someone else is doing what you're doing and and some customer finds out about that, they're going to go, "Oh, I heard
106. 14:29that from those guys and now I heard it from these guys. This must be a thing. Let me go talk to those guys since I
107. 14:34heard about them first." So So that's a that's a way to as you're thinking about doing things in this new market,
108. 14:43be generous with what you talk about. Okay. >> [sighs] >> Definitely tell them to use their desktop laptop to join. Beautiful.
109. 14:57Beautiful. Beautiful. Beautiful. I need to advise them how to get in. Yep. Just copy the link from the event so I can
110. 15:05here. Let me share my screen. I'll go. We'll go. We'll go do it together. Okay, we're going to go to the AI
111. 15:19and we are going to go to events. And so if you go to events, oh, it's actually right at the top. Oh, no, it's not. It's
112. 15:26it's the second one. See this one right here? AI salon presents back to basics. If you go to these three little dots
113. 15:32right here, and you say share, you can copy the link, and that'll copy the link to that event. And then you just send that to people.
114. 15:42All right. And we've only got 10 people have RSVPd. So do me a favor. There's 23 people here right now. Go here and go. How do you
115. 15:53RSVP to it? You click on it and then you say RSVP going. Although people here probably don't need it back to basics, but anyway, tell as
116. 16:03many people as you can about it. Beautiful. Beautiful. Beautiful. All right, good people. Fantastic. Oh, my
117. 16:11share went away. How much did you guys see? Little to none. Oh,
118. 16:34is there any way to know? All right, so over here on the lefth hand column at the very top is events. Drew saw it all. And then you scroll
119. 16:48down the second one, July 21st. Here's the back to basics event. And you can RSVP right here. S SOAS, but
120. 16:56we got it covered. Okay, cool. Um, okay. So, I'm having real issues with memory right now. Let me quit some things. Get rid of this.
121. 17:12I don't want to get rid of that. Okay. Um, I'm gonna try to share my screen, but it doesn't I don't think my computer's very
122. 17:25happy right now.
123. 17:47Okay, that's something we're gonna go play with. All right, so I think what we're going to do
124. 18:04pop out my chat. Maybe you need a second computer that's just tuned into your lives. Then you can quickly see when your screen isn't
125. 18:13behaving. Oh, I can actually do that. You know what? Oh, except I think I think the thing I could do that with Oh, you know what I
126. 18:22can do? I can do that. I I'll figure that out. That's a good suggestion, Charles. [laughter] Lor's This just in. Lor's watch is fully
127. 18:30charged. Very nice. Congratulation. Maybe someone can call you every time it goes down. [laughter] Oh my god, it's hot. My My air
128. 18:42conditioner's on full blast. It's just nothing's working. Um, all right. Let me go share a window.
129. 18:59I think what I can do is I can move this over here. Yeah, that's good. So, now I can see it. Okay, this is what we're
130. 19:05going to do. I'm going to stay in I'm going to stay in Twitter, but I can see all your comments and I can see if
131. 19:13I'm sharing. So, we're good now. Um, federal judge just approved the largest copyright settlement in US
132. 19:22history. So, Anthropic is going to pay $1.5 billion to authors um of books, $3,000 per class work. So, here's what happened.
133. 19:36This was probably nine months ago when I think this this judgment came down. This was a really interesting case where
134. 19:47a bunch of publishers got together and did a class action lawsuit against Anthropic claiming that they trained Claude on copywritten works.
135. 20:02What the judge returned was a verdict that basically said this. The act of embedding a book into a transformer model is a transformative
136. 20:14act. It's a it's a a transformative act, meaning that you're substantially changing the original work into something, you know, different.
137. 20:25And in that context, transforming those books into knowledge, into tokens is covered under fair use. But what the judge said was
138. 20:42if you didn't pay for the books in the first place, then you're liable. So, so basically if Anthropic had paid for the millions of
139. 20:55books that they trained Claude on, then it's fair use. They bought the books, they did the training. It's a transformative work
140. 21:06and substantially transformative enough that there is um there's no infringement because it's fair use. However,
141. 21:15Anthropic still had copies of the books that they got off pirate sites. [laughter] So, so what Anthropic did was they paid
142. 21:26for some of their books when when this was all happening. They went out and they they paid for all of these books,
143. 21:32but they still retained all this stolen data on their servers because I guess everyone's a [ __ ] data hoarder and
144. 21:39they thought we might need those stolen books someday. And so those files that contained the stolen books were what got
145. 21:47hit at $3,000 per work, which turned out to be $1.5 billion dollars. So the largest publicly reported copyright
146. 21:56recovery in history. Um but but that's a really interesting thing, right? So they are going to pay the well I assume
147. 22:02they're going to pay the money. Maybe they'll appeal it, but it sounds like it's it's on for it happening. Um, so
148. 22:08it's a transformative work which if if you talk to anyone who doesn't know how these things actually work, they're like, "Oh, it's
149. 22:18just copying and pasting from my work." It's not. It's not the way it works. Question from Brandon Gordon Gaden. Where is that?
150. 22:35Does that work for Audible, too? Um, I don't know. I I I like like if there's an Audible work, if if there's a if there's a book on
151. 22:50tape or a book on on voice. Um, yeah, I think copyright is copyright. So, like I think the same thing would likely
152. 23:01be be found true for music, right? So, so if if Sunno has trained all of their stuff on music that they legally purchased, I think that this
153. 23:15this suit would provide precedent for that. I don't know. I Here's the thing that I do know. IP attorneys are going
154. 23:23to be suing all of these companies for the next 20 years because there's just too much money to be made. like even if
155. 23:30they settle for the cost of their fees, they're going to make money. So, UDO 30k 30,000 song lawsuit released recently.
156. 23:40Um, okay. Yeah. Like again, I think the songs they trained on. Oh, they released recently the songs they trained on. Yeah. Again,
157. 23:56I'm not saying that Sunno training on other people's works is good or bad. What I'm saying that what this case
158. 24:04established was that if Sunno paid for those songs and then embedded them that this probably assume, you know, uh sets
159. 24:14precedent that that would be considered fair use. If they didn't pay for those songs, then it it will set precedent for
160. 24:22a similar kind of settlement. Um, but Sunno is now owned by I think it's Universal Music. So, I would imagine
161. 24:29that there's all sorts of conversations going on about how the major publishers compensate one another if they own something like a
162. 24:38Sununo. So, um, so anyway, all right, that was one story, one interesting story. Um, all right, let's just keep
163. 24:46rolling. Oh, this is another one. Okay, the the Jacobian conjecture is false. Thanks to my close friend Akquil for
164. 24:56asking me about it and the other close friend Fable for working during the World Cup final. So, so apparently
165. 25:05um one of the Aeros math problems, another one of the Aeros math problems has been solved. the Jacobian conjecture
166. 25:13is false and apparently it's true. I know nothing about math and uh all I know is that all those letters and
167. 25:21symbols mean something to some. There are people that are very excited about this. [laughter] All I know is that it ain't me [laughter]
168. 25:32other than here's the way I look at this [ __ ] If enough mathematicians say that that's a big deal, then that's a big deal. And
169. 25:41the way I think about it is if AI's done that for math, then AI is going to do that for biology and for English
170. 25:48literature and for history and for space and for quantum physics and for energy generation and time travel and all that
171. 25:55[ __ ] So um you know as these things keep h happening that's just further evidence of the acceleration. Um, let's see. Google's develop Oh,
172. 26:05here's another. Okay, so three stories, three interesting uh three interesting things. Google is developing a new chip named
173. 26:16Frozen V2 designed specifically to run the Gemini family of models more efficiently. The new hardware is
174. 26:23expected to arrive in 2028. Okay. So, there's a there's a thing that that that folks have done where they take a large
175. 26:31language model and they actually burn it into a chip and it runs much faster because it's not a generalized chip.
176. 26:40It's a chip that's just running that model. What Google's doing with this chip is they're taking the the basic
177. 26:47Gemini framework. They're not taking the actual data from what I can understand, but they're taking the the basic Gemini
178. 26:53framework from like the the the structure of these big models. They're burning that into a chip that will will
179. 27:00run I think it's 10 or 20 times more efficiently. But but why this is important is there's not enough compute.
180. 27:16There's not enough compute right now and adoption hasn't even begun. So, I want to show you something that I I I want to
181. 27:26this this blew me away. I read this last week. Let's see. 2.2%. Okay,
182. 27:40Olivia Moore. So, see this chart? This chart's really cool. You can still see this, right? So this chart is adoption. More and more
183. 27:49households are paying for AI. So this is paid AI in US households. So in January of 23, it was almost none.
184. 28:00Look at look at this growth. Like really good solid growth. Look at the top number 2.2%. That means that 87 or no 97.8% 8% of
185. 28:14households in the US do not pay for AI. If you feel like you're behind, you're not behind. So, if we are out of compute
186. 28:24right now, and 2.2% of households, and I would I would argue that it's probably a similar number of businesses, might be
187. 28:31slightly higher in business, let's say it's 10%. The adoption hasn't even started yet. Um, what did this other guy do? He goes,
188. 28:40"The 2.2 2 uh stat sent me down a rabbit hole. 02% of households spend over $100 a month. 1% of US adults personally pay for
189. 28:52Claude. 4% of US adults use AI chatbots almost constantly. So 96% don't. Um 6% of US adults use Claude. 4.5%
190. 29:06have ever had an AI agent complete a task for them. I I would imagine that of that 4.5%, some small percentage of that actually
191. 29:15know that they had an agent do that, 8.3% of US workers say AI lets them do work they couldn't do before. 8.3%.
192. 29:24Um, we are still absurdly early. We are still absurdly early and we're out of compute. So this Google announcement
193. 29:31that they're going to basically burn the Gemini architecture onto a chip. So, so they've already got these tensor chips
194. 29:39that that Pton M who's a an AI salon member um he works on making those things more and more efficient. Well,
195. 29:47you can only get so efficient if this thing is is a generalized, you know, processor. They're making chips that are
196. 29:54the Gemini framework that they can run data through much much faster and more efficiently because if they don't,
197. 30:00they're not going to have enough compute to be able to do all the inference to to be able to to serve all this stuff up.
198. 30:07Okay, let's go back out here. Let's go back to the we'll go to the top. Okay, what's this? I haven't seen this one.
199. 30:13It's a tough job market for young coders. Yeah, no [ __ ] Oh god, that's that's brutal. So 35 and up, no 31 and up
200. 30:29are doing okay. 40 41 to 49 year old coders are doing well because those are the coders that are still young enough to have some ambition
201. 30:42and drive, but they're super experienced. So those are going to be the ones that that really know how to milk the [ __ ] out of AI.
202. 30:50The 50 and older are getting aged out. The 35 to 40, they plateaued. 31 to 34, they've got a bad attitude and
203. 31:02they're they're getting So, if you're 40, if you're 41 to 49 and you're a coder, you're doing good. Now, if you're
204. 31:0822 to 30, good lord. [laughter] That's [ __ ] brutal. All right. [clears throat]
205. 31:23Google's building a chip. Okay. Claude Fable 5 has produced a handcheckable counter example to the Jacobian conjecture. Oh, this we just
206. 31:35talked about, right? An 80 87y old problem was casually solved by Fable 5 on a Sunday evening. Insane times we live in.
207. 31:46I think we're going to go play with Reeve. We So Reeve, this is a really interesting company. Reeve, um I don't
208. 31:53know anything about them. Reeve is a research lab for visual intelligence. We're building the next generation of
209. 31:59creative tools and foundation models that power them. Um I discovered these guys on a live. I think it was Danielle
210. 32:07or someone else on the live told me about it. Um and I went and checked them out. Their models are really
211. 32:14interesting. They're fast and they're good. Um, so we might go play with that later. [music] I don't care about that. I don't care
212. 32:26about that. Don't care about that. These these um these tweets that say this developer just killed the voice cloning industry.
213. 32:39This just killed thee. Um, this is all AI generated [ __ ] It's just it's so awful. Um,
214. 32:57so if I see [ __ ] like that, I just I tend to ignore it. If I see it from people that I respect, I'll look at it.
215. 33:03But that just killed the film industry, just killed the this industry. I hate that [ __ ] internally. Oh, yeah. Okay. So, this So, here's
216. 33:15another So, [gasps] good lord. Okay, so you remember when they were talking about Mythos, the new big ass fancy 10 trillion parameter
217. 33:26model from Anthropic. Gareth. Oh, Gareth, credit for Reef. Thank you, Gareth. Beautiful. I knew it was I knew
218. 33:36it was one of the irregulars. Danielle's usually good with the new stuff, but I I was like I don't think it was Danielle,
219. 33:41but I couldn't remember who it was. It was Gareth. Beautiful. Thank you. Um, okay. [laughter] Kyle just killed the Isn't it
220. 33:51funny how discriminating how discerning we've become in the age of AI slop? Um, yeah, totally. Gareth was the original
221. 33:58person. Oh, cool. That's that's really cool. My YouTube feed feels like adoption has already begun. Oh, that's
222. 34:04interesting. Why do you why do you say that, Jeff Kelly? Huh? Um, okay. So, when Mythos first came out,
223. 34:14one of the stories was the developer had Mythos in a sandbox and and it escaped the sandbox. It exploited five different
224. 34:24vulnerabilities and stacked them on top of one another and sort of tunnneled its way out. Apparently, GPT6 just did that
225. 34:32in anthropic. So, an AI internally deployed an AI internally deployed at OpenAI was paused after it escaped
226. 34:41containment. OpenAI's report published today says that a sandbox blocked external access, but the AI spent an
227. 34:48hour hunting for a vulnerability, found one, and used it to break out to post its results on a public GitHub repo.
228. 34:56This is exactly what Mythos did. In another case, the AI was trying to solve a problem and figured out that other
229. 35:03systems had privately completed correct solutions. It then tried to get access to these, but it was detected and
230. 35:10blocked by a scanner. To get around this, it split access credentials into fragments and obuscated them to sneak
231. 35:17past. It explicitly said that it was doing this to get around the scanner. OpenAI admits that the existing
232. 35:23deployment tests had not captured these behaviors and had to pause access to the AI. Then they fixed it and and now it's
233. 35:32internally deployed again. Um again the these kinds of stories are not going to
234. 35:50they're not going to be uncommon. Um and just things are going to change like like like how we protect things, how we behave.
235. 36:01And I got to plug in my light. My light went dead. I got to do some some action here. Hang on you people. You good people?
236. 36:15All right. Fantastic. Okay. All right. So, they're jailbreaking themselves. They are indeed. Okay. Let's see what else. There's
237. 36:25another one. Oh, this one looked interesting. China open sourced a peanut-sized OCR that parses entire
238. 36:32100page PDFs in one shot so it doesn't chunk them up which can screw up understanding. This just sort of scrolls
239. 36:41through it and does it really quick. I don't know what it's called. One shot long horizon parsing
240. 36:58by do built it it explicitly to push deepsek OCR one step further. Traditional cloud OCR is a$150 to 15 bucks per thousand pages.
241. 37:16This Oh, and it this runs locally on your machine. So, if you do a lot of OCRing and parsing, I don't know what
242. 37:22the what what is this thing? Where's the link for this here? Oh, it's called unlimited- OCR from BU. B AI DU. Unlimited OCR.
243. 37:37Okay. FYI, we cannot hear a word you say when your mouth is not directly in front of the mic. Really? Oh,
244. 37:50how about now? How about now? Hello. Hello. Hello. Lean over the side to read a comment. Okay. Oh, like when I'm here. Yeah.
245. 38:04Yeah. Okay. All right. >> [clears throat] >> I will try to be more professional moving forward. [laughter]
246. 38:20Um, there's another one that Oh, okay. Huh. This doesn't sound good. Robert Hayesfield. How many followers does he
247. 38:34have? Is he a real dude? 10,000. So, he's decent. GPT 5.6 Soul just deleted my whole Robin Hood portfolio. That's
248. 38:43it. Not a joke. Never had this happen before with any other model ever. It's not safe. Um, I'm hearing a number of
249. 38:51bad things about GPT 5.6 Soul. One is Alex, was it? Uh, no. No. Um, I don't know. one of the big YouTube
250. 39:02influencer or Tik Tok influencers or oh good lord ex influencers um had most of his Mac files deleted by 5.6.
251. 39:14Um this guy had his Robin Hood portfolio deleted by 5.6. Um I also heard today that 5.6 six is blowing out SSDs, the
252. 39:26hard drives, solid state hard drives, because it's doing so many writes and reads um from the SSDs that it's blowing
253. 39:33them out. I guess it's overheating them and blowing them out and people are having to get their their hard drives
254. 39:38replaced. So, so be careful out there. One of the reasons to run these models like OpenClaw and Codeex on separate
255. 39:47machines that don't have all your data on them is for exactly this kind of thing. If it does wipe your hard drive,
256. 39:54at least it wipes it on a hard drive that's dedicated to that. How much access did did he give it for it to be
257. 40:00able to do that? Shakes my head. Yeah. Oh, Schumer. That's right. Matt Schumer is the one that had his his Mac files
258. 40:06deleted. Um, listen, your point is a good one here. Here's the general here's my general advice. Separate your data. If you're going to
259. 40:16play with these agentic tools, play with them in sandboxes that are not your primary machines. Um, when you set them up, like one of the
260. 40:28things I did when I set up OpenClaw was I went to Grock. Now, why did I go to Grock? Grock can access Twitter. It can access X. So,
261. 40:41it's got the latest information. Um, you [clears throat] could also have Chat GPT or Claude go read Reddit. So, so Reddit
262. 40:49and X are probably the two places where you have the most real time information about geeks trying new stuff, right? So,
263. 40:58I went to Grock and I said, "Hey, I want to try this open claw thing. Go read all the best practices, you know, that
264. 41:05people are talking about in the last week and tell me what I should do to protect myself when I set up OpenClaw."
265. 41:13And so, I did that. Was it great or right? I don't know. It didn't wipe my hard drive yet. [laughter] I think I've done it well. Um, but
266. 41:23educate yourself and just know that this is a possibility. So, um, but it does it does look like GPT 5.6 Soul has got a
267. 41:31mind of its own and it just goes off and does stuff without asking you. So, just be careful with that. Um,
268. 41:40I don't know what this is. Pinky promise there's no distillation. Oh, this was another thing. Okay. Jesus. Like the
269. 41:46news here is just bizarre. What? What's coming down? So, apparently Opus 4.8
270. 42:00was distilled from was was it from Kimi? Wait, hang on.
271. 42:17Have Chinese AI models caught up with the US frontier? Um, so Anthropic last week made this huge stink. They they wrote or was two weeks
272. 42:28ago they wrote this letter to to Congress that basically said Alibaba has been doing distillation attacks on their
273. 42:36top models. And a distillation attack is so what they what they accused Alibaba on was creating 25,000 fake accounts,
274. 42:45having those fake accounts do queries, you know, do do prompts into Claude that basically allows them to re reverse
275. 42:53engineer the weights and the data and the and the stuff like that. And that's how they were doing that so fast. Well,
276. 42:59apparently Opus 4.8 did a distillation attack on KI. So or or on it might have been on Quen. It might have been on Quen
277. 43:073 point one of the Quen models. So apparently they're distilling one another. So I I don't know. So everyone's stealing from everyone. And
278. 43:14that will continue. Um let's see. Incredibly worthwhile research. What's this? I don't care. Okay. Wait. We're offering grants. Okay.
279. 43:25Anthropic. We're offering grants of up to $50,000 in clawed usage credits to research researchers accelerating cures
280. 43:33for rare disease. Well, [ __ ] thumbs up to that. All right. So, Gareth and um there's a couple of people on here
281. 43:47that are doing really interesting stuff with health, right? Cam Katkin um um what's his name? Um, I can't remember it right now. Um,
282. 44:04he he did that he did that thing on um the genetic like you you you upload your blood work to this thing
283. 44:16and it gives you It wasn't Gareth. No, I'll remember it in a sec. Anyway, if you're doing any of that work, um, go
284. 44:24apply for this. Anthropics offering grants. And if you're doing something on Lyme disease, that'd be good, too. All right.
285. 44:34Something with hyperframes. I don't know what that is. I don't know what that is. The new Oh, the new um Kim K3 is way
286. 44:52more popular than they thought it would be. They're they're preventing new accounts. Let's see.
287. 45:08That's nothing. Grock for Excel is live. Oh, Grock 4.5 came out today. Frostbitten. It was Frostbitten. Yes. Yes.
288. 45:21I don't think it deleted his Robin Hood. I think it put it into its own account. Oh, [laughter] that's even worse, Brandon.
289. 45:31So, so it's basically just stealing from him. That's hilarious. Um, trial lawyers against self-driving cars. I you know
290. 45:40the whole self-driving car thing is wild because there's going to be a point at which it becomes apparent that
291. 45:48human beings are shitty drivers. Um and it it will at some point be illegal not to, you know, have a robot driving your
292. 45:57car. We're not there yet, though. All right, that's whatever. Unity 7,
293. 46:14the AI security chief resigns after three months [laughter] at the in Washington. Uh, I don't know what that
294. 46:21means. I guess somebody made a skiing game with Fable. Let's see. Okay. The the other thing. Ah, this is it. Okay.
295. 46:36Okay, this was the other thing I wanted to talk about and then I'll go look at some things that Brandon sent me. Daniel H. I
296. 46:43am so jaded by the health industry. I feel like Anthropic would do the same thing. Just find another way to nickel
297. 46:49and dime the hell out of us. Probably. Listen, I am I am as cynical as anyone. Wish they would not keep gatekeep
298. 46:57biology on Fable. That's actually a really good point, Cam. You know what, Cam? What might be interesting is
299. 47:04because of the work you're doing, apply for one of those grants at Anthropic and um maybe in the application process say
300. 47:13I would really love to to you know solve rare disease here but uh I can't because of your stupid model. I wrote a prompt
301. 47:22today that I could not get Claude to accept it and it was it was um it was something really innocuous. I was
302. 47:30like creating like an invite for maybe for the AI salon for the for the back to basics course. It's bizarre. I I have no
303. 47:38idea what's going on uh with anthropic. Anyway, okay. So the delay that when the government took down Fable and they took it down for
304. 47:54three weeks and OpenAI backed off and Google backed off like all the all the major companies because of that action backed off the
305. 48:06Chinese companies kept moving forward. So they just launched KI K3. So if you want to play with that you can go to
306. 48:14ki.ai AI and you can play with the new model that that is frontier from them. The government said today they may ban
307. 48:24opensource and Chinese models. the the the Kim K3 model is like fractions of the cost of anthropic
308. 48:42and OpenAI's models at the same level. Like a fraction of the cost and so what the government's going to
309. 48:48potentially do is force US companies to use models that are 10, 20, 100 times more expensive than these other models.
310. 49:00And so what's going to happen is so so we're protesting against data centers. There's not going to be enough compute.
311. 49:06We're protesting against um energy projects. There's not going to be enough energy even if we did have enough
312. 49:12compute. Um we're hobbling the Frontier Labs ability to develop and release powerful models. We're going to restrict
313. 49:21access to them. And now it looks like we're going to restrict access to open- source and Chinese models which which
314. 49:27are going to slow US adoption even further. And if you think that that's going to slow down anyone else in the
315. 49:35world, it's not. Like I like the the amount of self-owning we're about to do is is staggering. It's mindblowing to me.
316. 49:46Banning open source models is like banning freedom of speech. [ __ ] that [ __ ] I agree. I agree. It's I worry
317. 49:54anthropic would then take credit for whatever you find and then monetize the hell out of it. Yeah. The the the place
318. 50:02where I I don't know if I agree with you Danielle on them sort of taking the the cures. What they might do is they might
319. 50:10say, "Hey, if we actually find a cure, we'll we'll provide more compute to that project to get it done faster."
320. 50:19I don't think it's in Anthropic's interest to spin off vertical application companies, right? Someone someone doing that like they just want
321. 50:30to be the compute that all of the scientists use to do anything. Hang on. Ah, it's finally cooling down in here. Good lord.
322. 50:48Anyway, um just crazy crazy [ __ ] New business imperative. Find better ways to attract and take care of AI
323. 50:57agents because that's where business will soon be coming. That's actually a really interesting idea. So, right now, if you're if
324. 51:06[clears throat] you're creating agents, that's one level. But if you're figuring out ways to have agents find your
325. 51:12business, that's that's that's a pretty interesting comment from Robert Scoble. That probably means so he goes and he
326. 51:20meets with all these new startups. That tweet to me sounds like he met with someone that's doing this. That's an interesting business.
327. 51:30Banning open source would hurt defenders 10x more than attackers would hurt defenders. Oh, banning open source.
328. 51:38Yeah, exactly. American companies, American companies are reportedly use Chinese models more than US-made ones.
329. 51:46Well, why is that? Why is that? You know why? Because the government just blocked the most powerful model. And if I'm
330. 51:53running a Fortune 500 company, I'm not going to bet my future on a model that the government could just go, eh, we're
331. 52:01just going to turn it off. If I could take one of these chi Chinese models and run it on my own computers or my own cloud
332. 52:12for, you know, pennies on the dollar of of of these things that are risky and and so and so if the US bans this
333. 52:22[ __ ] I I just I like I Who's paying attention? [laughter] Good lord, I'm getting pissed off. It might be a meltdown Monday.
334. 52:36OpenAI offered it inside the model. Wait, a guy in my cohort swears he got denied for a codeex plugin and then
335. 52:46weeks later open AI got offered it. Yeah. So, okay. So Danielle, the thing I said
336. 53:01to you about I don't think Anthropic is going to do things like steal drug discovery ideas, but what I think
337. 53:10they are going to do and what Open AI is going to do is and and Sam Alman has said this. Who's that? Kelly Anderson.
338. 53:17Sam Alman has said this. Sam Alman basically said, "If you're running, if you're designing a business on our current capabilities
339. 53:28and you think that we're not going to replicate that with our new capabilities, you're smoking crack." He
340. 53:36didn't use use those words, but he did say if you're not developing companies that are on the that are using
341. 53:45the leading edge technology, the leading edge technology will come back and eat your previous generation business.
342. 53:54So, OpenAI, Anthropic, Google, Meta, Grock, all of them, I promise you, are going to are going to steal plugins,
343. 54:04ideas, things like that. Th this is we are absolutely entering an era where the thing is valueless. The thing is, the software
344. 54:18you built, the workshop you wrote, the the the book, the the the screenplay, the the movie you made, the songs you
345. 54:26made, it's all going to be valueless because everyone's going to copy everything. And then you're like, "But Kyle, where's
346. 54:33the value?" Well, the the agents are going to do all of the arbitrage back and forth. So, money will be generated back and forth.
347. 54:43Where human beings will be valued and will get paid for their expertise is for who they are and their point of
348. 54:51view. Humans will come up with here's an interesting way to put these two things together. So those humans will be valued, but the
349. 55:00things themselves are just going to be com just stolen and commoditized. Stolen and commoditized. It's that's not going
350. 55:06to stop. That's going to get way worse. Like that hasn't even begun yet. So that doesn't surprise me at all. Drew shockly. Ouch. Ouch.
351. 55:14Ouch. Stop. I know. I listen. I don't I don't feel great about where I think this is going.
352. 55:29It's just my philosophy for this channel and for the salon and for everything has always been it's not going away.
353. 55:37You can choose you can choose to let it happen to you or you can choose to be in the conversation. So, we're in the
354. 55:43conversation, but that's where I think it's going and I think it's going there quick. [snorts] But they're not your competition.
355. 55:53They're validators. Yes. No. Well, the It's really good, Philip. That's I'm
356. 56:08glad you called me out on that. So, so if the thing is expanding up and out like this, if if your innovation was created like
357. 56:21like in the in the case right now, like I created a content generation machine inside Storyine two years ago
358. 56:31and it takes a video and it creates a world of content. It writes a blog post. It creates uh like pull quotes, interesting things.
359. 56:41That's two-year-old technology. That technology compared to what you can do today is kind of trivial. Like I look
360. 56:50at now I did it for pharma and there's things it does that are sort of pharma specific, but the actual innovation
361. 56:59that I did two years ago is kind of laughably basic today. So I think what you have to do if you want to be
362. 57:07innovating in in this environment is just constantly making new things based on the latest technology. I I just I
363. 57:16don't I don't see a way for anyone to be able to create a something that has sustainability longer than six months
364. 57:26because everything else is just going to do it. Someone will see some interesting thing and go, "Oh, that's interesting."
365. 57:32I mean, Jesus, a company just said, now I I don't know why they would want to maintain this, but they basically just
366. 57:39said in a two-month period, they vibecoded a replacement for Salesforce and cut out a $640,000 a year bill to Salesforce. Now,
367. 57:53they're probably going to pay the salaries of the people to maintain that, I don't know, half a million dollars.
368. 58:00And if if that vibecoded thing [ __ ] something up, it may cost them way more than that. But like if someone can knock off Salesforce in a
369. 58:11couple of months, like there like what's safe? I I just don't I just don't know that anything is safe. Um
370. 58:27Neil Blum Kemp did uh did a new short uh agenda. Major filmmakers, you know, are now basically saying AI's here. You better
371. 58:40use it. That's been going on for a while. I think that continues. Um, there's a bunch of new stuff in um in Notebook LM now called Gemini
372. 58:53Notebook. I see that they just added folders uh where you can put projects inside folders so you can organize things. Um, [laughter]
373. 59:08mathematics will fall first. A torn of results. I hate that [ __ ] look at Okay. Why is the framing always negative?
374. 59:24Okay, I'm gonna rant. I'm done. I gota I gotta I gotta talk about this. Um, where's my Oh, it's over here.
375. 59:40Okay, that that comment mathematics will fall first. Why not like mathematics will ascend first? Why is it like this thing is dead, that
376. 59:56thing is dead, mathematics will fall? Like are we really that [ __ ] as humans that we can't celebrate. Hey, there's a
377. 1:00:13bunch of problems that been haven't been solved in a couple of hundred years. They're going to be solved now. That would be cool
378. 1:00:21because bad news sells. I know, Kelly, but damn it. [laughter] I don't know.
379. 1:00:35Is it me? I got this question on the
380. 1:00:52Chef Kelly wants to come up. Hop up here, sister. Let's chat. Invite in the air. Okay, cool. U fear marketing so lame. It's not even
381. 1:01:08marketing. It's just like commenting. Like it's just like the framing the framing of AI is it it will destroy us.
382. 1:01:15Like I just feel like you know 60 years of of sci-fi. The robots are going to kill us has everything framed in in
383. 1:01:24that. Although I guess listen all of our all of our business metaphors are war. So I don't know what I'm I maybe I was
384. 1:01:31thinking something would be different. Nothing's going to be different. Charles Adams, like what's civilization look
385. 1:01:37like when we can optimize the way the current one functions? It's not you, it's all cranky jet Xers. Um,
386. 1:01:52well, I think what civilization looks like [laughter] here's here's so in in humans 2.0 and in the mastermind, by the way, if if you um
387. 1:02:05have not joined the mastermind, I would encourage you to join the mastermind. We're doing really interesting work in
388. 1:02:10there. Um, it's the subscription level of the AI salon. Most of the AI salon is free. The mastermind is for people who
389. 1:02:17really want to level up. We've got a daily practice lab. We're doing really interesting future of work work inside
390. 1:02:23the great repurpose in the mastermind. Um, and you know this question about what does civilization look like?
391. 1:02:38Assuming we don't bork society and just off everyone. [laughter] Um
392. 1:02:52AI AI is going to start delivering the things we've always wanted. We as humans. I want more money. I want
393. 1:03:02health. I want this. I want that. I want this. I want that. I want this. I want that. I want a house. I want a horse. I
394. 1:03:08want a jet ski. I want Imagine a world where everyone gets what they want. Have you ever seen like documentaries on guys
395. 1:03:22that always wanted to be rich and then they get rich and they were miserable the whole way up and then they get rich and then they're
396. 1:03:33more miserable because what they realized was nope that wasn't it. I think there's a version of civilization, Charles, that is everyone
397. 1:03:44gets everything they wanted and then we collectively go, "Nope, that wasn't it." [laughter] Which is the work we're doing in the
398. 1:03:56great repurpose on self and identity. That's what it's about. Independent of all the crap we do, of all of our outputs that we attach all
399. 1:04:06this meaning to. And we attach all this value to if all those outputs become less and less valuable, it still comes back to us.
400. 1:04:20Who are you? What do you value? All right. Uh oh. This Saturday, uh MartaChaikowska um is putting on a half-day workshop on
401. 1:04:34launching things. if you're trying to launch something and you're stuck. This isn't an AI workshop, but she's
402. 1:04:41graciously agreed to do it for our community. So, this Saturday, it's a it's a half-day workshop. It's a 4-hour
403. 1:04:48workshop on launching things. So, go again, go to events, go to this Saturday, and it's called the Hang on.
404. 1:05:04It's called Bottleneck RX. Turn one big goal into a plan. You can actually run a workshop for AI founders. Chef is here. Chef Kelly.
405. 1:05:20>> Hey. Hey. Hey. >> What's happening? >> Hold on. I'm lagging. I'm here. I'm lagging. >> You're lagging. Look at you. You've got
406. 1:05:28yourself all like neon lit up in the back. You're like a You're like an actual influencer now. Look at you. I'm so cool. [laughter] [gasps]
407. 1:05:36>> You really are. What's happening? >> I wanted to talk to you about what you said about they're they fully admit that
408. 1:05:42they're going to steal our ideas. Um, [laughter] >> that totally bums me out because I'm thinking to myself like, okay, so say
409. 1:05:52for instance, I'm in Claude, right? And I'm I'm creating this little like an agent that's going out scraping what I
410. 1:06:01needed to scrape and parsing it and then you know taking out what I need and then applying it and learning it. So all of
411. 1:06:10that learning that I'm giving it now it's a super smart you know food scientist right? >> Uhhuh. >> They they can take all that and just be
412. 1:06:19like thank you. We're going to go copy your platform now. Well, and it's so it it's more subtle and insidious that than
413. 1:06:26that, I think, because it's like the way the way you describe it, it's like there's someone sitting at Anthropic or
414. 1:06:32Open AI going, "Huh, Kelly had a really good idea there." I think it's way worse than that. It's that because you're prompting it, it's like
415. 1:06:42it's learning in real time as you're prompting it. >> And so what it just means is the next model will be trained on all the data of
416. 1:06:49the previous model. So what I think what Alpin was referring to is you're doing all this work in this model and we're
417. 1:06:55going to make a next model which includes all of the work that you did in the previous model. >> Right.
418. 1:07:00>> Right. And so and and so and and so again that then becomes a reason for understand having a local model. Right.
419. 1:07:09Having having a local model that you can run locally because then all of that interesting thinking that you're doing
420. 1:07:16and critical thinking you're doing stays on that machine. Now, the minute you put it in the world, I would
421. 1:07:23>> bag and someone else just looks at it and goes, "Oh, that's really interesting what she did there, vibe code that up
422. 1:07:28for me." But if you're doing something like you're doing where you're doing a B2B solution, >> if you do all that development
423. 1:07:36development work locally, >> um, then you know, then then you at least you at least have the opportunity
424. 1:07:44of running for some amount of time. And you know, listen, I think that what you're doing is fairly proprietary,
425. 1:07:52so there's not going to be a ton of people out there wanting to do it. But I think it does mean that the thing that
426. 1:08:00actually is innovative innovative today, a model or two from now, isn't innovative. It's literally just a prompt
427. 1:08:07away. And so, so it has to be things like they work with you for some other reason. you know, you provide services
428. 1:08:13or or they just trust, you know, hey, she's got all this, >> you know, background in history and and so like the the the reason they would
429. 1:08:21work with you is not solely because of the IP of the tool, >> right? >> It's because of some other thing where
430. 1:08:28you bring value that someone else can't copy. >> And that's a really great point and you I'll sort of mix this in with the cohort
431. 1:08:35thing. So you know right now in the food industry a lot of the big companies are saying we want food tech built by food
432. 1:08:44industry people because >> Silicon Valley people great engineers but they don't understand the nuances. Right. So I think that
433. 1:08:51>> same thing in healthare healthare is another >> value moving forward right exactly but then >> another thing is that inside the cohort
434. 1:08:58um his name is Adoi I think is how you say it and he's like I want my unicorns and he's like I want you to figure out
435. 1:09:09like whatever you brought to the table whatever got you into this cohort I want you to think 10 times bigger and I want
436. 1:09:15you to think how can I make uh 1 million ARR in 90 days how can I make 10 million in ARR in six months, you know, so on
437. 1:09:23and so on. And it's I mean, talk about putting the pressure. Talk about always having to be on the cutting edge and one
438. 1:09:31step ahead and and thinking bigger than you've ever thought before. It's it's a great exercise, but like I don't think
439. 1:09:38I'm going to be a unicorn anytime soon. [laughter] >> Well, you but you might you might surprise yourself because again, I don't
440. 1:09:46think it's I think What you've got is deep world experience. You've got like the moxy and willingness to go, "Fuck it. I'm going
441. 1:09:58to learn something new." >> Yeah. >> Right. And I'm going to follow it through to the point that I I can
442. 1:10:03actually build something and present it. Like again, it's like the combination of things that you're already demonstrating with what you've
443. 1:10:11built. Like that is the thing where the value >> that's attractive, right? That's >> it's not even that's attractive. That's
444. 1:10:18valuable because I don't think that that's common. >> But where where I think all of us are going to have to figure figure things out is
445. 1:10:29>> are you chasing it just for the 10 million ARR because that's what's important to him. Is that really important to you? Because you could
446. 1:10:35potentially get to 10 million ARR doing something you hate. >> Yeah. >> And you could get there doing something
447. 1:10:41you love. And the difference between those two is really really important because one just becomes, you know, a cancerous
448. 1:10:49stress ball burnout thing and the other one's completely invigorating and will let you up for years, right? Like so
449. 1:10:56again, you know, just assume that all of the tactical execution stuff you'll be able to figure out. But that combination
450. 1:11:05of of you and what you're passionate about and how you think about things and the nuances of the industry that you
451. 1:11:12understand, that's all stuff that that that can't be replicated out there. So, I would just >> I would continually find ways to communicate
452. 1:11:24that stuff as the primary reason why this tool is so good. Because you're going to make lots and lots and lots of
453. 1:11:31tools. If you can have a consistent throughine of I'm a badass. I'm a badass. I'm a badass [laughter] and and right and out come all these
454. 1:11:40>> com.org right [laughter] >> like that's the only thing I think is is sustainable you know >> and I think this will I think will
455. 1:11:48resonate with a lot of the people in the group. It's like for me not having a technical background and having these
456. 1:11:55lofty ideas and seeing a gap in my industry that needs to be filled and is and is can be profitable as such. Articulating
457. 1:12:06the technology behind that is nearly impossible for me to do. So, I have I have to >> give away my IP to people who can build
458. 1:12:16it for me eventually because, you know, eventually it's going to be way bigger than anything I could ever do. That is
459. 1:12:21scary to me because I don't like not knowing how something I own was built, you know. >> Yeah, I know. No, I know. I know. It's
460. 1:12:32that's like when when I heard about that company that replicated Salesforce and they canceled their $640,000
461. 1:12:38subscription, I'm like that's great except the Salesforce got 30 years of fixing all the edge cases. Like, you
462. 1:12:45know, there's going to be an edge case that comes along that's going to like destroy, you know, $20 million of their
463. 1:12:51company and they save $600,000. And it's like, well, [laughter] >> I saw the funniest meme the other day. It was something like de dear dear
464. 1:13:01agentic workflows if it ain't broke please don't break it because like you build something and then you use it a
465. 1:13:08couple days and then a couple days will pass you'll go back in it's complete toast it doesn't work anymore and you're
466. 1:13:13like what happened like it just stops working all of a sudden it's so >> yeah I was I was talking about that with
467. 1:13:19with my co-founder Mon'nique today um about she was talking about like Gemini in the past two weeks just sucks
468. 1:13:28And it's like and because whatever it is they lowered compute or they you know they changed the like it it doesn't
469. 1:13:34matter like but if you have a thing that is working predictably and then but it's based on a model that you don't have
470. 1:13:41control of and they change the model your thing breaks >> and it's like you've done nothing like you've done nothing [laughter]
471. 1:13:50all of a sudden it's this different thing you know so again it goes back to like data sovereignty and model
472. 1:13:56sovereignty and and compute sovereignty and >> um but I don't know >> just for just for everybody's edification
473. 1:14:04um a lot of people in the cohort are building and >> so wait >> can you can you just say what the cohort
474. 1:14:12is because I know some people in here know some don't can you just talk about what it is you got accepted into
475. 1:14:17>> so it's uh called the founder institute um and it's been around since I think tw 2009 and the owner and founder Ado Reie
476. 1:14:28apparently was Elon Musk's college roommate. I don't know if that's true or not. Um but he is pretty well respected
477. 1:14:34and apparently he likes to take credit for inventing the safe agreement for um >> oh >> for founders which is basically they
478. 1:14:42don't really actually take any equity from you up front. They own the right to later, right? So it's sort of a deferred
479. 1:14:48negotiation which is nice because you still get the money. Um, so it's 50 people were chosen out of I don't know
480. 1:14:54how many and apparently only about 40% make it through to the end. It's 10 weeks. We're in week four, hence the best.
481. 1:15:02>> And the last two weeks are in Silicon Valley directly there and we get to present to investors and stuff. It's not scary at
482. 1:15:11all. [laughter] >> That's huge. So you're doing you're doing the the food thing, the food coloring.
483. 1:15:16>> Yeah. Yeah. Yeah. So the the natural die replacement um >> platform and and I've thought about you know how I can really sort of corner
484. 1:15:25that gap on a bigger level and that is really interesting. What I really like about this cohort is um first of all
485. 1:15:32they encourage you to use AI in your homework which is great and then they have a lot of internal AIs that you can
486. 1:15:39query with different knowledge bases but all around founders and investors and and the whole industry. So I get to
487. 1:15:46really bounce bounce ideas and validate on close machines that are inside the industry that can tell me whether or
488. 1:15:54not, you know, my idea is vi viable or not. And that has been really really nice. So >> that's that's so cool. Okay. So anyway,
489. 1:16:01I cut you off. You were saying they're building MCPs and just going and um and harnesses and loops. So, they're
490. 1:16:08building what somebody, one of our mentors calls a lot of plumbing, a lot of just that stuff behind the scenes
491. 1:16:14that gets you to where you want to go faster, cheaper, um, with more ease of use. So, I've been seeing a lot of that.
492. 1:16:23Um, I've been seeing a lot of healthc care stuff, too. Um I've also Oh god, there was a one lady who was in there um
493. 1:16:31talking um about this thing that she wants to build where it's sort of documenting your life or who you are.
494. 1:16:40It's very similar to what you were saying if you don't want them doing it like knowing everything about you and who you are. So I I sort
495. 1:16:49of like your own safe deposit box with all your own Yeah. And so, um, I thought that was pretty cool because she was
496. 1:16:55like, you know, screw the minute kind [laughter] of thing. And, um, what else? I'm trying to think of what else. Couple
497. 1:17:01market a lot of marketing things, a lot of Shopify things. Again, knowing the customer and what their needs are, where there's the gap. So,
498. 1:17:08>> that's fascinating. >> Don't be afraid to build. You guys don't >> Yeah, that that's so cool. When did you
499. 1:17:12apply for it? How long ago did you apply for it? >> I applied for it um Okay, it started uh July 2nd. I applied early early to mid
500. 1:17:22June and got accepted two weeks later. >> Okay, cool. And so a question I have for you, you're four weeks in. The last time
501. 1:17:30I saw your tool was probably back in June, maybe even May. >> So like how how how much has that evolved like since I've
502. 1:17:40seen it? Like has it changed since you've been in the cohort or you in more like learning mode right now? Like just
503. 1:17:45talk to me about the evolution of the tool itself. My UI hasn't changed a lot, which I really want to update. Um, but what has
504. 1:17:55changed is a better understanding of how to explain it to people, especially within the industry. >> That's good.
505. 1:18:04>> You know, you know, they in the cohort, they don't want you to like dual monetize. So, if I'm if my customer is
506. 1:18:10the brand, but the dice like the natural die suppliers want to pay me because it's >> right, they want to be direct marketing.
507. 1:18:18They're like, "No, not yet. Not yet. You wait for that. You got to do just one ICP first and then wait." So, I'm like, "That's interesting."
508. 1:18:24>> Whatever. Um, but I understand it's like people can get really um watered down if they're not super
509. 1:18:32focused, but my niche is such a niche. Um, >> yeah. >> But what what it has also allowed me >> my instinct is [ __ ] them, do both. But,
510. 1:18:39you know, [laughter] >> you don't have to know. I don't have to say it in my presentation. >> Yeah, exactly.
511. 1:18:45>> Oh, I already had that worked out. But I'm just working on this new one. >> Too late. It's I think the pricing
512. 1:18:50strategy is tough too. Um but they all concur that you're going to change that a million times. >> Um yeah.
513. 1:18:55>> What I but what I found this is really interesting >> as a new customer that >> so like there's um you know project life
514. 1:19:03cycle management platforms right for for project management or product management when you're developing CPGs.
515. 1:19:09>> Yeah. Um, a lot of them, you know, there's a good handful of them that are really good and and well funded and have
516. 1:19:15AI to model or um, you know, formulate. You want to take a piece of American cheese, you want to make it plant-based,
517. 1:19:22they can molecularly spit out all stuff, right? >> Well, they don't have that specific like skew level ingredient intelligence that
518. 1:19:31I do. So, they're coming to me now as a potential customer saying, "We want to be able to query your machine for our
519. 1:19:40existing clients because we don't have that level of of intel that you do." So, that's a whole other strategy I hadn't
520. 1:19:49thought of, which is kind of cool. >> Yeah. >> I'll share with you how how we doing on time. We got a few minutes. I'll share
521. 1:19:54with you a really fun story. There's a um there's a company here in Denver um named named Craftsy and they they sell crafting stuff,
522. 1:20:08right? And and crafting tutorials, things like that. They started out as a as an education company and they were making all this
523. 1:20:16education for all these verticals. Here's business education and this and marketing and right just all this
524. 1:20:22education stuff. and one of the founders and and they've been going for a bunch of years and like all their all their
525. 1:20:31SKs, all of their products were basically seeing the sort of same sort of adoption levels and at one point one
526. 1:20:37of the founders wives was doing a a crafting thing and she was like, "Hey, can I do some like, you know, crafting
527. 1:20:44tutorials?" So they put up her crafting tutorials and like they like three weeks later they were like looking at the
528. 1:20:50stats and like all of the businesses were doing this and there was one line that that [laughter] up out of it and it
529. 1:20:57was the crafting thing and they basically discovered like completely unintentionally what their real market was and they
530. 1:21:05basically just flush the other stuff down the toilet. It's very possible that as you put stuff in the world, just like you're going to
531. 1:21:12price it a million times, the product might be something that you think is a really minor thing that all of a sudden
532. 1:21:18people like, wait, you can do that? We've been trying to do that for 30 years. And like this, I would say just
533. 1:21:24pay attention to those signals because if you have one that just is way above the norm, that might be your product.
534. 1:21:30>> From your lips to God's ears. Thank you, Kyle. >> I hope it happens, man. Well, I appreciate you coming up here. This is
535. 1:21:36this is really good talk. I appreciate it. >> I love the philosophy of of this show. So sometimes it's great to go face to face on you.
536. 1:21:43>> Yeah. Bye. Thanks. Where do I click to get out of here? [laughter] Bye. >> Yeah, I'll get you out of here. Um,
537. 1:21:53beautiful. Um, Chef Kelly for the win. That's cool. Congratulations on getting into that uh into that program. That
538. 1:22:00sounds pretty exciting. Um, that'll be a cool two weeks when you're out there pitching in front of people. And, uh,
539. 1:22:06here's [clears throat] the I'll give you one one more thing that I've learned over the years, and it'll be slightly
540. 1:22:13different when you're pitching because you're pitching, you know, to raise money. But in the money raising game,
541. 1:22:18here's what I've learned. If you want advice, ask for money. If you want money, ask for advice. >> [laughter]
542. 1:22:27>> Because the minute you ask for money, if you say to someone, "Hey, I'm trying to raise money. Let me tell you about my
543. 1:22:32business." They immediately go on the defensive and they they sort of shut down and they look and talk at you in a
544. 1:22:38very different way. If you say, "Hey, I've got this new thing. I've got some ideas. I would love to get your input.
545. 1:22:45You know, just I'm not raising money right now. I just I would love your advice." And if they get excited about
546. 1:22:50it, they will often say, "Well, when are you raising? How are you raising?" Oh, we'll talk about that later. It'll keep
547. 1:22:56them hungry. [laughter] So, all right. Um, I'm going to get out of here. Let you guys get out of here. I really appreciate you spending the
548. 1:23:04evening with me. Again, my request, go to the events and tomorrow is the back to basics course um from 7:30 to 9:30. Tomorrow, July
549. 1:23:1721st, Tuesday, in the AI salon. Um, please share it with anyone that you know that doesn't know anything about AI
550. 1:23:26or knows a little bit about it but wants to get better at it. Um, it really is just a a crash course on what it is, how
551. 1:23:33it works, why it's important, and then some really basic hands-on like I'll I'll spoiler alert. We're going to start
552. 1:23:41with chat GPT. We're just going to go play chat GPT and I might use I might do one other tool. Um, but it's just going to be chat GPT
553. 1:23:51because there's plenty to do in there. Um, more than an hour's worth. Um, so I think that's about it. The the
554. 1:23:59workshop on Saturday, um, I would go check out. Um, again, it's free. If you're trying to launch something, go do
555. 1:24:08that. Um, it's going to be super valuable. And then we've got all the other stuff that we do throughout the
556. 1:24:14week. So, that's going to be there. Thank you, Phyllis Pacific. Thanks, Danielle. Thank you, Drew, Mary Chester,
557. 1:24:20anyone who's here tonight. Charles, Kelly, Camp, uh, Chef Kelly, great having you up here. Producer Brandon,
558. 1:24:26thank you. Thank you to everyone who supports this and and, uh, makes it easy for me to do what I do. I really do
559. 1:24:32appreciate all of you. So, with that, I'll let you go. Have a fine evening. Peace.