9/11/2026 - From Corporate Chef to Tech Founder: Building and Funding an AI Startup with Chef Kelly

Transcript

1. 0:00[music] [music]
2. 0:20[music] Good evening, good people. Happy Friday night, date night. Welcome to the AI learning lab. We are inside the AI
3. 0:30salon. Hope everyone's doing well. Um I uh it it's 911 so uh 25th anniversary. So somber night uh somber day. Um I was
4. 0:45I was in New York City when that happened. We had offices about six blocks from ground zero. So uh it was pretty intense. So anyway, so um
5. 0:58many thanks to the um first responders for all you do, but uh especially those on that day. Um so yeah, just take a little moment. Uh
6. 1:14I got US Open tennis on. Um it's Friday night. It's Friday night date night. Uh tonight's when we talk about generous leadership
7. 1:23and what that looks like. Uh got some fun examples tonight. Uh generously lead is the the third leg of the cycle of AI
8. 1:32readiness. The first leg is play first. The second leg is create excellence. The third leg is generously lead. And if
9. 1:40you're wondering what that's all about, this is the cycle of AI readiness with you at the center is what we've discovered in the AI salon
10. 1:50makes for a really good recipe to reclaim your agency as a human being in the face of these AI tools getting smarter and
11. 2:01smarter and you know seemingly going out of control. Uh and the idea is to really understand who you are and what you
12. 2:07want. And play first is about learning. Learning through play. Um not making it such a burden to have to learn AI, but an
13. 2:19exploration where you get to learn AI. Create excellence is about taking what you learn and applying that, trying to
14. 2:24do good work for things that you're passionate about. And then generous leadership is about sharing what you
15. 2:30learn and putting yourself out there. So with that, welcome. Uh if you're new to the AI salon, I would encourage you. Hey Danielle. Hey
16. 2:39Val. Uh down below the live stream is the chat. You can chat and I can see that here. Um I'll be able to see a few
17. 2:47things on Tik Tok. However, I am producerless tonight. Producer Brandon's not here. Um so I think we might have
18. 2:55Chef Kelly joining us, but I'm not sure. um since I don't have a producer coordinating all that. So, if she shows
19. 3:03up back here, Silver Fox is in the house. Um we'll do that. What that likely means for Tik Tok is I might not
20. 3:10see as much over here because I don't have someone um uh looking for it. So, if you're if you're one of those people
21. 3:18doing both Tik Tok and the other thing um the uh the other thing, the AI salon, just let me know that there's a good Tik
22. 3:25Tok comment and I'll go look at it. Um, good. All right. If you're new to the AI salon, go check out all the activities
23. 3:32down the side. There's an events tab where you can figure out all the events that are coming up. They're really
24. 3:37amazing things coming up. Um, you can learn about the mastermind which is a subscription area of the salon which got
25. 3:44some remarkable programming in it, some really remarkable people in it. If you really want to level up your game with
26. 3:49AI, that's a that's a way to do it. It's a very powerful way to do it. Um, LDG's in the house. Is there a sound lag on the salon? Yes,
27. 4:01sound is good on my end. Um, you know, this was a little wonky. This was a little wonky. Um.
28. 4:17Oh, wait. Now it's lagging. I wonder if something's up. was a little wonky when I started it.
29. 4:34I'm going to leave the studio. Wait, here. I can do this. Let me try something. I'll be right back. I don't know that that's going to
30. 4:43stay there, but I'm going to try something.
31. 5:13Okay, I'm back. Is the sound better? Did that work? [laughter] Did that work? Now mine's good. On my iPhone, I have the sidebar with the
32. 5:23Okay, by the way, I'm producerless tonight. So, any any issues you have with chat or [ __ ] not working, you're on
33. 5:30your own. Okay, it's better now for me even before leaving the studio. Okay, I'm back. Uh, if Chef Kelly's here, feel
34. 5:38free to come backstage and I'll pop you up on screen. We can talk. Um, it's been a really eventful week uh in
35. 5:47the in the AI space. Um, I get the sense that it's better. Okay, good. Thumbs up. Great. Fantastic. Chef Kelly be there in just a few.
36. 5:58Awesome. Cool. So, we'll bring Chef Kelly up. So, so one of the things that that we do Friday night is generous
37. 6:04leadership. Um, so we'll we'll uh generally have a conversation with someone that is doing interesting things
38. 6:13in the in the world in the AI salon. And uh that should be good. Hang on. Checking out the score. 4-3 4-3 Shelton in the third set.
39. 6:25Not that anyone cares. I care. Um, so
40. 6:41so a dude quit anthropic this week and his his post on Twitter. He's got a he's got a a brand new Twitter account. Didn't
41. 6:57have many followers when he started. and his post on Twitter is now up to 6.6 million views. Uh it seems pretty clear
42. 7:06that that's all been amplified that this is all part of a coordinated effort to talk about how the AI is going to kill us. So [laughter] um
43. 7:19the the points he raises are not necessarily wrong, but I wouldn't take them at face value considering what's
44. 7:27happening. there's a lot of people doing sort of investigative work on who this guy is and what's happening and things like that. um
45. 7:37when you have literally trillions of dollars at stake um and the potential for tens of trillions of dollars um and
46. 7:48the power that goes with that um there's going to be a lot of games being played and I think what's clear what what
47. 7:53became really clear to me this week is the quality of AI has hit a tipping point where if we're not at AGI we're really close to it
48. 8:04and we're not going back. And so people in centers of power are sitting up and taking notice. They've they've been sort of
49. 8:17peripherilally playing around with noticing. Um but I think that I think that um it's now politically
50. 8:25advantageous to have a point of view on AI, good or bad. Um, AI has been kind of made the next uh the the next villain in a series of
51. 8:36villains, right? AI is the one that's taking your job away. You know, it was those other people and now it's AI and
52. 8:44six months from now it'll be another thing, but right now it's AI. Um, and the capabilities have gotten really quite remarkable.
53. 8:56One of the things that happened this week was um was uh anthropic and open AI both using models that have not been released to the
54. 9:11public yet have attacked um some of the hardest math problems known to man of the seven of them called
55. 9:26the Millennium Prize Problems. I I don't quite know exactly what the name is, but there's seven math problems and I think it's a
56. 9:35million dollars each if you solve them and some of them have been unsolved for nearly a century. Things like that.
57. 9:43Really hard. Really hard. It appears that Anthropic solved one of them this week and Open AI has solved three of
58. 9:51them. They haven't announced them all yet, but that's the rumor. So, four of the seven have been solved this week
59. 9:59using AI. Now, using a fancier, newer version of AI that we don't know about yet, that we can't we can't play with
60. 10:07yet. And they've spent a boatload of money solving these problems, but it looks like they're being solved. to which the math community
61. 10:23has basically said we don't like this. [laughter] We want to go back to how we used to solve problems where we were in an ivory
62. 10:33tower and we took a long time to do these things and we used our brains. So the people that gave us the math, that gave us AI
63. 10:43that's powerful enough to get good enough to solve really hard math problems is now pissing off the mathematicians. [laughter]
64. 10:53We talk in here about a lot of negative news this week. We talk Hey, hey, Cam Katkin. What's happening? LDG's in the
65. 11:01house. Todd, you know, we talk a lot in here about um the great repurpose and this idea of having your identity tightly coupled
66. 11:11with your job. Um I don't think anyone is immune from the fact that things are changing. How we do academic math is changing. And
67. 11:27what some of the top mathematicians in the world did was 25 of them penned a letter. Terrence Tao was I think the writer or
68. 11:36the architect of it and a thousand mathematicians have signed on and they basically said AI bad the way we did it before good.
69. 11:47Okay. [laughter] And it's here and it solved some number of these major problems. Um, and you know, their argument is is
70. 11:59something in the neighborhood of we learn more when we think, which I don't disagree with. And if we solve these
71. 12:07math problems, things progress in the world. Like there's fluid dynamics things and things that there's just all
72. 12:15sorts of, you know, downstream effects that can that can come from solving these problems. A good way to think about this is let's say
73. 12:29that instead of aiming these fancy new models at math problems, next week Open AI aims them at at curing cancer
74. 12:42and they figure out how to cure cancer. The response from these mathematicians is equivalent to the medical community
75. 12:50going, "Hold on there. It's our job to cure cancer, not some fancy machine. We wouldn't stand for that. It would be [ __ ] ridiculous. So,
76. 13:03that's kind of what's happening. [laughter] Why doesn't Google use its uh quantum computer for this? Well, the the the
77. 13:11math that quantum computers do is different than the math that um large language models require as far as I
78. 13:19understand it. I ain't a mathematician. Uh but I met one staying at a Holiday in Express. [laughter] Um
79. 13:32the math is different now. Google about a year and a half ago um came up with some software that allowed
80. 13:42quantum computers to be used to do AI math. I I don't know if it was inference or training, but I think Google has the
81. 13:52technology to be able to do that. I I I don't know the answer. The answer probably has to do with um with errors and [ __ ] like that. I I
82. 14:03don't know. I mean, quantum computers are not quite ready for prime time yet as far as I understand it. Brandon, I'm
83. 14:08building my soccer player agents right now on that aentic tournament that you shared in the salon. Oh, that's super
84. 14:15cool, Todd. Yeah, Brandon's not here tonight. Brandon is out. Uh, so good wishes to him. He's off doing some
85. 14:21holiday thing or other. If you're on Tik Tok and you're wondering what the hell is going on and why is he not looking at
86. 14:26the camera, the Tik Tok view, this is the backstage view, um, what I'm looking at on screen right now, uh, is at the AI
87. 14:34salon. So, if you want to join the live stream, go to my profile and click on the the hyperlink in there and that'll
88. 14:40take you over to the AI salon and you can watch the live stream. Um, I will pay attention here for Chef Kelly. Chef
89. 14:47Kelly's going to pop over and and hop in here and we'll have a chat with her in a bit. Um, I think what I want to do
90. 14:54though, and if anyone has any questions, it's Friday night. You can ask me any questions. Um, we're going to talk about
91. 15:00leadership. I mean, I I feel that the what the what the mathematicians are doing is kind of griping about things changing in the
92. 15:15guise of leadership. They're sort of saying, well, there's not alignment here with what the real power of mathematics
93. 15:22is. It's about human beings using our minds and interacting. And there's part of that that's true, but there's also part of that that's like
94. 15:33mathematicians were sort of at the highest rung of the intellectual, you know, the intellectual, you know,
95. 15:39quantum physicist, mathematicians, they're all sort of at the at the top of these ivory towers. And like they're the
96. 15:45only ones with the expertise. and a technology that is the result of their work up in these ivory towers is now toppling those towers
97. 15:56because the machines now have the capacity to do that level work. They don't like it. And so they're trying to
98. 16:07basically say here let me let me share something with you. Um I'm going to put let me share my screen.
99. 16:21Okay, that goes over there. Beautiful, beautiful, beautiful, beautiful. It's fantastic. Bob, hey Bob,
100. 16:36how you doing? We're doing pretty good there. Got Danielle in the house. Let's see. All right. Oh, these fruit flies are amazing. All
101. 16:49the all the fruitly stuff that's happening. Let me see if I can find the Let's look up calculator. Calculator. This is not the first time.
102. 17:05Where's the famous picture? This is not the first time that mathematicians have protested technology. Let's see. Calculator protest photo.
103. 17:22That's not a real photo. That's an AI photo. There's a real photo here somewhere. Calculator test.
104. 17:43this one. There you go. [laughter] Math teachers protest against calculator use. Turn them off until upper grades.
105. 17:57We don't We don't need those stinking machines that add faster than we do. This was in the 70s, I think. When does it say the date? 1965
106. 18:08was the year I was born. Mathematicians were protesting calculators. That's where it's where we've come to people. [laughter]
107. 18:18So now the mathematicians are protesting a highly powerful math machine. [laughter] Oh my god. So the other thing that I'm
108. 18:30sh showing here that's kind of all over So one of the things that's all over my timeline is um is the uh the
109. 18:39mathematician stuff and then the other thing is um fruit flies. And so let's let's look up fruitfly. In fact, let's look up fruitfly
110. 18:55Scoville. Robert Scoville cuz he did a whole post on this. Here we go. Um, the fly brain report. You've
111. 19:04seen the flies all over your feed. Um, here's a good list. Okay. So, what happened is scientists ha have
112. 19:13mapped they've created a map of 166,600 or and 700 neurons of a fruity's brain. Let's see what we got here. More doctors smoke camels. The
113. 19:29downstream effects of solving those complex problems is so important. I know. I know. And there there Yeah, it's
114. 19:36it's crazy. Anyway, so there is now a an accurate 166,000 neuron fruitfly brain simulator that is out on the internet that you can
115. 19:52you can use and you can train that brain to do stuff. And so people have been doing that and it's kind of crazy. So,
116. 20:00by the way, if you're on TikTok and you want to see the screen I'm looking at. I'm going to show some examples of these
117. 20:04fruitfly brain demos. Um, just go to my profile, click on the link there. It'll bring you into the AI salon. And if
118. 20:11you're new to the AI salon, welcome. Um, go check out all the the stuff going on. There's an events tab. There's a tab
119. 20:17called introduce yourself. If you're new here, please do that. Please introduce yourself. Um, but let's look at this. Um, okay.
120. 20:26Um, a fly brain plays Beat Saber. [laughter] [music]
121. 20:47[music] The one thing I don't know is how they're training these to do these and how they're mapping them. I assume
122. 20:59they're making a 3D model. I think Google today or yesterday released an open-source 3D model, like a rigged
123. 21:08model of a fruitfly body. So, you can take this simulated brain, train it to do [ __ ] and then put it in an actual
124. 21:15fruitfly simulated body and have it interact in the world. So, there there's a bunch of these things. Okay, so that's
125. 21:21that's that one. Let's see. Um, is this the fruitfly brain? Oh, here we go. Okay. Uh, a fly brain drives a Mini Cooper and
126. 21:37uses the blinkers. [laughter] Okay. So, you see the the little steering wheel on the bottom and then the brain like you see the neurons
127. 21:49firing that are having this thing do the driving. Um, I don't There's a fruitly doing the driving. [laughter]
128. 22:05Let's see. What's the guy say at the bottom that this is uh Let's see. Um, Guar Gupta, a fruitfly brain spent this
129. 22:13week playing Doom, trading Bitcoin, and driving a hexopod robot. None of this is fake. Um, so, so Cam, you know, you doing
130. 22:22your, uh, uh, your work on on, uh, on beings and things like that. Here, here's another one right here. You've
131. 22:29got a, you know, a simulation of a of a fruitfly brain learning and behaving. Let's see. Can you ask Astra to do
132. 22:39something with the fruitfly brain? It is an API. Um, yeah, we can do that. I could try that. Let me see if I've got any um
133. 22:48let me see if I've got any tokens left. Um hang on a sec. Let me go to What do we want to have it do?
134. 23:06I saw someone had a fruitfly do a uh a Rubik's cube. All right, hang on. Let me let me connect.
135. 23:22All right, there's that. How's my usage? Usage remaining. Okay, it's not bad. So,
136. 23:37so let me flip over to codeex and let me go to go to Astra.
137. 23:52[ __ ] it. I'll do it on Astra extra high. I've got an extra. We'll see. We'll see how far we can get. Okay. So, here's
138. 23:58what I'm going to do. I'm going to say um there's a new fruit fly brain simulator
139. 24:12with an API and I think also an open source fruit
140. 24:27fly body that is rig body. I'm going to put from Google. I think it's from Google. From Google question mark.
141. 24:47Okay. Um, I want you to find them and then
142. 25:03figure let's say figure out how people are training them and I want to train ours too. Uh, okay.
143. 25:19Cam, cat can or anyone make a jingle for popsicles. Mosquito. It's misk. It's a mosquito. No, it's a fruitly. It's a fruit fly.
144. 25:33Um, we want to train ours to play Pac-Man. No. Then he'd have to make Pac-Man, too. Well, that's okay.
145. 25:47What? Let's see. What What do we What do we want to train it to do? Um Oh, train ours to stack [snorts] cups. um
146. 26:05and make it easy for me to learn and demon with. Okay,
147. 26:21one of the things I do here on the AI learning lab, by the way, if you're on Tik Tok, welcome to the AI learning lab.
148. 26:25This is the backstage view. If you want to see what I'm working on, I'm going to have Astra train a fruitfly brain how to how to um
149. 26:37stack cups. [laughter] If you want to come see that, click on the uh on my profile and then click on
150. 26:46the link over to the AI salon and you'll be able to see the live stream. Okay, here we're gonna we're gonna go bang.
151. 26:53So, one of the things I do here is I do stupid prompts so you don't have to. If you're looking at that prompt and you're
152. 26:59like, "Kyle, that's one of the shittiest prompts I've ever I've ever seen." I don't disagree with you. Okay,
153. 27:06this is going to think and go do its thing. Let's see. I'll track down the brain simulator and rig fly. Check how
154. 27:12they're being trained. Work out a p practical cup stacking demo you can learn from and show. Okay. All right. Agents.md. It's doing
155. 27:23something. It's working. It's working. I finally saw that Charleston AI video. Crazy, right? Isn't that funny? Okay,
156. 27:32let's go look at some other things they're having to to do. I turned to fly bisexual blocked the mal output in 166,000 neuron model
157. 27:43and and it went over the the uh [laughter] it's at a disco. It's at a disco [laughter] and it likes the dude. That's awesome.
158. 27:59I [ __ ] love nerds. Okay. [laughter] Um, okay. Let's see. Fly a fly brain running inside Minecraft. For clarification, I'm not simulating a
159. 28:14fly's consciousness or uploading a living brain into the game. This demo uses a reconstructed map of a fruit
160. 28:22flies neural connections with simplified rules on how activity moves through the network. Minecraft events feed into the model and
161. 28:31activity is selected in selected neuron groups helps choose and adjust movements I program like flying, grooming, feeding or escaping.
162. 28:43The wiring comes from biological data, but the activity model, game inputs, and movement mappings involve engineering
163. 28:51choices. It's an interactive way to explore a connectto dome, not evidence of consciousness. Okay, so that makes kind
164. 28:59of sense, right? So, they took these 166,000 neurons, they wired them like a fruitflyy's brain is, and then human
165. 29:09beings are just basically using that neural network to do [ __ ] That makes sense to me. I'm backstage when ready.
166. 29:17Okay, beautiful. Beautiful. Chef Kelly, let's um let me go look at what is going on over here. The Google connection
167. 29:25checked out. Flybody is an open- source jointed mujo co-fly from Google DeepMind. I also found Eon's brain simulator. It's public code and
168. 29:36currently exposes neural simulation tools rather than a documented train this fly web API. Okay, so it's it's off
169. 29:45doing its thing. Let's let me go look at my usage here. 90%. All right. So, it's not blowing through it too fast. So, let me pull
170. 29:54this over here. We will come back. Chef Kelly, are you ready to pop up on stage? Hello. >> Hello. Can you hear me? All right.
171. 30:09>> I can hear you. Good. Good to see you, Chef Kelly. >> It's good to see you, too. >> Welcome to Friday night date night.
172. 30:17Great night. Date night. Where are those toaster waffles in [laughter] hot pockets? >> Hot chos. Hot pockets. Yeah, exactly.
173. 30:26Yeah. You were from the commercial commercial food industry. You know what's in a hot pocket? >> Hot pockets. Yeah. [laughter]
174. 30:32>> Um, so welcome. Um, it's really good to have you here. Um, Friday nights, um, we're we're we're we're starting to use
175. 30:41the AI learning lab to really live into the cycle of AI readiness. Mondays play first and Wednesdays create excellence
176. 30:47and Fridays generously lead. So I appreciate you um being a guest and having a conversation. So um I think a
177. 30:54lot of people here know who you are, but some probably don't. So would you do do us a kind favor and introduce yourself?
178. 31:00Tell us a little bit about who you are and what you have going on. >> Sure. Uh hi everybody. My name is Chef
179. 31:05Kelly. Um I'm not your average chef. It's not like the bear, you know. I don't own a restaurant. I don't have my
180. 31:11culinary tweezers, you know. Yeah. Um, I yeah, [laughter] I'm a I'm a corporate chef, so I work behind the scenes at a lot of famous
181. 31:19food brands that you've heard of, like Nestle and Impossible, Disney. Disney has this huge food innovation program.
182. 31:25Um, and I've been doing that for about almost 20 years now. Oh my god. And um, just very briefly, I I left my cushy job
183. 31:34in research and development for Nestle in 2024 because I was very curious about what AI would do for the food industry. And
184. 31:42>> oh, you left you you you left it for to explore AI. >> I did. Well, at the time Nestle was working on something in AI um and I kept
185. 31:51asking to be put on the project and they were like no absolutely not [laughter] >> really. >> And yeah, well it it turned out to be
186. 31:57for marketing. I mean they've already released it so it wasn't even in my division. But um I had that opportunity
187. 32:02to go to the MIT summer program um to get certified in AI and machine learning for data science. And so I did that and
188. 32:10I had to leave. And I everybody thought I was crazy. I probably am, but um you know, and long story short, last year
189. 32:19when the FDA said that they were going to require all food brands to remove artificial dyes, um I guess my training
190. 32:27kicked in and I just started working on a project that is now a full-fledged almost funded um >> congrats. >> B2B SAS. Yeah.
191. 32:36>> Yeah. [laughter] a B2B SAS for the food industry for the industry you left. >> Yeah. >> And have you talked to any of your ex ex
192. 32:43employees and shown them what you're working on? >> I haven't. I've talked to one who didn't ask me. Um but there a lot a lot of
193. 32:52Nestle folk are are like sneaking around my profile on LinkedIn. Like I see them all the time. >> You're being you're being stalked.
194. 32:59>> It's like a game of chicken. I'm like I'm not going to say anything first. You got to come to me. [laughter]
195. 33:04>> Wow. That's amazing. Um, so, so talk to me about you, you've been you've been really active in the AI salon. You've
196. 33:14been been active here. You ask really good questions. You've been at least from the outside deeply curious
197. 33:21about this stuff. Can you just talk about your journey from, you know, it sounds like you took an MIT course,
198. 33:27which I assume was more about, you know, how weights and embeddings and things like that work, which is not necessarily um what
199. 33:34>> it's not sexy. Yeah. [laughter] Yeah, it's not necessarily what you use to to prompt these things or use agents, but right
200. 33:41>> I assume understanding it's valuable. Can you just talk talk to us about what's been your journey in from you
201. 33:48know where you were and you know >> the program was really hard. I'm not going to lie. Um as dedicated as I was it it was really pushing my
202. 33:57self-sabotage buttons because it was so difficult to to like you know halfway through you're like I'm lost. I'm just
203. 34:03so lost. you know, good thing is you don't have to write the algorithms. You just got to sort of understand kind of
204. 34:08what they do, you know, like a decision model or, you know, so on and so forth. >> And um I felt like a total impostor once I got
205. 34:19my certificate, which was really I thought it was going to change the world. I was going to be on top, you
206. 34:24know, and I just felt like it was really hard and I was very proud of myself, but I was like, who am I to who do I think I
207. 34:31am to walk into the world with this new found skill? And yeah, >> what did you what did you learn? Did you
208. 34:38did were you learning generative AI or were you learning like the the underpinnings of how the models models work?
209. 34:44>> We were learning under the hood mostly like so we didn't learn prompt engineering. I mean, very briefly at the
210. 34:49end, you know, one shot and all that stuff, >> but it was really about the history of AI, how it came to be and and and what
211. 34:56it really does under the hood. Um, there wasn't a lot of talk about how it could destroy mankind or anything like that,
212. 35:03but it was really just about >> Yeah, back then it was powerful enough to threaten. >> Yeah, exactly. I mean we were still have
213. 35:10like nine fingers on one hand at that point you know and so I uh you know it was really interesting we did get into
214. 35:18data science a lot so analytics and you know aggregating cleaning parsing you know running the analytics against each
215. 35:24other trying to find the sort of hidden you know using outliers not using outliers all that stuff that was really
216. 35:29interesting to me and um then I was you know let loose in the world and I didn't know what to do with it and and truth be
217. 35:38told I found that I found that everything I learned was really valuable, but I was either too late or too early is what I felt like because
218. 35:45everybody else was learning prompt engineering and image generation and video generation and all that stuff.
219. 35:51>> And I was like, I don't I don't know what I'm going to do. So, um it was a weird transition to sort of walk back
220. 35:59into your industry thinking, oh, maybe I'll audit, you know, um businesses and and help get them AI ready or help them
221. 36:06implement things. and that didn't feel right. And so it took me it took me a good six months to figure out what am I going to do from here?
222. 36:16>> And and and that catalyst with the with the food dies was the um >> was my curiosity. I am relentlessly curious to a fault
223. 36:25>> and I was just very very curious about how like what color crops were they using and how were they grown and how
224. 36:31did they make them into dyes and that curiosity is what sort of sparked the the whole impetus for for die converter.
225. 36:39>> Yeah. And and when So you get the degree from MIT >> certificate >> or certificate? [laughter] >> I I'm not going to give myself more
226. 36:50credit than I have, sir. >> I went to MIT once. >> Yeah, he walked around. [laughter] Um, so you have that thing the gap between
227. 37:04understanding how large language models work or how AI you know works and getting started like where did you start
228. 37:11when you when you started playing with generative AI? Did you start with images? Did you start with catch EBT?
229. 37:16Like what was your progress after I got the degree? I loved research. Like I think um last uh festivist I did a little research um
230. 37:29showand tell for you guys because for me and this is nonsecuador but before I even became a chef I was a journalist. I
231. 37:37worked in in as an editor, West Coast editor for for Condast and I also was in PR for a while before social media and
232. 37:46so I have that that curiosity lended itself to journalism really well and research um was a pain in the ass and
233. 37:54citing your sources and getting those those those sources together in micro fish. You guys remember micro fish? Yeah.
234. 38:01>> And so for me coming out of that program and and you of course Chachi BT was all the rage. Claude was the writer. Chachi
235. 38:10PT had access to the web. That's kind of where I I was at the time. And >> and I could really now research what was
236. 38:18truly kind of going on in the food industry. I could pull on that thread and what would have normally taken me
237. 38:24weeks or months, you know, maybe you would needed a certain uh subscription to get into, you know, court documents
238. 38:30or, you know, you would you would it would have taken you months to dig through these file folders within, you
239. 38:35know, an old database could now be sourced lightning fast. And I could pull on that thread and find out things like
240. 38:43down to who voted for what bill in what state that passed in Congress that led to where we are now. Like all that stuff [laughter]
241. 38:52>> like SUPER SUPER QUICKLY, RIGHT? >> Super super refined. >> And I love that. So it was that and then also on the creative side, I've said
242. 39:00this before, I can't paint or draw or take a good photograph to save my life. and couple that with, you know, not even
243. 39:09knowing how to do any graphic design. >> And so playing around with the image generators and and and Canva and and and
244. 39:17um Gamma and all those things was a lot of fun for me. It's still kind of a creative outlet for me when I'm feeling
245. 39:23>> overwhelmed with stuff. Yeah. >> Yeah. Exactly. Um just if you're on TikTok, you want to see the person I'm
246. 39:29talking to, Chef Kelly, she's here. She's awesome. Um, just go to my profile, >> click on the link and uh come over to
247. 39:36the AI salon and you can see the live stream. Um, okay. That's really cool. So, what's been I I have two two
248. 39:44questions. Let me start with this one. Um, what's been the most surprising like thing you can do now that you never
249. 39:55thought you'd be good at? Like, is it the art stuff? like what's what's the thing that like you feel is the most like a superpower?
250. 40:06>> I think that um the whole die converter thing aside that's well let's put them into pillars. Yeah. the the fact that
251. 40:18whatever branding I stumbled upon that I use for all of my social media posts and all of my pitch decks and and everything
252. 40:27for die converter has really turned out to be great and I get a lot of compliments on it and I've been asked if
253. 40:33I worked in advertising. I'm talking to Kyle Shaden for crying out loud, right? [laughter] And people people at the cohort I was
254. 40:41just in were like, "You should have a workshop and show people how to make your pitch deck." And I was like [laughter]
255. 40:46like never would anybody say that my creative uh output was was worthy [laughter] of anything. >> And is it is it dieonverter.com?
256. 40:58>> Yeah. Dieconverter.com. >> All one word dieonverter.com. >> Yeah. So hang on. Let me We'll look at that.
257. 41:08Yeah. And if you go I think there's gallery or posts or something and and you can see but it's really evolved even
258. 41:14since what I still have on that um what I still have on that page. My my LinkedIn is probably the best place to
259. 41:22go find or Instagram too I think. >> Yeah. Anyway, >> but it's a very Appleesque kind of design. I mean, it's very lots of white
260. 41:29background, a product shot, just little handdrawn things that maybe ask a question or pose a thought, and people like it. >> Yeah.
261. 41:38>> And I and I cut you off. So So what what is the thing that that feels like the big the big level up for you?
262. 41:44>> I think that's that's what it was is is just stumbling on the ability to to make an an ad campaign or branding. >> Oh, I see. Yeah. Yeah.
263. 41:52>> Yeah. About, you know, around the product that I created that is actually Like it looks good. It doesn't look like
264. 42:00I piece milled it together. It >> right. >> Doesn't look like a bad AI. >> What's your What's your What's your
265. 42:08secret to it not sucking because it's really easy to make shitty content. >> It's very easy to make [ __ ]
266. 42:17>> Right. It's so easy. It's so quick. [laughter] >> Yeah. >> Yeah. I see it all the time. What's your process for like like what's
267. 42:25your filter or what's your process for not that? >> Um, don't overthink it too much. You know, one thing that I did a lot in the very
268. 42:36beginning was overedit >> or if I was trying to create a a post that looked welldesigned. I mean, I could ruin a good Canva template in 30
269. 42:45seconds flat, you know, [laughter] trying to think I know what font to use, right? keep it simple, you know, um,
270. 42:54look at the classics, you know, like like Apple or, you know, Chanel and and just see how very simply they they get
271. 43:02their point across. And now I mean when I think it was Chat GPT when they when they moved away from Dolly and started
272. 43:12calling it image 1.0, 1.5, 2.0, >> Yeah. >> it started doing product shots really really well. >> Yeah. Yeah.
273. 43:19>> And so whether I wanted something that was out of the package or in a generic or even I can I can ask for a Nestle
274. 43:25Crunch Bar and it will make me a picture of a Nestle Crunch Bar. It's wild. That's when it started to get really good. And
275. 43:33>> once upon a time Oh, and here's another thing. It used to be my Achilles heel, but I somehow turned it into sort of
276. 43:45superpower, I guess, is that I realized that I needed to tell Chat GPT whether it's the image generator or not.
277. 43:53I don't like telling you what I want. I like meaning I'm not going to tell you the camera angle, the lens to use, like
278. 44:00I'm not going to use all these fancy designer lingo. I'm going to say, "Here's an example. Here's kind of what
279. 44:07I'm looking for. Can you give me three options and go from there?" Because I'm so visual like that. I don't like to
280. 44:15spend an hour on a prompt. I don't like to engineer it. I just like to have a conversation about it. And so
281. 44:21>> if I think it kind of got accustomed to that with me and knew that it could be a little creative, it could push the
282. 44:27envelope one way or the other and then we would refine it together. And I don't know if it's because it learned it
283. 44:32through me or just because we happen to find a good solid branding model that we like it. It works. So
284. 44:40>> well and I assume as you as you navigate you get three options. You pick one. You get three more. You pick one. You get
285. 44:45three more. You pick one. It's kind of it's kind of learning, you know, your your aesthetic or whatever. Whatever
286. 44:51it's prompting is. Yeah. Your pattern. >> It's it's learning from that. That's really cool. Um, so one of the things I
287. 44:59talked about this a little bit on Wednesday, but this is this is kind of my new my new thing. Well, I don't I
288. 45:05don't know if it's a new thing. It's a it's a new awareness for me, which is building something with AI and building
289. 45:14it well is very very fulfilling and that's very different from launching something into the world. you know,
290. 45:21whether you built it with AI or you built it with your hands or right, you know, putting something in the world is
291. 45:26just a fundamentally different thing than building a thing. So, I I remember when you when you first showed me like
292. 45:33your first version of die converter, you're like, "Hey, I think I've got this thing here." And you showed it to me and
293. 45:37it and like was like, "How did you do this?" You're like, "I just kind of, you know, was it [laughter] I think you
294. 45:42started in lovable, right? >> It coded it and here it is." And so I I would love to understand from you the
295. 45:50difference between those two things. The difference between that initial build and sort of getting it to like okay I've built a thing
296. 45:57and how is that different than launching a thing? >> Um I'm going to use a word that people any entrepreneur or any tech founder out
297. 46:06there knows and has heard and probably hates the word grit. Like and they they're not lying when they say you've
298. 46:14got to have grit. it. You've got to be relentless with yourself. Um, you've got to eat the elephant one bite at a time.
299. 46:23>> And a lot of people will say, you know, stop stop learning about it. Stop looking at YouTube courses and stop stop
300. 46:30reading up about it. Stop joining communities about it and just do it. Sure. But if you want to do it right, you're going to stay with your
301. 46:37communities. You're going to keep watching the YouTube tutorials. you're going to keep bettering yourself to
302. 46:41understand, you know, how to build what you need to build the right way. And maybe that's uh, you know, one of my imperfections is is is it
303. 46:54wanting to be right, >> not just wanting to be the most viable product, but a viable product that can actually be work and be sold. And and
304. 47:04that takes continual learning and tweaking. And it takes a lot of what I don't usually have, which is patience.
305. 47:10Pardon my French, but these [ __ ] machines sometimes can mess up your [ __ ] so fast. >> Oh. [laughter] Oh, it's brilliant.
306. 47:18>> I'm in the middle of it right now. All of my automations with codecs are broken since Astra and I cannot get it to fix
307. 47:24them. It keeps like it I went in today. I said, "Why did my stuff not run?" It's like, "Oh, because I disabled your cron
308. 47:30jobs." I'm like, "Why did YOU DISABLE MY CONT TO DO THAT?" >> YEAH. [laughter] IT SAID IT said you're
309. 47:34right to be pissed. I'm like, "I know I am, you loser." [laughter] >> I love it when they're sick of fantic
310. 47:40after they've done something wrong. >> Yeah, exactly. Oh, you're right to call me out on [laughter] that. >> Yeah, exactly. >> Um, so
311. 47:50I So, you've been in market now. Are you having conversations with customers? like has the product evolved because of
312. 47:57conversations with people looking at it that are potential buyers? Like how's how's that journey been?
313. 48:03>> So the journey for me was kind of odd. Um and I and I will say this, not getting beta testers earlier on was
314. 48:12probably one of my biggest uh obstacles. Like I didn't I think I'd be further along if I had if I had launched it a
315. 48:21little sooner. But um and listening to customer feedback and I maybe I was still too attached to it at the time
316. 48:29because it had been my baby for almost a year, you know, and um >> you you knew what it was and why everything worked and Right. And then
317. 48:37and then other people >> and if I if I had shown it to somebody Yeah. Some were like, "We don't get it." I'd be I'd be crushed. Right.
318. 48:43>> Yeah. Yeah. But [laughter] but I I I finally so I started getting a lot of phone calls from some big names in in the food industry,
319. 48:53>> but they were again they were sniffing around. They weren't real sure. I was new on the scene. I mean they some of
320. 48:58them knew who I was and my LinkedIn you know my my resume speaks for itself but >> right >> who who who is this tech startup you know
321. 49:06>> and I really had to do the dog and pony show. I went to five trade shows, summits in five months.
322. 49:14>> I joined that summer cohort because I was like, I don't think I can really launch this the way I want to launch it.
323. 49:20I'm a solo founder and I wanted to keep it that way, right? >> Yeah. >> But I am only one person and you hear so
324. 49:26much noise coming at you about what you should do, you should do this. I mean, even by the experts, one expert will say
325. 49:32one thing and another will say a different thing. >> Absolutely. And so I said, I'm going to go to this cohort and devote my entire
326. 49:38summer to go to go to market and one voice, one curriculum, just stick with it and do it. And I think that was the
327. 49:46best decision I ever made. And I was really unsure about it. I was like, you know, it's kind of an expensive program
328. 49:52and why why isn't like Y Combinator and they're paying us, you know, but you but the payoff was that it was hardcore and
329. 50:01it got me ready and it got me in front of investors and it >> So, this was this was an accelerator.
330. 50:07>> It was an accelerator over the summer. Yeah. >> Over the summer. How long was it? >> Two months. It was from July 1st and
331. 50:13we're just now wrapping it up. Yeah. >> Wow. And was it in person? It was the first uh six weeks were in a uh online
332. 50:22and then we did the last two weeks up in Silicon Valley in person. >> Nice. And >> and then there were two summits at the
333. 50:28end that we went to. >> Yeah. Describe what were the uh like like what did you learn? what were the outputs like what are you building
334. 50:33toward or you know >> so this was through the founder institute >> and um Adeo Rei is the is the the main
335. 50:41founder and it's been around for since I think 2009 >> and um he had gone to he kind of left and gone to create this founder uh
336. 50:52funding group on the side and then came back this summer after 10 years just to teach this cohort and he he was intense [laughter]
337. 51:00>> and we had what were called deliverables every week by Thursday morning at 9:00 a.m. when our live sessions started. And
338. 51:07these were they were a lot a lot of work. Now, they had internal AIS that you could use to help you get through
339. 51:13them. But I mean, he said at the beginning of the summer, we chose 50 of you and 40% or more will not make it to
340. 51:22the end. And he was absolutely right. 23 of us ended up being there at the end >> out of 50. >> Yeah. Yeah. And not because there were
341. 51:29bad ideas or that people weren't cut out to do it. It's just it's takes a lot of your time away from everything else. And
342. 51:37even when you're a solo founder, I had to pull back on customer acquisition if I was going to focus really hardcore on
343. 51:44getting to market, right? And so I did feel like I I missed a little time there where I could have been building more
344. 51:49traction, but um at the end of the day, the payoff was well well worth it. and and I needed that um I needed that
345. 52:00>> sort of, you know, I needed deadlines. I needed homework. I needed to to know that I I had to be responsible to
346. 52:06somebody else because when you're an entrepreneur, it's very easy to push it off till tomorrow, you know, >> or or to do stuff that's like,
347. 52:14>> you know, exciting, right? [laughter] >> I'll get I'll get to sorting the list later. >> Yeah. Yeah. >> Yeah.
348. 52:22>> Let me make some more pretty branding. [laughter] >> Yeah. Yeah. Exactly. Um, and what were the what were the nature of the deliverables?
349. 52:30>> It was a lot about validating your idea at first. Um, and then it was a lot of getting out and talking to customers and
350. 52:37that was so scary to me. Yeah. Um, and then there were also things that we learned how to do which was to build
351. 52:44synthetic focus groups and so that you could if you weren't like for me I was very unsure about my pricing. I'd been
352. 52:50struggling with that for a long time. And so we created all my different ICPs and and this is stuff that people know
353. 52:57if they're doing it, but like but when you're forced to sit down and really create your three different ICPs and who
354. 53:03who's going to be your champion, who's going to write check? >> What's ICB stand for? >> Uh what is ICP? It's it's customer profile.
355. 53:14>> What's the I stand for? >> I don't know. >> Oh my god. It's something customer profile. your intimate customer profile
356. 53:20or something like that. >> Hang on. [laughter] I I I'm an expert at this. I have chat. >> I know what I know what an MCP is, but
357. 53:27all I know is the word protocol. [laughter] >> Something >> internal. It's >> thinking. It's thinking or no, it's not. Come on. Chat GPT.
358. 53:42Uh try again. chat GPT not >> ideal customer protocol >> ideal okay ideal customer profile okay got it >> so doing ICPS
359. 53:54>> so for me like the the the people who will champion me who will like my posts who will engage with my posts who might
360. 54:01want to schedule a call with me are the R&D teams and the food scientists but they're not the ones who can approve the
361. 54:08purchase they're not the ones who will write the check so I have to make sure that I'm also appealing to the the
362. 54:16people that do. And not only does my content need to reach them, but I need to know how they think and what language
363. 54:21they're using and why this my my [clears throat] prof or my platform would be valuable to them. And so that
364. 54:28was really great through these synthetic focus groups being able to create the different personalities and then query
365. 54:36them on what would you pay for this? Do you have anything like this? You know, what what would it do for your revenue?
366. 54:42what would it do for your for your reputation or your your customer perception? And that was really um
367. 54:49amazing. And then we had to this is the big one. We had to file all of our incorporation paperwork, create our cap tables,
368. 54:58>> all of that stuff. And >> oh wow. >> So you have your [ __ ] [laughter] together in [clears throat] case someone
369. 55:03wants to invest in case, right? Because because they investors will not give you the time of day if you're going to not
370. 55:11not be available immediately if you can't take that check right away. Yeah. >> Yeah. >> Right. Wow. Fascinating. And now you're
371. 55:19at the end of the the program. Are there pitches? Like what's it look like? What's the what what what's the
372. 55:24opportunity um at the end of it? Is it just that you have your [ __ ] together now and you know what you have? Like what's it look like?
373. 55:31So first let me just tell you the story of how I almost completely sabotaged myself. Okay. So first of all everybody
374. 55:39from across the globe these last maybe 30 people that whittleled down to 23 >> were coming into Silicon Valley right
375. 55:47and and theo and his wife Mercedes Bankston were creating all these parties and all these networking events for us.
376. 55:55And then we had to you know we had to get our final pitch deck together. We had to sort of audition our pitch deck
377. 56:01in front of all the mentors also that came in. I mean it >> and I and then also we had the opportunity to Airbnb it with other
378. 56:10founders who were also coming in town. >> Nice. Nice. Yeah, it was cool until about day five when you're sharing a
379. 56:17room and a bed with [laughter] another founder from France you've never met before and you're not getting any sleep
380. 56:25and everyone's up at all and you've got your dog that you brought with you [laughter] from Cal Southern California
381. 56:31>> and it was just like it was wild and fun but it really wore me out faster than than than I had expected. And the we the
382. 56:41night before we were going to do >> our it was called demo day and it was leading up to what was called founder X
383. 56:48which was a a summit that had um a lot of investors coming in and demo day was the day that you could get up and do
384. 56:54your pitch and roll your pitch deck in front of a room full of investors and I'm not afraid of you know getting for
385. 57:00the camera getting up on stage right not me >> and the night before and I I had been in the top five of the cohort for the last
386. 57:09three or four weeks leading up to it. >> I got up on that stage and I choked. I >> no >> I had been sitting there I was so tired
387. 57:18by this point. I'd been sitting there at my table vigorously writing on index cards because I could not remember my
388. 57:24speech and we were not allowed to use notes. We were not allowed to use like uh our phones even. It had to have been
389. 57:30memorized. And I was just and I just worked myself up so bad that I got up there and I got through the first two
390. 57:38slides and I was like, "All right, you got this. You're gonna be okay." And then I just I clicked and I was like
391. 57:44>> You didn't know what to say? >> Uh yeah. Yeah. And then it was and then I just crumbled a big gold geek up there and I'm sorry.
392. 57:56And I could not I mean like Mercedes her face just fell cuz she had such like high hopes for me >> and uh I said um oh then so it was the
393. 58:08end of the night we were all called back in and I was not even in the top 10 anymore. >> Wow. Wow.
394. 58:13>> And a deo goes so you know I see some of you with falling bases out there. >> Was this the practice? Was this the
395. 58:19practice? This was the this was the last round to get to get picked or chosen to >> to to pitch your deck in front of the
396. 58:26investors tomorrow and they were only going to choose the top 10. >> Got it. >> And um he said, "You have 24 hours,
397. 58:33those of you who didn't make the cut to prove yourself." And I went right up to Mercedes and I said, "What time do you
398. 58:38want me there tomorrow?" And she goes, "What happened?" She's like, "Are you afraid to speak in front of people?" I was like, "No."
399. 58:43>> And and I said, "You tell me what time to be there tomorrow and I will blow your socks off." And I mean I did barely
400. 58:50inside. I was dying, but I got up there and I and she she was like, "Okay, you're on, kiddo." So I >> I got to pitch my dick
401. 58:59>> and I learned to to get sleep. Number one. Number two, don't talk yourself into a mind [ __ ] Just don't do it. Just don't do it.
402. 59:08>> Yep. Yep. >> Yep. >> And how did the pitch go? >> It went really well. Um there weren't a lot of food tech people in the audience.
403. 59:18Most of them are tech, right? >> Um but then the following two days there was the founder X summit and then there
404. 59:24was also something called step SF which was another founder investor summit and lots of networking parties, lots of
405. 59:31switching, you know, LinkedIn profiles and doing the nano nano iPhone to iPhone thing. >> And um I I got my advisor out of it. He
406. 59:41had been mentoring me almost throughout the entire program through uh founder institute. Um and last week I got my
407. 59:49first um offer for funding. Uh so I'm in diligence now which means they're going in and checking all my incorporation papers. I
408. 59:58>> was going to say yeah. of your capital. [laughter] >> Got to make sure that's all in place and and it's not it's not a ton and it's not
409. 1:00:06it's only just to sort of kick off um what what my seed round would be. These are words I didn't know two months ago.
410. 1:00:13>> Um but I have a very uh big venture capitalist firm in my space in food tech that agreed to take a call with me next
411. 1:00:22week. And so I'm just preparing for that. And we knock on wood, we'll see what happens. Yeah. >> Lovely. And I
412. 1:00:29>> It's all happening, Kyle. [laughter] >> That's beautiful. It's really good. I you know, I don't I don't have a ton of
413. 1:00:34advice on the on the investing front other than this >> is that the minute you get some momentum, the fact that you've got
414. 1:00:42someone now triangulate as much as you can. Tell everyone about tell everyone everything that keep them updated on,
415. 1:00:50oh, there's momentum. someone's interested because they love to follow the herd. No one wants to go first. So, the fact that you
416. 1:00:57have someone that says they're interested, especially if they say they're willing to lead around like everyone, you know, because this
417. 1:01:04happened with Mon'nique and I where I went to New York and I was I was pitching. This is for our seed round for Storybine.
418. 1:01:10And I talked to an investor in New York and he said he was willing to lead the round. And then we just we started we
419. 1:01:18called everyone we knew and a week later we were going to an investor conference in in Salt Lake City and by the time we
420. 1:01:26got there people were like, "Oh, Story Vine, I've heard of you." like like the like the the the network had done it. >> Yeah. They were all
421. 1:01:34>> had done it thing about it >> and they were all chatting about it and it was really cool but it was all
422. 1:01:38momentum like it wasn't the story. It wasn't the thing. It was just the fact that someone else was interested and
423. 1:01:44then other people were talking about it >> was the thing that ultimately I feel got us funded. >> So So just keep being a blabbermouth
424. 1:01:51about it. Just keep talking. >> Be an absolute blabbermouth about it and just and even even the people that maybe
425. 1:01:56you haven't talked to in a while. Hey, just let want to let you know I got this thing going. I've got some thing if you
426. 1:02:01know anyone. Just any of that because they'll start talking to each other and at some point you'll pick up the phone
427. 1:02:06and say, "Hey, this is Chef Kelly with with Die Converter." Like, "Oh, I've heard of you." Right. And that's, you know,
428. 1:02:12>> I would die. I would die. >> I know. It's good. It's good. [laughter] It's really awesome. Really >> great advice. Thank you. Yeah.
429. 1:02:18>> Um, so listen, thank you for for coming up here. I don't want to take your time. It's It's 8:30. It's late at night, but
430. 1:02:24um really appreciate you being here and and sharing your story and and just >> and I and I want to let you know that
431. 1:02:30without you and she leads AI as well. You know, two festivi ago, I stumbled into that room and you guys have been
432. 1:02:40lifesavers ever since and champions of me. So, I want to thank you. So, >> yeah. Well, follow your dreams out
433. 1:02:45there, guys. Follow your dreams. >> Yeah. Follow your dreams. Yeah. Yeah. Figure out what you're passion about. Go
434. 1:02:48do it. and just I want to acknowledge you and thank you for your participation in the salon and and she leads and here
435. 1:02:55um it really makes a difference and and uh you know being out in the world. So thanks >> really beautiful to see you. All right,
436. 1:03:02see you later. >> Have a good weekend everybody. Bye >> bye. >> Beautiful. That was awesome. Um okay. Um Jeff Kelly, generous leadership.
437. 1:03:14That's what it looks like. Build [ __ ] Learn [ __ ] Build [ __ ] Launch it. Go get it funded. Then talk
438. 1:03:25about it. Share your story. Share how you got there. Um, share the good stuff. Share the bad stuff. Um, so thank you to
439. 1:03:31Chef Kelly for that. Um, let's see. What do I want to talk about? Generous leadership. Generous leadership. Um, I'm going to go show you a
440. 1:03:53want to show you a new piece of techn Oh, you know what? Wait, we'll go back for a second. I want to I want to show
441. 1:03:58you another thing here. Um, I ran out of tokens in uh in Codeex, but let's see what what uh what it came up
442. 1:04:11to. So, let's see. Let me go look at my rate limits here. So, I have 33% weekly through the 14th and I have 0% until 1:00 a.m. tonight.
443. 1:04:22Okay. But let's go see what this thing did. Okay. So what we were looking to do is there are these brains out there, these these
444. 1:04:35neuron connected things based on a fruitfly brain, a fruit flies brain um that are out there that you can you can train to do
445. 1:04:50stuff. Okay, so it says the Google connection checks out. Flybody is an open-source jointed mujo co fly from Google deep mind and
446. 1:05:03hhmi janelia. I found's brain simulator but it's okay. Okay. So let's see. Eon's technical explanation revolves around an
447. 1:05:13important point. The demo uses neurommechfly a separate EPFL body and relies on existing movement controllers. E says,
448. 1:05:23"Learning and plasticity are largely missing from that model. I'm going to set up a local fly cup lab
449. 1:05:32with a clearly labeled learning demo and the real fly body assets that gives us something usable while we tackle the
450. 1:05:40research step. Making a fly controller grasp, lift, and release tiny cups. [laughter] By the way, we're going to teach uh a
451. 1:05:52fruitfly brain how to stack cups. [laughter] Oh, thoughts on custom GPTs migrating to plugins. I'll talk about that in a
452. 1:06:02second here. Okay. The real Flybody model is installed on an isolated environment on your Mac. Okay, cool. The training research
453. 1:06:13points to a useful sequence. First imitate successful movements, then improve the rewards. A recent Fly GM paper
454. 1:06:23applies that approach to controller structured. Okay, let me just tell you something right now. So, what this thing is describing
455. 1:06:32is that it went out and it found the the brain model. It went out and it found the body model. It went out and it read
456. 1:06:41all of the documentation and it went out and it figured out from that documentation how we're actually going
457. 1:06:47to train this thing like in in what sequence in what order. As someone with ADD, I wouldn't have made it past step
458. 1:06:55two of that process. I would have realized, oh, there's all this documentation I have to read and there's
459. 1:07:00probably math involved and concepts and so so getting agents to go do that [ __ ] for you is pretty damn good. Okay.
460. 1:07:09For the first lesson, I'm making cup order and navigation learnable with visible rewards and repeatable tests.
461. 1:07:19Grasping will be explicitly assisted in that lesson. Learning physical grasping with the flies leg is the next harder
462. 1:07:26milestone. Um, use Google Drive integration. Okay. Edited files. Okay. The first learning run works. The abstract wait edited files.
463. 1:07:40Okay. The first learning run works. The abstract cup controller completed all 120 starting arrangements
464. 1:07:50while a random controller completed none in the same test. This measures cup order and navigation learning only. The actual fly body also
465. 1:08:01renders and advances through Mojo Co. successfully. I'm now checking a small run of Eon's full connections so the lab can let you
466. 1:08:11explore three things separately. Neural activity, the articulated body, and a reward based and reward-based learning.
467. 1:08:21The real brain model runs here too in a 30 millisecond simulated sugar simulation test. The supplied 138,639
468. 1:08:33neuron network produced measurable downstream spikes. The neural simulation took about 10 seconds on GPU.
469. 1:08:42The lab [clears throat] now has separate tabs for cup learning, the real body, and the real brain. I'm checking the
470. 1:08:49controls and writing a demonstration guide with the remaining braintobody integration clearly marked. Okay, I'm
471. 1:08:55doing something on Google Drive. The browser demo is working. The learn controller completes its stack. You can inspect each decision
472. 1:09:06and the real flies pose controls work. I'm checking the no simulation brain control and the one click launcher. Now
473. 1:09:15the guide includes a fiveminute presentation script. The next research milestone is specific. Lift and release
474. 1:09:24one cup with the actual fly body. Then expand to stacking and test whether the connect to me base controller helps
475. 1:09:35use Google drive. And so let's see command decline. So I ran out of I ran out of [ __ ] So it says updated document
476. 1:09:47got document fly cup lab research. Wow. This is kind of bonkers.
477. 1:10:02agents.mmd created launch flycup lab command launch.python. Huh. All right. Well, I'll keep I'll
478. 1:10:16keep playing with that. That's kind of fun. Um, [clears throat] okay. So, let me see what questions do you have.
479. 1:10:29And by the way, if you're on TikTok and you want to see what I'm looking at, just go to my profile and click on the
480. 1:10:34link. That'll bring you over to the AI salon where the live stream is happening where you can see the screen. This is
481. 1:10:40obviously a live stream, too. But, and what's going on with the tennis match? Um, Shelton is up 43.
482. 1:10:54They're on served in the fourth. Okay. So, I decided to give Oh, yeah. So V6 is free access through 914. Um,
483. 1:11:11uh, what's his name? Um, uh, Gareth. Gareth has been playing with Suno V6. He says it's quite good. Um, I was kind of
484. 1:11:21poo pooing it because they've they've gotten rid of all their previous models because they were trained on uh
485. 1:11:30copywritten music and uh and without paying for it and so they're now owned by BMG and Sony and something else or I
486. 1:11:37don't know whoever else doesn't matter. Um, big music labels. So V6 is now trained on um music that BMG has the
487. 1:11:46rights to. Okay. So let's see. Frosty. So, I decided to give up on Opus 5 and Fable 5. Opus 5 was flagging any bio
488. 1:11:54health mention. Um, not sure why Anthropic flags almost any research health paper all of a sudden this week. Now, there listen
489. 1:12:05all of these well anthropic in particular is leaning into this AI is going to kill us thing. Anything health related, they're looking at
490. 1:12:12bioteterrorism. So, I think that's what they're trying to prevent. That's probably what that's about. Good
491. 1:12:18question from brother 52 on Tik Tok. Thoughts on custom GPTs migrating to plugins. So we talked about Festivus earlier. The
492. 1:12:27year before Festivus, we had this event three years ago called GPT for good or GPTs for good and it was fantastic. We
493. 1:12:36built custom GPTs for nonprofits. And um December 11th this year, so two months, custom GPTs are out of here. They're
494. 1:12:51done. They're being retired. You can convert your custom GPT to a plugin. I don't know how you do that.
495. 1:13:00You can ask chat GPT. It'll figure it out. But for those of you that have made dozens, if not hundreds of custom GPTs,
496. 1:13:09know that this is happening. So if you've got things that you would like to save, turn them into plugins. Now,
497. 1:13:17you're like, "But Kyle, like what's a why a plugin? What's how's it different?" Custom GPTs were designed
498. 1:13:24for human beings to log in and learn how to use your thing with your custom prompts and your buttons and all that.
499. 1:13:33and they had no way to monetize them and and they were powerful and very cool if you were nerdy enough to know what to
500. 1:13:41do with them. Plugins are designed for agents to be able to use them. So, if you've got some cool sort of pil
501. 1:13:51Pennsylvania Dutch recipe maker custom GPT, you could make a Pennsylvania Dutch recipe plugin that you can now make
502. 1:14:00available to like codeex and chat GPT work where those agents can use your plugin to do things. I don't know if
503. 1:14:08there's monetization. I assume there's not, but um that's happening. My thoughts on it are that's progress. That's
504. 1:14:18One of the risks of being an early adopter is you put a bunch of time into some piece of technology that then goes, I'm out of here.
505. 1:14:30Life in a big city. Um, also next Tuesday, oh, back to basics will be on, I believe. So, next Tuesday,
506. 1:14:44um, we have Back to Basics. I'm I'm not going to be here. I'm out of town. I'm at a conference. I'm going to be busy
507. 1:14:49with conference stuff. Um Dr. J, Claire Jacobs, and Rick McCaulay and producer Brandon are putting on Back to Basics.
508. 1:14:58It's going to be on music generation. There's a new model. Um I guess they're going to cram figuring out how to use
509. 1:15:06that over the weekend because apparently it's quite different. Um it's it's figure outable, right? But if if you had very specific prompting
510. 1:15:15techniques that worked with the previous models, apparently they're quite different now. Um, so that is next
511. 1:15:21Tuesday at 700 p.m. Eastern time. Uh, back to basics and ready to go. Cammy Ugg, I have 200 GPTs. Yay for being an early adopter. [laughter]
512. 1:15:38That's that's that's life in a big city there, boys. Yeah. You know, you know how it goes. Um anyway, you know, Brandon said to me
513. 1:15:51earlier today, does does this mean there's going to be a sin a Sin City Sipper plugin? I said, I guess there's
514. 1:15:57going to have to be. Um it's probably not a bad idea. Maybe next uh let's see, next Wednesday. I'm probably not going
515. 1:16:04to be here because I'm in conference land. Um, but at some point here, I think I think I will probably take a
516. 1:16:10custom GPT and on the live we'll figure out how to convert a custom GPT to a plugin. I can't imagine that it's that
517. 1:16:17hard. It's probably, you know, some different setup with some amount of system prompts and choose the model and
518. 1:16:24all the same [ __ ] you have to do. So Vicki, tomorrow, oh that's right. Tomorrow, LOL's are back. Thank you,
519. 1:16:32Vicki. Great to see you. Um, tomorrow Sid Harrow uh is doing an LOL and so go to that. It's 400 PM Eastern time on
520. 1:16:42Saturday and uh LOL's are awesome, right? It's people within the AI salon teaching other people within the AI
521. 1:16:49salon how to do stuff. Um, so please and Sid's Sid's just an awesome guy. By the way, if you're in the mastermind and you
522. 1:16:56haven't been to a practice lab in a while, Sid now leads the practice lab. He's an amazing uh facilitator of of the
523. 1:17:03practice lab. So, if you're looking to create a daily practice around how you use AI and I strongly recommend you do that,
524. 1:17:12um that group's a really powerful group and Sid heads that up. So, um if you want to RSVP for that, you go to the
525. 1:17:19events tab, which is toward the top of the the list down the side. You go to the top of that and uh click on the LOL
526. 1:17:26for tomorrow and RSVP. Support Sid. All right. And support your brain and go do that. Okay. Beautiful.
527. 1:17:35Um, when are we going to talk of Agentic with Claude? Um, I don't know. I haven't been using Claude. I I used Claude Code or Claude Co-work
528. 1:17:49once and then I used Codeex and I just like Codeex better. Um, this week I've kind of actually the last two weeks I've kind of hated codeex
529. 1:18:00because they put the Astro model in it which is this big ass powerful model but it's totally [ __ ] up my simple automations.
530. 1:18:11So I'm very frustrated with codeex right now. So maybe it is time to do that. Um, do you have that in your class? I don't
531. 1:18:19really have a class. I don't really do classes here. Um, and I haven't, you know, it's funny. I haven't really leaned too heavily into
532. 1:18:30teaching, um, setting up aic workflows on this channel because it's really [ __ ] boring. And so, so I could do it, but it's just
533. 1:18:43like I don't know. It's it's just it's just not fun. Um, have Astra do everything. Well, Astra does everything, including eat up
534. 1:18:51all your tokens. Vicki, sorry, I was just watching TV and realized I couldn't comment, so I had to log in. Oh, that's
535. 1:18:57funny. Beautiful. All right, so the uh LOL tomorrow. Next Tuesday is the um is back to basics. Um, Monday and Wednesday are
536. 1:19:12probably not going to happen uh for the for the learning lab just because I'm I'm out of pocket. Um, but next Friday
537. 1:19:21I'll be here. So, we'll have Friday night date night next Friday and we'll have all the all the rest of the
538. 1:19:25activities during the week. Um, by the way, if you're in the mastermind, Brandon's Codeex workshop. So, you know,
539. 1:19:32asking about agentic work. So within the mastermind, Brandon is in the middle of a four-week course on codec. So if
540. 1:19:40you're a member of the mastermind, those are available on demand and next Thursday is um his his third session um
541. 1:19:49about setting up codecs and setting up an agentic workflow and and actually applying that to your business. It's
542. 1:19:56it's really powerful. Um, okay. So, last little thing I'll show you before we we blast on out of here is
543. 1:20:05I want to show you a new thing. I showed this on office hours today. Um, but this is really quite something. Let me look at
544. 1:20:16Okay, still on serve in the fourth set. Okay. Um I think this is the tab. Is this the tab? Yes. Okay.
545. 1:20:37So, one of the things that has me excited about Astra, the the new AI fancy model, is that it's been optimized for computer use and for MCPs
546. 1:20:50and for agentic use of tools. It's really good at using other tools. So, it can it can just use applications running
547. 1:20:57on your computer. It can use MCPS and it can do it really well. I saw this demo of this this new thing called Blueprint blueprint.io.
548. 1:21:09Um, and it's not running on Astro that I know of. Uh, it's it it looks like it's some sort of, you know, proprietary
549. 1:21:15thing built on APIs and things like that. Um, but it's it very much points to a future that I think we're about to
550. 1:21:24enter where the tools that we use are actually a very simple front end to to a complicated network of services and
551. 1:21:31tools sitting underneath and this is one of those. So, I want to share it with you and let me know. I assume you can
552. 1:21:38hear this. I think I've got this shared correctly. You should be able to hear this. [music]
553. 1:21:53>> [music] [music]
554. 1:22:10[music] [music] >> We don't Black you
555. 1:22:25[music] know [music] [music]
556. 1:22:41[singing] you come on to >> [music] >> on. [music]
557. 1:22:57All I know. [music] [music]
558. 1:23:17[music] All right. So, if you are not a mechanical or electrical engineer or a product designer, an industrial
559. 1:23:33industrial designer, but you've always wanted, you've had ideas for products, but you don't know [ __ ] about products.
560. 1:23:42Now, you can now you can just do this. Um, let me, if you go to it's it's blueprint.io.
561. 1:23:57Let's see. Here's my This is one I did. So, this was this was my very first one. My prompt was um design me a cute dog hat that goes on
562. 1:24:13like a collar but has a gimbbled camera where the where the hat would be on the dog's head. So it looks like a cute hat,
563. 1:24:20but it takes smooth 2K or better videos. And so all right, on serve in the in the fourth set, 65 Shelton. So looks like
564. 1:24:30they might go to a tiebreaker in the fourth set. Um, so so it did the it did the this this design of the of the cute hat. It it's
565. 1:24:42very cute with the fake dog ears in there. Um, but that looks like something, you know, you could see in a
566. 1:24:48in a store. But then look, if I go to the parts, it figured out, here's all the parts that need to be used on this.
567. 1:24:55And if you're like, "Yeah, but I I don't know what these parts do. I don't know how they get wired together." Well,
568. 1:25:00it'll tell you how that happens, too. So, you can go in here and you connect things to, I don't know, to the whatever
569. 1:25:07the parts are. And then it's got sort of the mechanical thing. I guess this is a simulator where it connects all the
570. 1:25:14parts and figures out how the thing would run. Uh there's instructions. You can order all of the parts. Um are
571. 1:25:24these any good? Like do these work? Is it ordering the right parts? That I have no idea about. But the reason I wanted to share this is that
572. 1:25:36we're now far enough into AI generation where this the startups that are starting to happen are using
573. 1:25:47lots and lots of different models, lots and lots of different services. anthropic creating the MCP, the model
574. 1:25:53context protocol that allows AI agents to use other services and other websites and be able to do things like, you know,
575. 1:26:03log into a website and go pull all the products from it and compare Amazon products with Alibaba, with Teimu or
576. 1:26:09wherever they're sourcing them. um is this remarkable thing where the the front end here is this very simple thing where you can literally
577. 1:26:20describe a product, it'll design it and then it sort of does the leg work of how you put all this together. I would
578. 1:26:26imagine that there's a version of this business where you can probably push a button that says assemble this for me,
579. 1:26:35right? Or there could be a whole sub business of, you know, advertising for people who've designed these things and
580. 1:26:42ordered the parts, but now they're stuck. They don't know how to do them for you to go fix their thing and
581. 1:26:48prototype for them. So anyway, um I just thought that was one worth paying attention to going into the weekend. Um
582. 1:26:57in terms of weekend homework, um I would say this. I would say just be don't get caught up in the doomer versus
583. 1:27:07dreamer battle. It's all bad and it's going to kill us. It's all good and shangrila. Don't live on the edges of that. Live in
584. 1:27:15the live in the the paradox of the middle of understanding that some of these things are remarkable and powerful and yes,
585. 1:27:24they're scary and they're going to displace a lot of people. And you know, I would say just even though there's
586. 1:27:29going to be a lot of noise, you know, coming from the anti- AI doomer side, um just hang in the pocket, just just
587. 1:27:39think critically, do your research. Um and uh you know, in the end, none of this stuff ultimately matters. What matters is us as humans
588. 1:27:54and figuring out who we are, what we want, and using AI to amplify that. That's what the cycle of AI readiness is
589. 1:28:00all about. So, I'll close it where I opened it up. The cycle of AI readiness, you and your values and and who you care
590. 1:28:09about and what you want to do in the world is um is everything. And then how you learn is through play. playing with these AI tools without
591. 1:28:20expectation like I did tonight. I'm like, "Hey, Astra, go take one of these fly brains and let's see if we can get
592. 1:28:26it to stack cups." Um, and we're going to see if that works. Um, create excellence is, you know, build things
593. 1:28:33with integrity and generously lead is is talk about it in in the world. Talk about it with community. So, with that,
594. 1:28:41happy weekend everybody. Um, again, take a moment if you haven't today to just think about uh what happened on 9/11 and
595. 1:28:51uh acknowledge our first responders uh cuz they're they're out there doing all this still and in spite of uh that that
596. 1:29:00remarkable day. All right, so with that, I appreciate all of you and uh hope you have a beautiful weekend. Bye.
597. 1:29:12Thank you, Claire. Appreciate it.