You Need A Real Plan For What Comes After AI — Transcript

Full transcript

1. 0:00So, let's let's get into AI a little
2. 0:01bit.
3. 0:02Um
4. 0:03>> You're asking me the questions. I don't
5. 0:05know anything.
6. 0:05>> So, actually, um
7. 0:08not many people remember, but you were
8. 0:10one of the first people to key in on
9. 0:13ChatGPT as a as a tool for business.
10. 0:15>> Yeah.
11. 0:16>> And I remember some of your early
12. 0:17videos.
13. 0:18>> three almost two three years ago. Yeah,
14. 0:19it was a lot.
15. 0:20>> Yeah, it was like a couple weeks after
16. 0:21ChatGPT was launched, right? And like um
17. 0:25was it November '22 or something like
18. 0:27that? December '22.
19. 0:29>> Yeah, I remember.
20. 0:30>> Yeah. Um so uh and I was very interested
21. 0:34cuz I I always watch your world and I
22. 0:36watch the Silicon Valley world and I see
23. 0:37like what is hopping over. And when I
24. 0:39saw that, I was like, okay, this is
25. 0:41going to be huge.
26. 0:43>> [laughter]
27. 0:43>> Um so so what what what like what
28. 0:46sparked the interest for for ChatGPT?
29. 0:48What what what vision did it create for
30. 0:50you?
31. 0:53>> I mean, I I think that there's there's
32. 0:55fundamental questions of like, what is
33. 0:56it So, I'm such a such a Like, what do I
34. 1:00got or like, what do I really enjoy a
35. 1:01lot? Um is studying learning and
36. 1:03behavior. And you can hear it through
37. 1:05all my my definitions and my business
38. 1:07books. Like, that's the through line
39. 1:08that that marries everything. Um
40. 1:11and so then then
41. 1:14I mean, AI obviously begs the larger
42. 1:15question of like, what does it mean to
43. 1:16be human? And I think and I have this
44. 1:18philosophical slant to my own interest
45. 1:20and I think that's probably why I was
46. 1:21drawn to it disproportionately. I also
47. 1:22am a writer and so language is a natural
48. 1:24thing for me. And so the kind of the the
49. 1:27confluence of multiple these things
50. 1:29together, um made ChatGPT really
51. 1:30interesting. Um
52. 1:33but just in like
53. 1:34we we as humans tend to be very romantic
54. 1:36about our ourselves and thinking that
55. 1:38like no one can do it what we can do and
56. 1:39like we've been proven time and again
57. 1:40that that we're we're not as special as
58. 1:43we think we are.
59. 1:44>> Um that fits in your
60. 1:45>> Yeah, that fits in my world. Yeah, so
61. 1:47it's totally aligned with my world.
62. 1:48Yeah.
63. 1:48>> Yeah.
64. 1:49>> Um a lot of people are very upset by
65. 1:50that. But
66. 1:51but with with each thing that it proves
67. 1:53it can do because if if
68. 1:56artificial intelligence learns almost
69. 1:58exactly the same way that humans do
70. 2:00through reinforcement training. And so
71. 2:02you have you do a thing and you get an
72. 2:03outcome, yay or nay. And I believe at
73. 2:06the most foundational level when you ask
74. 2:07the question like why why are you so
75. 2:08driven or why are founders so driven?
76. 2:10We've gone through some reinforcement
77. 2:12training which we may have been aware or
78. 2:13not aware of that reinforced this set of
79. 2:16behaviors that when, you know, stacked
80. 2:18together so you see
81. 2:19a skill as a behavior chain of multiple
82. 2:22adaptive skills that are put in sequence
83. 2:24to create an outcome that's maybe you
84. 2:26know ideal or that adaptive chain
85. 2:28becomes more complex set of
86. 2:30behaviors that becomes a skill, sorry.
87. 2:32Um
88. 2:33And so to the same degree AI basically
89. 2:35learns the same way humans do. And so if
90. 2:37it learns the same way humans do then it
91. 2:38will be able to do what humans do.
92. 2:39>> Okay. Yeah, that's a very first
93. 2:41principles kind of thinking especially
94. 2:42watching ChatGPT. It kind of sucked back
95. 2:44then but like being able to key on that
96. 2:47was really
97. 2:47>> just keeps tweaking. And so um
98. 2:51it it's it obviously starts with with
99. 2:52just, you know, language but then it
100. 2:54just continues to move move out and I'm
101. 2:56curious what you think about like
102. 2:57Tesla's because, you know,
103. 3:00it trying to us trying to clean data
104. 3:02sets to train on larger and larger, you
105. 3:04know, you know, big data that it's that
106. 3:06it's working off of. Um obviously that
107. 3:09will have limits and then creating high
108. 3:11quality data will become the constraint
109. 3:12of the learning and so then it it will
110. 3:15have to learn from first principles
111. 3:16itself which is going to be
112. 3:17experimentation in the world which means
113. 3:18that it has to have some sort of link to
114. 3:21the real physical world which is exactly
115. 3:22how we learn.
116. 3:23>> Yes.
117. 3:23>> Right? And so we're able like we have A
118. 3:25So everyone who feels like, you know, AI
119. 3:26is not as smart as humans it's like,
120. 3:27well, because we've had a learning
121. 3:28advantage because we take 10,000 inputs
122. 3:31a day or whatever the number is.
123. 3:32>> 100%.
124. 3:32>> Um if we take all of our senses times,
125. 3:34you know, hours and whatever we can take
126. 3:36in. Um as soon as we have robo babies
127. 3:38and I say that not in a way like not in
128. 3:39a weird way but like um where it can
129. 3:41experiment and touch and taste and do
130. 3:43all that stuff in the world I think the
131. 3:44the reinforcement loops will happen so
132. 3:45quickly and the other part of it that
133. 3:46will happen is that it doesn't forget.
134. 3:49And I think that's the that's the crazy
135. 3:50part. So we've, you know, we've
136. 3:51obviously trained AI for different
137. 3:52functions within our business
138. 3:54um and business use cases. And one of
139. 3:56the things that was astonishing to me is
140. 3:57that we were trying to train a
141. 3:58an SDR.
142. 4:00Um so, a sales development rep who you
143. 4:02know, do outreach or do follow-up.
144. 4:04And what was really interesting to me is
145. 4:05that um obviously the hallucinations are
146. 4:07still a problem right now
147. 4:08uh with it and I think it will always be
148. 4:10a problem. I'm sure you've I don't know
149. 4:11the French guy, but the guy who Google
150. 4:14listens to and he's some you know,
151. 4:15>> French philosopher, right?
152. 4:16>> Might be. Yeah. Um but he basically was
153. 4:18saying that like LLMs will never truly
154. 4:20get into AGI because if you have 3%
155. 4:22error, it gets expounded, you know, ad
156. 4:25infinitum. And so,
157. 4:27uh but
158. 4:27>> He did moderate his view, but I'll tell
159. 4:28you in a second.
160. 4:28>> Yeah. Yeah. Oh, yeah. Please. Yeah, tell
161. 4:29me. Um but one of the things that I
162. 4:31found it so cool for me though is that
163. 4:33like if you know, it would it would make
164. 4:35a mistake and I would you know, we'd be
165. 4:36like, "Hey, do this next time." And
166. 4:38immediate and every time it would do
167. 4:40that. I was like, "Man, I've trained a
168. 4:41lot of sales people like
169. 4:43scary in a cool way of just how quick it
170. 4:45can learn and it just doesn't make a
171. 4:46mistake after that. Obviously, there's
172. 4:47hallucinations, but in terms of
173. 4:48following the directions, it would
174. 4:49follow. And so,
175. 4:51if if you
176. 4:52It's funny cuz there's there's a lot of
177. 4:54like really cute-isms in in Silicon
178. 4:56Valley like, you know,
179. 4:58being a master is something that like
180. 5:00some skills cannot be taught and only
181. 5:02learned. Like, they're these very cute
182. 5:03like rhetorical devices, but they're
183. 5:04just not they make no sense.
184. 5:05>> Yeah.
185. 5:07>> If you learn it, you learned it. So,
186. 5:08which means someone could teach it. It's
187. 5:09just not structured. And we didn't know
188. 5:11how we taught it cuz there's multiple
189. 5:12factors. But like if we if you control
190. 5:14the [clears throat] conditions, you can
191. 5:15control the outcome. At least that's my
192. 5:16view point of the world. And so,
193. 5:18it just means that there's more complex
194. 5:19environments in the real world to learn.
195. 5:20And so,
196. 5:21um I think I've just taken a lot of
197. 5:22interest in that because you know, the
198. 5:24obvious of like ChatGPT enters Optimus
199. 5:26or you know, Grok enters Optimus in the
200. 5:28real world is like,
201. 5:29"Oh, okay. Well, it can interact with
202. 5:32all the tools that we've already made
203. 5:33around the world to to interact with us
204. 5:34humans and then never make a mistake
205. 5:36after that." And then that just gets
206. 5:37into lots of questions around existence
207. 5:39and all that kind of stuff, which I
208. 5:40>> I I