1
00:00:03,120 --> 00:00:08,360
Speaker 1: Bloomberg Audio Studios, Podcasts, Radio News.

2
00:00:20,200 --> 00:00:23,640
Speaker 2: Hello and welcome to another episode of the Odd Lots Podcast.

3
00:00:23,720 --> 00:00:26,040
Speaker 3: I'm Joe Wisenthal and I'm Tracy Alloway.

4
00:00:26,200 --> 00:00:27,880
Speaker 2: Tracy, let's talk about AI some more.

5
00:00:29,960 --> 00:00:32,760
Speaker 1: Okay, Well, we could just have AI write the script

6
00:00:32,760 --> 00:00:34,360
Speaker 1: for us. We could give ourselves some time.

7
00:00:34,479 --> 00:00:36,360
Speaker 2: No, I don't think the technology is there yet, you

8
00:00:36,400 --> 00:00:39,680
Speaker 2: know what, so I can't say who. Maybe I was

9
00:00:39,720 --> 00:00:43,000
Speaker 2: trying to a professor recently and she said something really

10
00:00:43,040 --> 00:00:45,720
Speaker 2: interesting to me. I'm not supposed I'm not sure it's fine,

11
00:00:45,720 --> 00:00:48,279
Speaker 2: I think, And she said, like, you know, there's all

12
00:00:48,280 --> 00:00:51,879
Speaker 2: this anxiety about you know, kids cheating on their essays

13
00:00:51,960 --> 00:00:55,279
Speaker 2: or having Chad g GBT write their essays for them,

14
00:00:55,320 --> 00:00:58,120
Speaker 2: and like, you know, supposedly professors are like tearing their

15
00:00:58,120 --> 00:01:00,400
Speaker 2: hair out trying to figure out what to do about this,

16
00:01:00,480 --> 00:01:03,200
Speaker 2: and the AI detector don't really work all that well particularly,

17
00:01:03,560 --> 00:01:06,399
Speaker 2: but apparently like it seems like the solution is to

18
00:01:06,560 --> 00:01:09,800
Speaker 2: just grade them as regular essays no matter what. And

19
00:01:09,840 --> 00:01:12,600
Speaker 2: it sounds like at this point all the chat GPT

20
00:01:13,040 --> 00:01:16,160
Speaker 2: essays are basically solid c essays, and so if you

21
00:01:16,200 --> 00:01:18,039
Speaker 2: just sort of like take them at phase value, even

22
00:01:18,080 --> 00:01:20,920
Speaker 2: if you think they might be AI generated at least

23
00:01:20,959 --> 00:01:23,160
Speaker 2: at this point, it doesn't seem like a way at

24
00:01:23,200 --> 00:01:25,960
Speaker 2: least for college students to write good essays.

25
00:01:25,760 --> 00:01:29,520
Speaker 1: Yet, so our baseline, our average is chat GPT.

26
00:01:29,680 --> 00:01:32,080
Speaker 3: Now, yeah, that's basically can you beat the bot?

27
00:01:32,240 --> 00:01:35,560
Speaker 1: Did you see the thing someone was tweeting about one

28
00:01:35,640 --> 00:01:39,200
Speaker 1: of the tells for AI generated words is if you

29
00:01:39,400 --> 00:01:42,119
Speaker 1: use the word delve. Yes, I saw that, which, as

30
00:01:42,160 --> 00:01:45,600
Speaker 1: someone who I'm sure has used delve numerous times on

31
00:01:45,640 --> 00:01:47,640
Speaker 1: this podcast and in my writing, I thought was a

32
00:01:47,680 --> 00:01:50,320
Speaker 1: little unfair. It is a little it's a cliche, but

33
00:01:50,520 --> 00:01:52,520
Speaker 1: that doesn't mean it's from AI.

34
00:01:52,720 --> 00:01:55,960
Speaker 2: Although being said, there's more to AI than just chat

35
00:01:56,000 --> 00:01:59,160
Speaker 2: GPT obviously in chatbots, and this sort of has come

36
00:01:59,240 --> 00:02:02,840
Speaker 2: up on a couple episodes recently, but only like very tangentially.

37
00:02:03,240 --> 00:02:06,840
Speaker 2: People talk about the use of like AI and industrial applications,

38
00:02:06,880 --> 00:02:08,280
Speaker 2: and I've seen a lot of stuff. There have been

39
00:02:08,280 --> 00:02:12,760
Speaker 2: a couple Bloomberg articles about some startups that sort of say, like, okay, well,

40
00:02:12,800 --> 00:02:15,800
Speaker 2: like what if we trained robots the same way we

41
00:02:15,880 --> 00:02:18,480
Speaker 2: trade large language models, where you feed them just like

42
00:02:18,680 --> 00:02:21,280
Speaker 2: tons and tons and tons of real world data, and

43
00:02:21,360 --> 00:02:23,360
Speaker 2: so that yeah, like you sure you still have to

44
00:02:23,360 --> 00:02:26,600
Speaker 2: solve the mechanical engineering part. But then what if that

45
00:02:26,760 --> 00:02:30,080
Speaker 2: allows them, like all this training data to do more

46
00:02:30,280 --> 00:02:34,760
Speaker 2: advanced industrial things like I don't know, like make a

47
00:02:34,800 --> 00:02:38,560
Speaker 2: pizza or you know, be a more powerful humanless assembly

48
00:02:38,600 --> 00:02:39,079
Speaker 2: line or.

49
00:02:39,040 --> 00:02:39,840
Speaker 3: Something like that.

50
00:02:40,240 --> 00:02:42,800
Speaker 2: Where it's like we see all these oppressive robots and

51
00:02:42,880 --> 00:02:45,680
Speaker 2: videos like Boston Dynamics, but I never know like if

52
00:02:45,720 --> 00:02:48,000
Speaker 2: any of this is like quite there yet in terms

53
00:02:48,000 --> 00:02:49,000
Speaker 2: of having its value.

54
00:02:49,160 --> 00:02:52,679
Speaker 1: Yeah. So the robotics aspect of AI is something that's

55
00:02:52,800 --> 00:02:55,799
Speaker 1: incredibly interesting to me. It kind of makes me think

56
00:02:55,960 --> 00:02:58,760
Speaker 1: about the world that we want to see. So it

57
00:02:58,800 --> 00:03:02,280
Speaker 1: would be great if if we had physical robots who

58
00:03:02,320 --> 00:03:04,799
Speaker 1: are able to do stuff like clean up a house

59
00:03:05,120 --> 00:03:08,480
Speaker 1: or take care of an elderly family member or something

60
00:03:08,560 --> 00:03:11,320
Speaker 1: like that. It's not so great if all of our

61
00:03:11,320 --> 00:03:15,959
Speaker 1: technological prowess basically goes into writing satirical lyrics. Yes, via

62
00:03:16,080 --> 00:03:19,000
Speaker 1: chat GPT, like that's fun. I can do that myself,

63
00:03:19,400 --> 00:03:22,600
Speaker 1: But what I really need is someone to vacuum or

64
00:03:22,680 --> 00:03:23,600
Speaker 1: dust the house.

65
00:03:23,520 --> 00:03:24,320
Speaker 3: Do the laundry.

66
00:03:24,440 --> 00:03:26,760
Speaker 2: Yeah, exactly, really nice. All right, Well, I just want

67
00:03:26,760 --> 00:03:29,400
Speaker 2: to jump right into it because we really do have

68
00:03:29,680 --> 00:03:31,800
Speaker 2: the perfect guest. We're going to be speaking to someone

69
00:03:31,800 --> 00:03:34,960
Speaker 2: who has been investing in AI for a long time.

70
00:03:35,280 --> 00:03:39,040
Speaker 2: A lot of vcs like started investing in AI last

71
00:03:39,120 --> 00:03:42,560
Speaker 2: year obviously, but this is someone who has been investing

72
00:03:42,560 --> 00:03:45,360
Speaker 2: in AI for quite some time before it became the

73
00:03:45,360 --> 00:03:48,360
Speaker 2: hot new thing. We spoke to him last July and

74
00:03:48,440 --> 00:03:50,560
Speaker 2: had a great conversation about what he was seeing in

75
00:03:50,600 --> 00:03:53,400
Speaker 2: the space. So I'm really pleased to welcome back on

76
00:03:53,440 --> 00:03:55,240
Speaker 2: the show. We're going to be speaking with Josh Wolfe,

77
00:03:55,560 --> 00:03:59,360
Speaker 2: co founder and managing partner of Lux Capital. Josh, thank

78
00:03:59,360 --> 00:04:01,120
Speaker 2: you so much for coming back on ODLTS.

79
00:04:01,120 --> 00:04:03,440
Speaker 4: Great to be on. I feel like I should say

80
00:04:03,480 --> 00:04:04,920
Speaker 4: hello and like a robot voice.

81
00:04:06,560 --> 00:04:08,840
Speaker 3: So what's interesting to you these days? What are you

82
00:04:08,840 --> 00:04:10,360
Speaker 3: seeing out there that gets you excited?

83
00:04:10,720 --> 00:04:12,440
Speaker 4: Well, you know, you guys started this off with AI

84
00:04:12,720 --> 00:04:15,960
Speaker 4: and with an AI. Look, we've had what I would

85
00:04:15,960 --> 00:04:19,120
Speaker 4: call maybe a little bit pejorative, a little bit looting crude.

86
00:04:19,440 --> 00:04:21,880
Speaker 4: We've had chips everybody knows that we talked last time

87
00:04:21,880 --> 00:04:25,320
Speaker 4: about in AMD. We've got chatbots. You already have some

88
00:04:25,360 --> 00:04:27,479
Speaker 4: of these guys that are starting to fail. They've raised

89
00:04:27,480 --> 00:04:30,680
Speaker 4: billions of dollars and in some cases just you know,

90
00:04:30,800 --> 00:04:35,720
Speaker 4: relatively undifferentiated, big debate between open source that is approaching

91
00:04:35,760 --> 00:04:39,480
Speaker 4: the ASSYMP tote of achievement that the big private models

92
00:04:39,520 --> 00:04:42,799
Speaker 4: have and then being a little bit lout you've got chicks.

93
00:04:42,880 --> 00:04:44,719
Speaker 4: What do I mean by that? Most of the applications

94
00:04:44,720 --> 00:04:47,760
Speaker 4: in AI are the mundane and passe on the one side,

95
00:04:47,760 --> 00:04:51,520
Speaker 4: which would be like customer service and basic call center

96
00:04:52,000 --> 00:04:55,400
Speaker 4: supplementation or substitution. And then at the other end you

97
00:04:55,440 --> 00:04:57,400
Speaker 4: have people that are spending tens of thousands of dollars

98
00:04:57,440 --> 00:05:00,599
Speaker 4: a month in some case on AI girlfriends and people

99
00:05:00,640 --> 00:05:02,800
Speaker 4: that are doing what they often do with technology which

100
00:05:02,800 --> 00:05:05,840
Speaker 4: is used for prurient interests. So that to me the

101
00:05:05,880 --> 00:05:08,400
Speaker 4: two barbel extremes in AI, where people are actually making

102
00:05:08,440 --> 00:05:12,640
Speaker 4: money and profits serving demand for basic human instincts I

103
00:05:12,640 --> 00:05:16,680
Speaker 4: guess and needs. That's interesting. Overall, you're seeing a big

104
00:05:16,680 --> 00:05:19,719
Speaker 4: shift from the compute piece to the energy piece, meaning

105
00:05:19,960 --> 00:05:24,400
Speaker 4: people now recognize this bottleneck in AI is not going

106
00:05:24,440 --> 00:05:25,880
Speaker 4: to be so much about the chips. We also talked

107
00:05:25,880 --> 00:05:28,800
Speaker 4: about this months ago when we said, look, you don't

108
00:05:28,839 --> 00:05:31,960
Speaker 4: necessarily need these in video chips for inference, the part

109
00:05:32,000 --> 00:05:33,960
Speaker 4: that most people do when they're querying all these models.

110
00:05:33,960 --> 00:05:37,000
Speaker 4: You do need them for training, but the power levels

111
00:05:37,040 --> 00:05:39,599
Speaker 4: on these things are just enormous. There was a Dell

112
00:05:39,720 --> 00:05:42,960
Speaker 4: earnings call that they think sort of accidentally leaked that

113
00:05:43,040 --> 00:05:45,720
Speaker 4: this in VideA B one hundred Blackwell chip is going

114
00:05:45,760 --> 00:05:48,320
Speaker 4: to be a thousand wide draw of power, which is

115
00:05:48,360 --> 00:05:51,039
Speaker 4: like forty fifty percent more than these eight one hundred chips.

116
00:05:51,560 --> 00:05:53,840
Speaker 4: Why does that matter because now you got to figure

117
00:05:53,880 --> 00:05:56,279
Speaker 4: out how do you supply the energy for that. And

118
00:05:56,320 --> 00:05:59,880
Speaker 4: a tidbit which I thought was interesting was Amazon acquired

119
00:06:00,120 --> 00:06:03,520
Speaker 4: a nuclear powered data center in Pennsylvania. They've spent six

120
00:06:03,560 --> 00:06:06,080
Speaker 4: hundred and fifty million dollars, they got about a gig

121
00:06:06,200 --> 00:06:08,680
Speaker 4: lot of power. And I think that that is going

122
00:06:08,720 --> 00:06:09,880
Speaker 4: to be a trend. I think it's actually going to

123
00:06:09,960 --> 00:06:11,840
Speaker 4: usher in a wave of what I like to call

124
00:06:11,880 --> 00:06:14,880
Speaker 4: out elemental energy, but demand for nuclear power to power

125
00:06:14,920 --> 00:06:16,240
Speaker 4: these big AI data centers.

126
00:06:16,480 --> 00:06:18,160
Speaker 1: Yeah, it's kind of funny when you think about it.

127
00:06:18,200 --> 00:06:20,920
Speaker 1: I don't think anyone expected uranium to end up being

128
00:06:20,960 --> 00:06:23,680
Speaker 1: an AI play, but here we are. I want to

129
00:06:23,720 --> 00:06:26,280
Speaker 1: go back to something you said, and you actually brought

130
00:06:26,279 --> 00:06:28,960
Speaker 1: it up the last time we spoke to you last year,

131
00:06:29,240 --> 00:06:31,800
Speaker 1: and it was the idea of I guess the novelty

132
00:06:32,160 --> 00:06:36,200
Speaker 1: of some of these more public facing AI projects, And

133
00:06:36,360 --> 00:06:39,119
Speaker 1: I think you pointed out last year it's really fun

134
00:06:39,200 --> 00:06:42,080
Speaker 1: to use some of these things, like generate a bunch

135
00:06:42,080 --> 00:06:45,160
Speaker 1: of cartoon versions of yourself or whatever, but it might

136
00:06:45,200 --> 00:06:47,480
Speaker 1: not be a sustainable business model, and it might end

137
00:06:47,560 --> 00:06:51,920
Speaker 1: up being a functionality that is eventually incorporated into another

138
00:06:52,000 --> 00:06:55,520
Speaker 1: platform or a different project. Have you seen any example,

139
00:06:55,680 --> 00:07:01,080
Speaker 1: specific examples of more public facing novelty AI start to

140
00:07:01,320 --> 00:07:04,039
Speaker 1: like go away? I think you mentioned a few failures recently.

141
00:07:04,400 --> 00:07:06,280
Speaker 4: Yes, You've had a whole bunch of companies that basically

142
00:07:06,320 --> 00:07:09,200
Speaker 4: took chat, GPTA, GPT three or four and put a

143
00:07:09,200 --> 00:07:12,440
Speaker 4: wrapper around it, basically meaning give the average user who

144
00:07:12,480 --> 00:07:14,720
Speaker 4: doesn't know how to use these or even do prompts,

145
00:07:14,760 --> 00:07:17,080
Speaker 4: some means to interact with it. And those things raised

146
00:07:17,120 --> 00:07:18,720
Speaker 4: a bunch of money. They made these things accessible, and

147
00:07:18,720 --> 00:07:21,400
Speaker 4: they've sort of gone away. The foundation models themselves that

148
00:07:21,440 --> 00:07:24,560
Speaker 4: are undergirding all of this themselves are also starting to

149
00:07:24,560 --> 00:07:27,640
Speaker 4: be relatively commoditized away. And some of these things have

150
00:07:27,760 --> 00:07:29,560
Speaker 4: prominent people and have raised a lot of money, but

151
00:07:29,600 --> 00:07:32,640
Speaker 4: they are what I would consider failing. Take inflection. You've

152
00:07:32,600 --> 00:07:35,360
Speaker 4: got Mustapha Solimon, super smart guy, co founder of deep

153
00:07:35,400 --> 00:07:37,640
Speaker 4: mind raised I think a billion and a half for

154
00:07:37,680 --> 00:07:39,480
Speaker 4: that company. You know, I'm gonna be careful here because

155
00:07:39,520 --> 00:07:43,000
Speaker 4: Microsoft has been probably the savviest actor in this entire game,

156
00:07:43,320 --> 00:07:45,880
Speaker 4: figuring out that they can acquire things by doing things

157
00:07:45,920 --> 00:07:48,480
Speaker 4: in a clever way, skirting FTC and DJ oversight. They

158
00:07:48,480 --> 00:07:51,000
Speaker 4: effectively control open ai. As I think we also talked

159
00:07:51,000 --> 00:07:53,480
Speaker 4: about heard, particularly going into the end of last year

160
00:07:53,520 --> 00:07:55,960
Speaker 4: when there was all the drama around open Ai. Sadia said, look,

161
00:07:56,160 --> 00:07:58,120
Speaker 4: if open ai went out of business, we own it,

162
00:07:58,200 --> 00:08:00,600
Speaker 4: control it. We've got all the data up left, right center,

163
00:08:00,680 --> 00:08:03,320
Speaker 4: you know, all around them. Same thing with this company Inflection.

164
00:08:03,400 --> 00:08:04,840
Speaker 4: They did I think a six hundred and fifty six

165
00:08:04,920 --> 00:08:07,840
Speaker 4: hundred and seventy five million dollar license for the technology

166
00:08:08,240 --> 00:08:10,480
Speaker 4: that basically was a payment above and beyond what the

167
00:08:10,640 --> 00:08:13,360
Speaker 4: venture investors had made. Venture investors made a little bit

168
00:08:13,400 --> 00:08:16,320
Speaker 4: of money, not a lot. Key management went over to Microsoft.

169
00:08:16,360 --> 00:08:19,000
Speaker 4: But Microsoft has been very clever. So back to your question,

170
00:08:19,440 --> 00:08:20,920
Speaker 4: I think the big are going to get bigger and

171
00:08:20,960 --> 00:08:25,120
Speaker 4: are going to be most of the beneficiaries here, Microsoft, Adobe, Amazon,

172
00:08:25,280 --> 00:08:28,000
Speaker 4: Amazon themselves. Coming up on the one year anniversary of Bedrock,

173
00:08:28,040 --> 00:08:30,240
Speaker 4: they're going to announce that they've got the best performing

174
00:08:30,360 --> 00:08:33,320
Speaker 4: model with Anthropic, which they've made billions of dollars of

175
00:08:33,360 --> 00:08:36,200
Speaker 4: investments in now sort of competitor to Opening Eye CHATGPT.

176
00:08:36,280 --> 00:08:38,400
Speaker 4: They're also going to renounce something with one of our

177
00:08:38,440 --> 00:08:41,240
Speaker 4: companies that hasn't yet been publicly disclosed in biology. That's

178
00:08:41,280 --> 00:08:44,000
Speaker 4: going to be one of the two biggest waves next

179
00:08:44,040 --> 00:08:47,719
Speaker 4: biology and what you started the conversation with robotics. If

180
00:08:47,720 --> 00:08:51,319
Speaker 4: you just take robotics as an example, I think in

181
00:08:51,960 --> 00:08:53,839
Speaker 4: one of our companies, Hugging Face, which is one of

182
00:08:53,880 --> 00:08:56,880
Speaker 4: the main repositories for all these open source models, there's

183
00:08:56,920 --> 00:09:01,440
Speaker 4: something like sixty thousand text generations models. You know, it's

184
00:09:01,440 --> 00:09:03,720
Speaker 4: like fifty nine thousand, seven hundred something now, but just

185
00:09:03,760 --> 00:09:06,040
Speaker 4: an enormous number of text generation models. This is in

186
00:09:06,120 --> 00:09:08,520
Speaker 4: like two or three. It's like everybody's doing this. Everybody's

187
00:09:08,520 --> 00:09:10,280
Speaker 4: trying to do it all, basically trying to predict the

188
00:09:10,280 --> 00:09:13,760
Speaker 4: next word based on the prior one. What this transformer technology,

189
00:09:14,160 --> 00:09:17,520
Speaker 4: which was invented at Google, ended up parlaying into guess

190
00:09:17,520 --> 00:09:20,240
Speaker 4: how many robotic models there are? Fifty nine thousand in

191
00:09:20,280 --> 00:09:22,200
Speaker 4: text generation? Guess how many robotic models there are?

192
00:09:22,520 --> 00:09:23,120
Speaker 1: A fraction.

193
00:09:24,559 --> 00:09:27,600
Speaker 4: I mean, I'm obviously leading with my phone are but

194
00:09:28,360 --> 00:09:31,319
Speaker 4: nineteen robotic models. Okay, so you got that to me.

195
00:09:31,960 --> 00:09:34,040
Speaker 4: As a venture investor, we're just always looking at where's

196
00:09:34,040 --> 00:09:39,200
Speaker 4: their abundance, where's their scarcity? Their scarcity of robotic models? Now, why, Well,

197
00:09:39,240 --> 00:09:42,920
Speaker 4: it's relatively easy to train on the open Internet. You've

198
00:09:42,920 --> 00:09:45,400
Speaker 4: got Wikipedia, you've got YouTube videos. You know, whether you're

199
00:09:45,400 --> 00:09:47,120
Speaker 4: not you're supposed to be doing that or not. Like

200
00:09:47,120 --> 00:09:50,000
Speaker 4: the woman who was asked it, soa hey, how did

201
00:09:50,040 --> 00:09:51,560
Speaker 4: you train these things? Was it on YouTube? And she

202
00:09:51,600 --> 00:09:54,280
Speaker 4: gave it and you probably saw that, So there's gonna

203
00:09:54,280 --> 00:09:56,400
Speaker 4: be all kinds of copyright stuff on that. Robots are hard.

204
00:09:56,440 --> 00:09:59,400
Speaker 4: Why Most of the robotic stuff that has been out

205
00:09:59,400 --> 00:10:01,480
Speaker 4: in the world, like you talked about in your intro,

206
00:10:02,200 --> 00:10:06,320
Speaker 4: is constrained in manufacturing facilities, in work cells on an

207
00:10:06,320 --> 00:10:11,600
Speaker 4: assembly line, very specific, parametrically constrained, so very few degrees

208
00:10:11,600 --> 00:10:13,960
Speaker 4: of freedom of what they're actually doing. The robots themselves

209
00:10:14,040 --> 00:10:16,520
Speaker 4: might have multi access scripts and controllers, but they're not

210
00:10:16,559 --> 00:10:19,880
Speaker 4: moving around very freely. You have exceptions with like Amazon

211
00:10:20,120 --> 00:10:24,200
Speaker 4: which acquired Kiva, moving the warehouse inventory stuff around, but

212
00:10:24,280 --> 00:10:30,360
Speaker 4: again relatively xyz access not unstructured environments. You and I

213
00:10:30,840 --> 00:10:33,880
Speaker 4: and our listeners. We all thrive every day in unstructured

214
00:10:33,960 --> 00:10:38,240
Speaker 4: environments and that is where you need enormous training data.

215
00:10:38,440 --> 00:10:40,280
Speaker 4: You can't search the internet for that, So how do

216
00:10:40,320 --> 00:10:43,079
Speaker 4: you do it. There's a few things that have emerged,

217
00:10:43,080 --> 00:10:45,920
Speaker 4: and you mentioned some articles. We funded a company that

218
00:10:45,960 --> 00:10:48,400
Speaker 4: recently came out of STELL called Physical Intelligence stead of

219
00:10:48,480 --> 00:10:51,559
Speaker 4: artificial intelligence, Physical Intelligence, and it is the krem Dela

220
00:10:51,600 --> 00:10:54,320
Speaker 4: Creme team from Stanford and Berkeley. You've got some open

221
00:10:54,360 --> 00:10:57,240
Speaker 4: ai folks, You've got Google Deep Mind folks. They took

222
00:10:57,240 --> 00:10:59,600
Speaker 4: investment from open ai US and a bunch of other

223
00:10:59,679 --> 00:11:03,560
Speaker 4: vcs and they are just twenty four to seven training

224
00:11:03,640 --> 00:11:07,360
Speaker 4: robots doing all kinds of crazy things like folding, laundry, pouring.

225
00:11:07,440 --> 00:11:12,240
Speaker 4: Determined but to let these robots encounter unstructured environments and

226
00:11:12,280 --> 00:11:15,480
Speaker 4: then be able to thrive in them. The next thing

227
00:11:15,480 --> 00:11:18,480
Speaker 4: that you're going to see are visual models where you're

228
00:11:18,520 --> 00:11:21,760
Speaker 4: effectively giving, like an ikea sketch or you're drawing something,

229
00:11:21,840 --> 00:11:24,800
Speaker 4: and you're being able to instruct the robot to have

230
00:11:24,840 --> 00:11:27,439
Speaker 4: a sense of intuitive physics of how the world works

231
00:11:27,440 --> 00:11:29,559
Speaker 4: and how things might connect to each other and then

232
00:11:30,120 --> 00:11:33,360
Speaker 4: learn from that. And then we're also training these robots

233
00:11:33,360 --> 00:11:36,560
Speaker 4: with simple verbal cues. So there's a video you can

234
00:11:36,600 --> 00:11:39,440
Speaker 4: see online from some of the researchers where they are

235
00:11:39,480 --> 00:11:42,560
Speaker 4: picking nuts and m and ms and separating them, you know,

236
00:11:42,640 --> 00:11:44,440
Speaker 4: just as a task of being able to sort and

237
00:11:44,480 --> 00:11:47,480
Speaker 4: filter with precision and dexterity, and if they picked a

238
00:11:47,480 --> 00:11:50,120
Speaker 4: wrong one, you can actually instead of physically grabbing the things,

239
00:11:50,120 --> 00:11:53,760
Speaker 4: say stop grab the Eminem's not the nuts. And now

240
00:11:53,800 --> 00:11:57,040
Speaker 4: it knows that. So I think that we're about to

241
00:11:57,120 --> 00:12:01,000
Speaker 4: unleash in robotics what will become a chat GPT like

242
00:12:01,080 --> 00:12:03,439
Speaker 4: moment where people are so used to seeing robots and

243
00:12:03,480 --> 00:12:05,679
Speaker 4: they see the arms, and they've seen West World and

244
00:12:05,720 --> 00:12:08,600
Speaker 4: this kind of stuff, and suddenly something happens that just

245
00:12:08,679 --> 00:12:11,240
Speaker 4: blows your mind. And I think that's coming soon.

246
00:12:27,120 --> 00:12:29,960
Speaker 2: That's pretty exciting because, like I said, you know, for

247
00:12:30,120 --> 00:12:33,000
Speaker 2: like at least a decade, I've been watching those Boston

248
00:12:33,480 --> 00:12:36,440
Speaker 2: whatever online, those youtubes, and at this point I'm sort

249
00:12:36,440 --> 00:12:38,960
Speaker 2: of convinced that it's like basically a content generator because

250
00:12:38,960 --> 00:12:41,280
Speaker 2: it never seems like they're like crazy robot dogs or

251
00:12:41,320 --> 00:12:44,040
Speaker 2: anything like ever become commercial. But maybe this is the

252
00:12:44,040 --> 00:12:46,880
Speaker 2: missing link. But you've brought up like three different avenues

253
00:12:46,920 --> 00:12:48,400
Speaker 2: we could go on and I want to sort of

254
00:12:48,400 --> 00:12:51,120
Speaker 2: eventually hit on any of them. Here's a specific question,

255
00:12:51,120 --> 00:12:53,920
Speaker 2: and then we can maybe get back to robotics. This

256
00:12:54,080 --> 00:12:57,760
Speaker 2: element where there is such a shortage of advanced, cutting

257
00:12:57,880 --> 00:12:59,840
Speaker 2: edge talent, people who really know how to do this.

258
00:13:00,200 --> 00:13:02,960
Speaker 2: And you mentioned that guy that got hired by Microsoft

259
00:13:03,000 --> 00:13:07,440
Speaker 2: from his other companies as an investor in AI or

260
00:13:07,559 --> 00:13:11,160
Speaker 2: robotics startups. Is this a dynamic that's different than in

261
00:13:11,280 --> 00:13:15,400
Speaker 2: other software or other tech investing. Basically, this sort of

262
00:13:15,440 --> 00:13:18,520
Speaker 2: like highly skilled tech qman risk.

263
00:13:18,600 --> 00:13:22,640
Speaker 4: Basically yes, in that you always are looking for what's

264
00:13:22,640 --> 00:13:25,040
Speaker 4: scarce and you want scarce talent. If anybody could do this,

265
00:13:25,120 --> 00:13:27,920
Speaker 4: it's just not that valuable companies would get funded. Vcs

266
00:13:27,920 --> 00:13:30,440
Speaker 4: would fund forty of them at the same time, or

267
00:13:30,520 --> 00:13:33,120
Speaker 4: maybe four hundred of them. Contrast to things where it's

268
00:13:33,240 --> 00:13:35,840
Speaker 4: very web based or the old groupons of yesterday. This

269
00:13:35,920 --> 00:13:40,160
Speaker 4: is highly technical, often PhD scientists. The vast majority of

270
00:13:40,200 --> 00:13:43,080
Speaker 4: the founding teams that we've backed at companies like Covariant

271
00:13:43,360 --> 00:13:46,760
Speaker 4: or format or this new company Physical Intelligence, they're all

272
00:13:46,840 --> 00:13:50,319
Speaker 4: PhDs that are coming out of Stanford, Carnegie, Mel and MIT.

273
00:13:50,800 --> 00:13:53,640
Speaker 4: Some of the best robotics programs in the world, and

274
00:13:53,840 --> 00:13:57,160
Speaker 4: there's lineage of these great professors. Many of them have passed,

275
00:13:57,160 --> 00:13:59,320
Speaker 4: but for example, there's this one guy, Hans Morvec used

276
00:13:59,320 --> 00:14:00,679
Speaker 4: to be a carnegiemail and I got to meet him

277
00:14:00,679 --> 00:14:02,199
Speaker 4: when he was still alive, but he was one of

278
00:14:02,240 --> 00:14:04,880
Speaker 4: the very early pioneers in robotics. And he's got this

279
00:14:05,000 --> 00:14:09,199
Speaker 4: paradox that insiders in the robotics world called the Morvek paradox,

280
00:14:09,520 --> 00:14:12,320
Speaker 4: which is this weird counterintuitive phenomenon which is basically like

281
00:14:12,679 --> 00:14:15,120
Speaker 4: all the stuff that we think is really hard is

282
00:14:15,120 --> 00:14:17,920
Speaker 4: actually pretty easy for AI, and all the stuff that

283
00:14:17,960 --> 00:14:20,760
Speaker 4: we find totally intuitive and easy, like riding a bike,

284
00:14:21,000 --> 00:14:24,440
Speaker 4: that's really hard for robots. So there's this great paradox

285
00:14:24,480 --> 00:14:26,560
Speaker 4: that some of the most brilliant researchers are working on,

286
00:14:27,400 --> 00:14:29,040
Speaker 4: which is how do we do the kind of stuff

287
00:14:29,040 --> 00:14:31,280
Speaker 4: that a four year old can do very intuitively with

288
00:14:31,360 --> 00:14:33,560
Speaker 4: these very complex, expensive machines. And there's all kinds of

289
00:14:33,560 --> 00:14:35,760
Speaker 4: considerations we could talk about about where are these arms

290
00:14:35,840 --> 00:14:39,200
Speaker 4: coming from? The acquisitions that China has been making from

291
00:14:39,240 --> 00:14:41,520
Speaker 4: what historically was a lot of German companies. I mean

292
00:14:41,520 --> 00:14:43,760
Speaker 4: when I say arms meeting the robotic arms they can

293
00:14:43,840 --> 00:14:46,760
Speaker 4: move thing. And then there's this great philosophical debate that

294
00:14:46,760 --> 00:14:48,600
Speaker 4: it hasn't yet come to four. But I believe will

295
00:14:49,000 --> 00:14:51,120
Speaker 4: and investors are sort of lining up on this. I'm

296
00:14:51,160 --> 00:14:55,680
Speaker 4: on the opposite side of some people are funding humanoid robots.

297
00:14:56,120 --> 00:14:57,600
Speaker 4: And the reason that I say I'm on the opposite

298
00:14:57,600 --> 00:14:59,640
Speaker 4: side of it is I don't really believe in them. Yes,

299
00:15:00,040 --> 00:15:01,480
Speaker 4: you would want somebody to help take care of your

300
00:15:01,520 --> 00:15:04,920
Speaker 4: grandma and maybe provide some companionship. But this idea of

301
00:15:04,960 --> 00:15:08,960
Speaker 4: the movies of these ex machina kind of robots that

302
00:15:09,160 --> 00:15:13,800
Speaker 4: embody a human form. We know that engineering is better

303
00:15:13,920 --> 00:15:16,840
Speaker 4: than evolution. If we were inventing a car tomorrow, it

304
00:15:16,840 --> 00:15:19,400
Speaker 4: would be a terrible idea to take Fred Flintstone and

305
00:15:19,520 --> 00:15:22,280
Speaker 4: use his feet, you know, to power these stone wheels.

306
00:15:22,320 --> 00:15:24,640
Speaker 4: We know that an actuator and the axle and an

307
00:15:24,640 --> 00:15:27,680
Speaker 4: engine or just better, and evolution didn't create that. Why

308
00:15:27,680 --> 00:15:31,480
Speaker 4: would we create these humanoid hands where if I'm twisting

309
00:15:31,520 --> 00:15:33,640
Speaker 4: the cap off of a bottle, you know I can

310
00:15:33,720 --> 00:15:35,440
Speaker 4: only I have to turn my hand like seven times

311
00:15:35,440 --> 00:15:38,600
Speaker 4: to do that. Whereas if I was just designed the

312
00:15:38,600 --> 00:15:40,800
Speaker 4: perfect robot, I'd have a little suction cup that would

313
00:15:40,800 --> 00:15:42,640
Speaker 4: go on top. It'd have like a drill bit mechanism.

314
00:15:42,680 --> 00:15:44,720
Speaker 4: It would quickly twist it off and then it would

315
00:15:44,920 --> 00:15:47,360
Speaker 4: Swiss army knife, you know, swap out for the next

316
00:15:48,280 --> 00:15:52,280
Speaker 4: technical gripper capability. So I think that people are misguided

317
00:15:52,800 --> 00:15:54,600
Speaker 4: and they're basically going to end up doing things for

318
00:15:54,680 --> 00:15:59,040
Speaker 4: like prosthetics or you know, something that's sort of Westworld like.

319
00:15:59,480 --> 00:16:02,120
Speaker 4: But I think the practical robots that we're going to

320
00:16:02,200 --> 00:16:03,680
Speaker 4: all be using in our homes are going to look

321
00:16:03,680 --> 00:16:05,160
Speaker 4: nothing like these humanoid robots.

322
00:16:05,360 --> 00:16:08,320
Speaker 1: This is funny. This is very reminiscent of a weird

323
00:16:08,400 --> 00:16:10,920
Speaker 1: conversation I used to have with my dad. He had

324
00:16:11,040 --> 00:16:16,040
Speaker 1: like some sort of bugbear around the shape of aliens hands,

325
00:16:16,400 --> 00:16:18,640
Speaker 1: and he was like, why are they always shown or

326
00:16:18,720 --> 00:16:24,040
Speaker 1: depicted in these illustrations as having like human like hands

327
00:16:24,160 --> 00:16:26,760
Speaker 1: or sometimes even three fingers, Like why wouldn't they have

328
00:16:26,960 --> 00:16:32,520
Speaker 1: just evolved to the next level of very very efficient physiology. Anyway,

329
00:16:32,800 --> 00:16:34,920
Speaker 1: One thing I wanted to ask, and I'm trying to

330
00:16:34,920 --> 00:16:37,560
Speaker 1: think how to frame this question or what the right

331
00:16:37,600 --> 00:16:43,320
Speaker 1: word is, but how open source is robotics in the

332
00:16:43,360 --> 00:16:46,720
Speaker 1: sense of, like how much of the technology is shareable

333
00:16:47,000 --> 00:16:49,600
Speaker 1: or replicable? Because I feel like one of the reasons,

334
00:16:49,640 --> 00:16:51,360
Speaker 1: and you touched on it earlier, but one of the

335
00:16:51,360 --> 00:16:55,280
Speaker 1: reasons we have seen this boom in AI is because

336
00:16:55,320 --> 00:16:58,880
Speaker 1: you can go to places like hugging Face and download

337
00:16:59,000 --> 00:17:02,160
Speaker 1: a bunch of code, open source code, and build off

338
00:17:02,200 --> 00:17:05,919
Speaker 1: of it and it sort of multiplies around itself. But

339
00:17:06,080 --> 00:17:09,159
Speaker 1: is there any aspect of that all in robotics or

340
00:17:09,200 --> 00:17:11,119
Speaker 1: is it just much more proprietary.

341
00:17:11,520 --> 00:17:15,040
Speaker 4: The hardware piece has historically been very proprietary, although there's

342
00:17:15,080 --> 00:17:18,120
Speaker 4: lots of knockoff things. There is a Chinese company which

343
00:17:18,160 --> 00:17:20,480
Speaker 4: is increasingly dominating the field. A lot of people don't

344
00:17:20,480 --> 00:17:23,960
Speaker 4: know the name called Unitry Uni Tree Unitry that is

345
00:17:24,000 --> 00:17:27,200
Speaker 4: sort of copying the Boston Dynamics robots that Joe was

346
00:17:27,240 --> 00:17:29,720
Speaker 4: talking about and that you see in Black Mirror episodes

347
00:17:29,880 --> 00:17:33,560
Speaker 4: and those kinds of things. On the software side, it

348
00:17:33,720 --> 00:17:38,000
Speaker 4: really because it has the same kernel, the same origins

349
00:17:38,400 --> 00:17:41,840
Speaker 4: as many of the AI software things that led to

350
00:17:41,920 --> 00:17:45,600
Speaker 4: large language models and came from transformers academic roots and

351
00:17:45,640 --> 00:17:48,080
Speaker 4: academics like to share and publish, and of course you

352
00:17:48,119 --> 00:17:50,880
Speaker 4: can patent certain things, but by and large, the early

353
00:17:51,720 --> 00:17:55,560
Speaker 4: systems something called ROSS, as you might guess, robot operating system.

354
00:17:55,840 --> 00:17:58,159
Speaker 4: People that were doing something called r duino, which is

355
00:17:58,520 --> 00:18:02,119
Speaker 4: sort of for hobby programmers with hardware and software at

356
00:18:02,160 --> 00:18:06,119
Speaker 4: the intersection Poppy. There's a handful of these things. But again,

357
00:18:06,200 --> 00:18:08,480
Speaker 4: now you're in this mode where you need to find

358
00:18:08,560 --> 00:18:11,840
Speaker 4: training data and you need to do the work and

359
00:18:11,920 --> 00:18:14,560
Speaker 4: spend the time, and that costs money. So you will

360
00:18:14,600 --> 00:18:17,280
Speaker 4: have a mix of open and closed models. And if

361
00:18:17,320 --> 00:18:19,760
Speaker 4: you take a company like Physical Intelligence, their mo their

362
00:18:19,800 --> 00:18:23,200
Speaker 4: motive is is we want to build the operating system

363
00:18:23,200 --> 00:18:25,639
Speaker 4: that any robot can basically use to navigate the world.

364
00:18:25,640 --> 00:18:28,800
Speaker 4: They want to build the brains for the robots as

365
00:18:28,800 --> 00:18:32,800
Speaker 4: opposed to the robots themselves. And there's an interesting philosophical

366
00:18:32,840 --> 00:18:36,800
Speaker 4: and scientific tangent which Barbara Tavski, who is a friend

367
00:18:36,840 --> 00:18:39,879
Speaker 4: and she was the partner of now the late Danny Connoman,

368
00:18:39,920 --> 00:18:42,439
Speaker 4: who was also a friend, her work and she was

369
00:18:42,600 --> 00:18:44,880
Speaker 4: far less famous, but I think actually more important work

370
00:18:45,240 --> 00:18:49,359
Speaker 4: is all about motor function, and her hypothesis is that

371
00:18:49,440 --> 00:18:52,520
Speaker 4: the entire reason for the existence of the brain, like

372
00:18:52,560 --> 00:18:56,640
Speaker 4: the sole purpose of it is to actually produce movement,

373
00:18:56,880 --> 00:19:00,239
Speaker 4: movement towards food or a mate, or to run from

374
00:19:00,280 --> 00:19:02,119
Speaker 4: a prey or predator, which in turn is doing the

375
00:19:02,160 --> 00:19:04,800
Speaker 4: same kind of thing, and I think that some of

376
00:19:04,840 --> 00:19:11,680
Speaker 4: the most interesting philosophical questions about consciousness and memory, spatial perception,

377
00:19:12,160 --> 00:19:16,600
Speaker 4: embodied cognition, gesturing. I mean, I'm wildly gesticulating while I'm

378
00:19:16,600 --> 00:19:18,600
Speaker 4: talking now with my hands. It's just an innate thing

379
00:19:18,760 --> 00:19:21,040
Speaker 4: being able to do mental simulations, like when we think

380
00:19:21,080 --> 00:19:24,640
Speaker 4: about human brains and machine brains merging. I actually think

381
00:19:24,680 --> 00:19:28,800
Speaker 4: it's really going to be very revelatory as these robotic

382
00:19:28,800 --> 00:19:31,320
Speaker 4: systems advance and us understanding that's a lot of the

383
00:19:31,400 --> 00:19:34,879
Speaker 4: purpose of thinking and intelligence is actually just about moving.

384
00:19:35,359 --> 00:19:37,720
Speaker 4: And so that's a pretty cool side effect that I

385
00:19:37,720 --> 00:19:39,400
Speaker 4: think is going to come from all the commercial and

386
00:19:39,640 --> 00:19:42,400
Speaker 4: speculative ventures stuff that we do and funding these things.

387
00:19:42,720 --> 00:19:46,080
Speaker 2: You keep saying things that like jog me onto something

388
00:19:46,080 --> 00:19:47,560
Speaker 2: else that I was meaning to ask you, but you

389
00:19:47,680 --> 00:19:49,639
Speaker 2: mentioned this impulse.

390
00:19:49,240 --> 00:19:52,360
Speaker 3: That academics like to share their work, and.

391
00:19:52,320 --> 00:19:54,479
Speaker 2: That reminded me of something that someone was telling me.

392
00:19:54,640 --> 00:19:57,280
Speaker 2: So you know, maybe I'm sure you are on this

393
00:19:57,320 --> 00:19:59,679
Speaker 2: site a lot. But for listeners, there's this site archived

394
00:19:59,720 --> 00:20:03,320
Speaker 2: dot org where people like publish research and sort of

395
00:20:03,359 --> 00:20:06,560
Speaker 2: like an open source, sort of ungated manner about all

396
00:20:06,640 --> 00:20:09,760
Speaker 2: kinds of scientific and computer things and just today on

397
00:20:09,840 --> 00:20:12,879
Speaker 2: the artificial intelligence page, there's like fifteen new papers and

398
00:20:12,880 --> 00:20:16,520
Speaker 2: they have headlines like autonomous evaluation and Refinement of digital

399
00:20:16,560 --> 00:20:20,520
Speaker 2: agents or a modular benchmark framework to measure progress and

400
00:20:20,640 --> 00:20:23,760
Speaker 2: improve LLM agents. And this guy was who I was

401
00:20:23,760 --> 00:20:25,920
Speaker 2: talking to, made the contention that's like all this stuff

402
00:20:26,000 --> 00:20:29,400
Speaker 2: is being published and investors, a lot of vcs don't

403
00:20:29,440 --> 00:20:32,640
Speaker 2: really have these sort of like technical chops to judge

404
00:20:32,680 --> 00:20:33,720
Speaker 2: a lot of this research.

405
00:20:34,359 --> 00:20:35,200
Speaker 3: It's like the hear.

406
00:20:35,080 --> 00:20:39,320
Speaker 2: Take my money meme, I'm curious, like what you see

407
00:20:39,440 --> 00:20:42,040
Speaker 2: in this space where it's like there must be a

408
00:20:42,040 --> 00:20:44,720
Speaker 2: lot of investors like yourself who are like wowed by

409
00:20:44,960 --> 00:20:47,200
Speaker 2: PhDs like doing all kinds of stuff. It's like, we

410
00:20:47,280 --> 00:20:50,879
Speaker 2: got a one hundred x breakthrough in the energy efficiency

411
00:20:50,920 --> 00:20:53,760
Speaker 2: of this in video chip by training the model differently,

412
00:20:53,840 --> 00:20:56,560
Speaker 2: like how do you evaluate the science and how risky

413
00:20:56,640 --> 00:20:59,680
Speaker 2: is that for investors with a sort of like plethora

414
00:20:59,800 --> 00:21:02,240
Speaker 2: of ostensible breakthrough is happening left and right.

415
00:21:02,240 --> 00:21:04,480
Speaker 4: Well, you're right, there's endless I mean it is why

416
00:21:04,520 --> 00:21:06,320
Speaker 4: there's this endless frontier and sort of the batter for

417
00:21:06,400 --> 00:21:08,959
Speaker 4: bush sense, and it's always very exciting and you need

418
00:21:09,000 --> 00:21:11,040
Speaker 4: to have a very high filter for these things. Not

419
00:21:11,080 --> 00:21:13,720
Speaker 4: everything is commercent. Sometimes there's just a breakthrough, but maybe

420
00:21:13,760 --> 00:21:15,879
Speaker 4: that breakthrough can be licensed to a company. And so

421
00:21:15,920 --> 00:21:18,440
Speaker 4: the people who are actually commercializing this and thinking about

422
00:21:18,440 --> 00:21:22,399
Speaker 4: capital allocation and recruiting teams and then deciding these are

423
00:21:22,400 --> 00:21:24,920
Speaker 4: our top three priorities, and even though there's forty other

424
00:21:24,960 --> 00:21:27,080
Speaker 4: really exciting things that we could and maybe should be doing,

425
00:21:27,200 --> 00:21:28,880
Speaker 4: we're just not going to do that. Now. That's really

426
00:21:28,960 --> 00:21:32,280
Speaker 4: what company building is about. And so oftentimes we might

427
00:21:32,280 --> 00:21:34,440
Speaker 4: have a brilliant scientist, but maybe that person just isn't

428
00:21:34,480 --> 00:21:36,960
Speaker 4: a good salesperson. They can't tell a narrative and convince

429
00:21:37,000 --> 00:21:39,159
Speaker 4: people to move across the country and join them, they

430
00:21:39,160 --> 00:21:40,879
Speaker 4: can't raise capital, and therefore they're not going to be

431
00:21:40,920 --> 00:21:43,360
Speaker 4: a great entrepreneur and they're probably better off as a scientist.

432
00:21:43,640 --> 00:21:47,480
Speaker 4: But when we're making evaluations, it's how much money is

433
00:21:47,480 --> 00:21:49,600
Speaker 4: going to accomplish what in what period of time? And

434
00:21:49,600 --> 00:21:51,480
Speaker 4: who's going to care. It's like if you were playing poker.

435
00:21:51,520 --> 00:21:53,120
Speaker 4: You're looking at your hand, You're figuring out how much

436
00:21:53,119 --> 00:21:54,760
Speaker 4: money do you have to anti up for the next round,

437
00:21:55,000 --> 00:21:57,800
Speaker 4: And then what's the exogenous what's the outside view, what's

438
00:21:57,800 --> 00:21:59,880
Speaker 4: the market going to say, and will they care. That's why,

439
00:22:00,440 --> 00:22:01,719
Speaker 4: you know, we talked a little bit about this, but

440
00:22:01,800 --> 00:22:04,120
Speaker 4: very skeptical about other fields that people are funding right now,

441
00:22:04,720 --> 00:22:08,520
Speaker 4: in fusion or in quantum computing. I'm so skeptical about

442
00:22:08,560 --> 00:22:11,960
Speaker 4: them in part from hard and cynicism of twenty years

443
00:22:12,000 --> 00:22:14,800
Speaker 4: of people pitching us things that are always about unbreakable

444
00:22:14,840 --> 00:22:19,119
Speaker 4: cryptography and femtosecond and kneeling of this quantity, and I'm like, so,

445
00:22:19,280 --> 00:22:22,040
Speaker 4: what most of the things that people promised, you know,

446
00:22:22,119 --> 00:22:26,320
Speaker 4: about unbreakable cryptography or molecular modeling for people are doing.

447
00:22:26,359 --> 00:22:29,040
Speaker 4: They're just not using quantum computers. They're using GPUs, they're

448
00:22:29,080 --> 00:22:31,760
Speaker 4: using in video chips, they're using new algorithms. So I've

449
00:22:31,760 --> 00:22:34,320
Speaker 4: been very skeptical about that, also been very skeptical about

450
00:22:34,320 --> 00:22:37,200
Speaker 4: fusion for the same sort of reason that to your point,

451
00:22:37,240 --> 00:22:40,760
Speaker 4: there's like this ignorance arbitrage that people take advantage of.

452
00:22:40,800 --> 00:22:43,280
Speaker 4: They take advantage of that Investors don't fully understand something.

453
00:22:43,520 --> 00:22:45,440
Speaker 4: It's hot, it's on the front page of a newspaper

454
00:22:45,520 --> 00:22:48,679
Speaker 4: or magazine or you know whatever. It's a buzz and

455
00:22:48,720 --> 00:22:50,280
Speaker 4: they want to play in it, and so they invest

456
00:22:50,359 --> 00:22:52,760
Speaker 4: in it. And that's how you get frauds. So we're

457
00:22:52,760 --> 00:22:55,720
Speaker 4: always looking and basically trying to say, is this academic

458
00:22:55,800 --> 00:22:59,040
Speaker 4: practitioner commercially minded. They're probably not going to leave their jobs,

459
00:22:59,040 --> 00:23:00,760
Speaker 4: so we're only getting them twenty percent of their time.

460
00:23:00,880 --> 00:23:03,399
Speaker 4: Is the intellectual property reel. And then oftentimes to you're

461
00:23:03,400 --> 00:23:06,440
Speaker 4: pointing about the papers, there is a high reference ability

462
00:23:06,440 --> 00:23:09,680
Speaker 4: in papers. A paper that is cited enormously or immensely

463
00:23:09,960 --> 00:23:13,640
Speaker 4: has a lot more credibility because you have the vainglorious

464
00:23:13,960 --> 00:23:17,880
Speaker 4: error detection and correction of other scientists who are trying

465
00:23:17,880 --> 00:23:19,800
Speaker 4: to shoot that person down who has all the fame

466
00:23:19,840 --> 00:23:22,720
Speaker 4: for doing it, trying to seize that mantle. And so

467
00:23:23,240 --> 00:23:26,080
Speaker 4: scientists are not a benevelent bunch. They're just as competitive

468
00:23:26,080 --> 00:23:29,119
Speaker 4: as an investigative journalist trying to break the story or investor,

469
00:23:29,440 --> 00:23:31,359
Speaker 4: or an A and R REP for a music band

470
00:23:31,400 --> 00:23:34,080
Speaker 4: trying to get there before everybody else. And we're no different.

471
00:23:34,119 --> 00:23:36,800
Speaker 4: They're no different, But that's how all this stuff progresses.

472
00:23:37,960 --> 00:23:40,199
Speaker 1: Talk to us about the robotic arm. Yeah, I'm going

473
00:23:40,280 --> 00:23:40,840
Speaker 1: to take the bait.

474
00:23:41,040 --> 00:23:41,560
Speaker 3: Yeah.

475
00:23:41,640 --> 00:23:43,160
Speaker 4: The first point I would make is I don't believe

476
00:23:43,160 --> 00:23:45,760
Speaker 4: in these humanoid robot arms with like fingers and you know,

477
00:23:45,840 --> 00:23:48,879
Speaker 4: high dexterity. I think it's cool Parler trick, but I

478
00:23:48,920 --> 00:23:51,240
Speaker 4: just I think we should have robotic arms that look

479
00:23:51,320 --> 00:23:53,960
Speaker 4: more like Swiss army knives that are swapping and moving stuff.

480
00:23:53,960 --> 00:23:55,920
Speaker 4: And you can see a bunch of these things online.

481
00:23:56,000 --> 00:23:57,960
Speaker 4: You know, different tools for different tasks and be able

482
00:23:57,960 --> 00:24:01,280
Speaker 4: to do them instantaneously when you out to the industry structure.

483
00:24:01,600 --> 00:24:04,760
Speaker 4: You've got Fanik, which is a major Japanese player. They

484
00:24:04,800 --> 00:24:08,200
Speaker 4: started with industrial robots. They do factory automation, but they've

485
00:24:08,200 --> 00:24:10,720
Speaker 4: been a key player, probably a twenty five thirty billion

486
00:24:10,760 --> 00:24:15,000
Speaker 4: dollar enterprise value company. You've got ABB, which is Swiss

487
00:24:15,000 --> 00:24:19,240
Speaker 4: Swedish multinational industrial robot arms. Those are most of the

488
00:24:19,240 --> 00:24:22,200
Speaker 4: things that you would see in even a Tesla gigafactory

489
00:24:22,280 --> 00:24:24,520
Speaker 4: or something where Elon's talking about all their automated things

490
00:24:24,520 --> 00:24:26,919
Speaker 4: and they've got some of these ABB arms or the

491
00:24:26,960 --> 00:24:30,160
Speaker 4: other one is Kuka. Kuka which is a German company

492
00:24:30,520 --> 00:24:32,320
Speaker 4: and they were one of the great leaders. They got

493
00:24:32,320 --> 00:24:34,399
Speaker 4: bought by a Chinese company, I want to say twenty

494
00:24:34,440 --> 00:24:37,200
Speaker 4: sixteen or twenty seventeen, and China's made a bunch of

495
00:24:37,520 --> 00:24:41,440
Speaker 4: I think very smart investments acquiring technologies that were a

496
00:24:41,480 --> 00:24:43,840
Speaker 4: little bit before their time, And I think it presents

497
00:24:43,840 --> 00:24:46,640
Speaker 4: some geopolitical things that probably in five or ten years

498
00:24:46,680 --> 00:24:48,800
Speaker 4: will be looking at and saying, my gosh, how is

499
00:24:48,840 --> 00:24:53,399
Speaker 4: the dominant robot arms supplier or robot body supplier be

500
00:24:53,480 --> 00:24:57,439
Speaker 4: akin to like being dependent on TSMC, a Chinese company.

501
00:24:57,720 --> 00:25:02,879
Speaker 4: So I think there's going to be nash robot companies

502
00:25:03,000 --> 00:25:04,959
Speaker 4: that form in the same way you're starting to see

503
00:25:05,320 --> 00:25:07,000
Speaker 4: national AI companies form.

504
00:25:07,359 --> 00:25:11,160
Speaker 2: How does a company like Physical Intelligence, how are they

505
00:25:11,200 --> 00:25:14,440
Speaker 2: solving the data problem? Because like you said, there isn't

506
00:25:14,480 --> 00:25:17,000
Speaker 2: just the equivalent of a robot Internet where they can

507
00:25:17,040 --> 00:25:19,640
Speaker 2: watch millions of hours of a robot or I'm trying

508
00:25:19,680 --> 00:25:22,760
Speaker 2: to do something, or humans doing something or whatever. What

509
00:25:22,880 --> 00:25:24,919
Speaker 2: is the approach by which the sort of the token

510
00:25:25,000 --> 00:25:26,080
Speaker 2: problem is being solved?

511
00:25:26,359 --> 00:25:28,240
Speaker 4: I would call it the easy way to do it,

512
00:25:28,280 --> 00:25:30,959
Speaker 4: which is actually quite trivial and hard, is doing what

513
00:25:31,000 --> 00:25:33,720
Speaker 4: people originally did with robotic surgery. So we had a

514
00:25:33,720 --> 00:25:36,359
Speaker 4: company called Ors Surgical Robotics. We saw that JJ for

515
00:25:36,359 --> 00:25:40,880
Speaker 4: six billion dollars. It started with surgeons operating these things

516
00:25:41,240 --> 00:25:43,560
Speaker 4: in like a telerobot, so they had little pinchers on

517
00:25:43,600 --> 00:25:46,280
Speaker 4: their fingers and you know, from five feet away or

518
00:25:46,320 --> 00:25:49,280
Speaker 4: in a totally clean operating room. They were operating, but

519
00:25:49,320 --> 00:25:53,119
Speaker 4: it was their hands being teletransmitted to the device. And

520
00:25:53,200 --> 00:25:55,320
Speaker 4: so that is the first way which has come up

521
00:25:55,359 --> 00:25:59,600
Speaker 4: with one hundred different tasks, maybe the highest frequency things

522
00:25:59,760 --> 00:26:04,080
Speaker 4: like washing dishes, folding clothes, again unstructured environments, being able

523
00:26:04,080 --> 00:26:06,520
Speaker 4: to do them in multiple different houses, multiple different heights,

524
00:26:07,160 --> 00:26:09,920
Speaker 4: multiple different you know, wet clothes, dry clothes. Being able

525
00:26:09,960 --> 00:26:13,040
Speaker 4: to pour coffee, being able to have the dexterity to

526
00:26:13,080 --> 00:26:15,560
Speaker 4: open up a ca cup. I actually don't drink those.

527
00:26:15,560 --> 00:26:17,640
Speaker 4: I think they're disgusting, but you know, put it into

528
00:26:17,640 --> 00:26:18,480
Speaker 4: a coffee machine.

529
00:26:18,560 --> 00:26:19,520
Speaker 3: They don't drink any of that.

530
00:26:19,640 --> 00:26:23,760
Speaker 4: Slow and there it's an engineer that is operating them

531
00:26:24,280 --> 00:26:29,919
Speaker 4: and the movement of compensating for gravity, how much force,

532
00:26:30,040 --> 00:26:32,760
Speaker 4: how much tension, how much pressure. That's all information. It's

533
00:26:32,760 --> 00:26:35,800
Speaker 4: information that historically has not been captured and some of

534
00:26:35,840 --> 00:26:38,239
Speaker 4: that is then extensible. And this is a really cool thing.

535
00:26:38,240 --> 00:26:39,719
Speaker 4: You can see some of these robots. You might have

536
00:26:39,880 --> 00:26:42,560
Speaker 4: five different robots, but they have this amazing thing called

537
00:26:42,600 --> 00:26:45,960
Speaker 4: transfer learning. You teach one robot a thing, and suddenly

538
00:26:46,000 --> 00:26:48,840
Speaker 4: the other robot, which is disconnected, you know, to or

539
00:26:48,920 --> 00:26:51,560
Speaker 4: it's connected through the Internet through it, but it can

540
00:26:51,600 --> 00:26:54,680
Speaker 4: actually learn what that robot just learned and perform the task.

541
00:26:54,760 --> 00:26:57,359
Speaker 4: So that's actually pretty eerie and pretty cool. It's the

542
00:26:57,359 --> 00:26:59,919
Speaker 4: same sort of thing like if I had one roll

543
00:27:00,680 --> 00:27:03,160
Speaker 4: that saw where I tossed a ball in a room,

544
00:27:03,680 --> 00:27:06,040
Speaker 4: but three other robots didn't know, they would instantly know

545
00:27:06,440 --> 00:27:08,520
Speaker 4: because they have the eyes of the first robot. So

546
00:27:08,640 --> 00:27:11,320
Speaker 4: there's all kinds of training like that. Then there's schematics

547
00:27:11,320 --> 00:27:13,000
Speaker 4: and drawings. So I was alluding to this before of

548
00:27:13,040 --> 00:27:16,439
Speaker 4: like Ikea drawings, but once you can take diagrams and

549
00:27:16,440 --> 00:27:20,879
Speaker 4: schematics and actually use visual language models, it's something that

550
00:27:21,240 --> 00:27:23,439
Speaker 4: Opening Eyes a partner here with physical intelligence, and they

551
00:27:23,440 --> 00:27:25,719
Speaker 4: have pioneered some of that. That's going to be wild too,

552
00:27:25,760 --> 00:27:29,199
Speaker 4: where you can literally just show an Ikea drawing and

553
00:27:29,240 --> 00:27:32,960
Speaker 4: the robot goes with a constraint set of limited pieces

554
00:27:32,960 --> 00:27:35,200
Speaker 4: that they have to put together, a set of screws

555
00:27:35,520 --> 00:27:38,840
Speaker 4: and wrenches and whatnot, and they can completely assemble whatever

556
00:27:38,880 --> 00:27:41,720
Speaker 4: it is, a nursery thing or furniture or desk, which

557
00:27:41,720 --> 00:27:43,080
Speaker 4: I think is going to be pretty wild for people

558
00:27:43,080 --> 00:27:43,479
Speaker 4: to see too.

559
00:27:43,720 --> 00:27:46,040
Speaker 1: That would actually be a major quality of life cure,

560
00:27:46,720 --> 00:27:49,840
Speaker 1: not having to put together ikea furniture. That'd be amazing.

561
00:27:50,359 --> 00:27:53,040
Speaker 1: Could you talk a little bit more about how robots

562
00:27:53,200 --> 00:27:55,880
Speaker 1: are learning and I guess, like what the different types

563
00:27:56,200 --> 00:27:59,880
Speaker 1: of learning are or different patterns of learning, and what

564
00:28:00,119 --> 00:28:02,360
Speaker 1: you've seen that's been most promising so far.

565
00:28:02,920 --> 00:28:05,680
Speaker 4: You have two main categories or maybe three. You've got

566
00:28:05,680 --> 00:28:11,320
Speaker 4: supervised learning, so there you've got input data. The robot

567
00:28:11,400 --> 00:28:14,959
Speaker 4: is learning, it's being corrected, it's being told in some

568
00:28:15,000 --> 00:28:18,040
Speaker 4: cases like I described before, whether it's voice or a

569
00:28:18,119 --> 00:28:22,240
Speaker 4: gesture or nudge. Then you have unsupervised learning, where the

570
00:28:22,320 --> 00:28:26,080
Speaker 4: robots are basically training on unstructured data. They get to

571
00:28:26,080 --> 00:28:30,200
Speaker 4: discover patterns, they encounter the world, they encounter boundaries, gravity,

572
00:28:30,400 --> 00:28:32,960
Speaker 4: those kinds of things, and it might be slower in

573
00:28:33,000 --> 00:28:36,880
Speaker 4: that case, but they're reducing the dimensions for error. There's

574
00:28:36,920 --> 00:28:40,160
Speaker 4: something that I coined this term. I call it MBTFU,

575
00:28:40,240 --> 00:28:43,680
Speaker 4: which is meantime between f ups. But you want that

576
00:28:43,720 --> 00:28:45,600
Speaker 4: to be as long as possible. If you go back

577
00:28:45,640 --> 00:28:48,720
Speaker 4: to like the early rumbas, you know, a roomba would

578
00:28:48,760 --> 00:28:51,280
Speaker 4: not know if it was cleaning up a spill of

579
00:28:51,320 --> 00:28:53,760
Speaker 4: chocolate milk, or if your dog made a mess and

580
00:28:53,840 --> 00:28:56,320
Speaker 4: it smears. It like to tell all over your floor, right.

581
00:28:56,520 --> 00:28:59,520
Speaker 4: You want to increase as long as possible the meantime

582
00:28:59,560 --> 00:29:02,840
Speaker 4: between me, you know, basically error reduction. Men, you've got

583
00:29:02,880 --> 00:29:06,520
Speaker 4: reinforcement learning, imitation learning, where you might be controlling the

584
00:29:06,600 --> 00:29:09,720
Speaker 4: robot or having it mimic you. There's this idea of

585
00:29:09,720 --> 00:29:12,040
Speaker 4: transfer learning, where a single robot learned something, but it

586
00:29:12,040 --> 00:29:15,720
Speaker 4: can transfer it to different robots or from a different domain.

587
00:29:16,240 --> 00:29:19,000
Speaker 4: So people are trying lots of different approaches, and the

588
00:29:19,000 --> 00:29:22,920
Speaker 4: more different mechanisms you have, people are then going to

589
00:29:22,960 --> 00:29:25,280
Speaker 4: figure out, okay, which is the least data intensive, or

590
00:29:25,280 --> 00:29:27,840
Speaker 4: which has the lowest latency and is the quickest, or

591
00:29:28,120 --> 00:29:31,560
Speaker 4: which is the best system for training a robot that

592
00:29:32,400 --> 00:29:36,160
Speaker 4: you put it in a totally unstructured environment and without any training,

593
00:29:36,200 --> 00:29:38,000
Speaker 4: sort of what they call it zero shot learning. It's

594
00:29:38,000 --> 00:29:41,400
Speaker 4: able to figure out from prior knowledge. I know that

595
00:29:41,440 --> 00:29:44,560
Speaker 4: I can go through that chair. I know that it swivels,

596
00:29:44,600 --> 00:29:46,640
Speaker 4: I have to turn it this way. I know how

597
00:29:46,720 --> 00:29:49,200
Speaker 4: much force I need to pick up a general coke

598
00:29:49,280 --> 00:29:54,120
Speaker 4: can and those kinds of things. And I think, again,

599
00:29:54,160 --> 00:29:57,520
Speaker 4: it's going to be enlightening of how much we as

600
00:29:57,560 --> 00:30:01,880
Speaker 4: we navigate any given minute in our life. Take for granted,

601
00:30:02,040 --> 00:30:05,680
Speaker 4: all this intuitive tacit knowledge that we have about the

602
00:30:05,680 --> 00:30:09,480
Speaker 4: physical world, and there really is this intuitive physics of

603
00:30:09,800 --> 00:30:11,800
Speaker 4: how we move around. Robots are going to learn.

604
00:30:11,680 --> 00:30:15,440
Speaker 2: That is really simple question. In the next five years,

605
00:30:15,880 --> 00:30:18,400
Speaker 2: is it plausible that I'll have a robot in my

606
00:30:18,480 --> 00:30:20,320
Speaker 2: house where I could take all the clothes out of

607
00:30:20,360 --> 00:30:23,120
Speaker 2: the dryer, drop it into something and have it turned

608
00:30:23,160 --> 00:30:27,480
Speaker 2: into folded clothes. And or if not that, what could

609
00:30:27,560 --> 00:30:30,920
Speaker 2: be that chat gpt of robots that's right around the corner.

610
00:30:31,840 --> 00:30:34,240
Speaker 4: Well, one thing which I posited to the team because

611
00:30:34,280 --> 00:30:36,720
Speaker 4: I was thinking about exactly that. You know, what would

612
00:30:36,720 --> 00:30:38,920
Speaker 4: be a really cool thing. I lose stuff in our

613
00:30:38,960 --> 00:30:41,280
Speaker 4: apartment all the time. We have a bunch of different rooms.

614
00:30:41,920 --> 00:30:44,960
Speaker 4: I would love to basically say, has anybody seen which

615
00:30:45,000 --> 00:30:46,720
Speaker 4: is often what I do to my wife and three kids?

616
00:30:46,760 --> 00:30:49,920
Speaker 4: Has anybody seen my wallet? Has anybody seen my glasses? Okay,

617
00:30:50,360 --> 00:30:53,840
Speaker 4: just announcing that you can see that a robot with

618
00:30:53,920 --> 00:30:55,760
Speaker 4: a series of things in your home that would have

619
00:30:55,840 --> 00:30:58,720
Speaker 4: visual identifiers and visual learning machine learning to be able

620
00:30:58,760 --> 00:31:03,440
Speaker 4: to spot object in a video frame could say yes, Josh,

621
00:31:03,480 --> 00:31:06,120
Speaker 4: I know exactly where they are and go and retrieve them.

622
00:31:06,360 --> 00:31:09,040
Speaker 4: So fetch and retrieving objects in the home to me

623
00:31:09,640 --> 00:31:11,480
Speaker 4: would be a pretty cool thing. Where'd I put that

624
00:31:11,520 --> 00:31:14,160
Speaker 4: remote or whatever? And the robot basically knows because they

625
00:31:14,200 --> 00:31:16,520
Speaker 4: can go through the DVR of the home and they

626
00:31:16,520 --> 00:31:18,040
Speaker 4: basically know where it is and they can fetch it

627
00:31:18,040 --> 00:31:20,720
Speaker 4: and retrieve it with the right physics. Folding laundry. I

628
00:31:20,720 --> 00:31:23,160
Speaker 4: don't know how to handicap that, and we can do

629
00:31:23,240 --> 00:31:24,520
Speaker 4: that now pretty.

630
00:31:24,160 --> 00:31:26,960
Speaker 2: Crudely through That's the one I want because I have

631
00:31:27,040 --> 00:31:28,960
Speaker 2: a swall in New York City apartment. I don't lose

632
00:31:29,000 --> 00:31:31,680
Speaker 2: things like too much. Basically, like, I need that fold

633
00:31:31,760 --> 00:31:32,880
Speaker 2: I need that folding robot.

634
00:31:32,960 --> 00:31:34,520
Speaker 4: What are you gonna pay for that? Though? I don't know.

635
00:31:34,600 --> 00:31:37,320
Speaker 4: You know, would you pay five grand ten grand for

636
00:31:37,640 --> 00:31:40,840
Speaker 4: a robot that folded your clothes? Probably not. So it's

637
00:31:40,920 --> 00:31:42,640
Speaker 4: it's gonna have a high. That's why most of these

638
00:31:42,680 --> 00:31:46,880
Speaker 4: things have found their way into industrial uses first, and

639
00:31:46,960 --> 00:31:48,880
Speaker 4: over time they'll get cheaper and cheaper, they'll get better

640
00:31:48,960 --> 00:31:51,120
Speaker 4: and better. But you know, look, I'm one of the

641
00:31:51,160 --> 00:31:54,280
Speaker 4: few people that have an Amazon astro It's a robot

642
00:31:54,320 --> 00:31:56,840
Speaker 4: that you know, rolls around the house and you can

643
00:31:56,840 --> 00:31:58,480
Speaker 4: tell to go into a room. You can put something

644
00:31:58,520 --> 00:32:00,160
Speaker 4: in the back of it and take it to It

645
00:32:00,200 --> 00:32:02,640
Speaker 4: can do facial identification for my family, so I can

646
00:32:02,720 --> 00:32:06,040
Speaker 4: say where's quinner body, and they'll find my younger kids

647
00:32:06,080 --> 00:32:09,760
Speaker 4: and I can send a message noise the hell out

648
00:32:09,800 --> 00:32:11,400
Speaker 4: of my wife. But I think it's sort of cool,

649
00:32:11,400 --> 00:32:13,160
Speaker 4: and yeah, every robot that comes out, I'll be an

650
00:32:13,160 --> 00:32:16,240
Speaker 4: early adoptor, and yeah, we like funny things.

651
00:32:33,560 --> 00:32:36,440
Speaker 1: So one of the interesting things about the current tech

652
00:32:36,560 --> 00:32:39,840
Speaker 1: cycle and all this enthusiasm for AI is that so

653
00:32:40,040 --> 00:32:43,680
Speaker 1: far it's been the big incumbents who seem to be

654
00:32:43,720 --> 00:32:47,040
Speaker 1: winning here. And part of that is because the capital

655
00:32:47,080 --> 00:32:50,440
Speaker 1: investment needed is so large, the amount of data needed

656
00:32:50,680 --> 00:32:53,360
Speaker 1: is so large when it comes to robotics, would you

657
00:32:53,400 --> 00:32:56,280
Speaker 1: expect to see a similar thing? And then added on

658
00:32:56,360 --> 00:32:59,280
Speaker 1: to that, could you have a situation where, you know,

659
00:32:59,360 --> 00:33:03,120
Speaker 1: if you are a large manufacturer, or perhaps you are

660
00:33:03,560 --> 00:33:07,040
Speaker 1: a company that has a lot of proprietary data, like

661
00:33:07,160 --> 00:33:11,480
Speaker 1: an insurance company or a financial company or something like that,

662
00:33:11,840 --> 00:33:14,240
Speaker 1: could you develop your own robotics Like would that be

663
00:33:14,320 --> 00:33:17,840
Speaker 1: your edge here? And could you potentially just do it yourself?

664
00:33:18,480 --> 00:33:21,880
Speaker 4: On the first case, if I look at the current landscape,

665
00:33:22,320 --> 00:33:24,800
Speaker 4: the one company in sort of the big magnificent seven

666
00:33:25,080 --> 00:33:27,760
Speaker 4: would be Amazon, just because they have already been investing

667
00:33:27,800 --> 00:33:31,200
Speaker 4: in this and for a long time. Jeff Bezos is

668
00:33:31,280 --> 00:33:34,440
Speaker 4: very passionate about robots. So I think there's a DNA

669
00:33:34,560 --> 00:33:38,520
Speaker 4: there where doing things that enable them to do the

670
00:33:38,520 --> 00:33:41,240
Speaker 4: three things that Jeff loves to do historically, which is

671
00:33:41,560 --> 00:33:46,120
Speaker 4: increased choice and availability, increase convenience for customers and lower prices,

672
00:33:46,160 --> 00:33:51,000
Speaker 4: and factory automation, warehouses delivery. Even when they bought our

673
00:33:51,000 --> 00:33:53,400
Speaker 4: company for a little over a billion dollars called Zookes

674
00:33:53,440 --> 00:33:55,800
Speaker 4: that was doing autonomous driving, they have a long term

675
00:33:55,800 --> 00:33:57,800
Speaker 4: intention to be able to do twenty four to seven

676
00:33:57,880 --> 00:34:00,520
Speaker 4: right hand turn lanes, navigating around the city with a

677
00:34:00,600 --> 00:34:03,200
Speaker 4: human that's basically delivering last miles kind of stuff. And

678
00:34:04,000 --> 00:34:07,720
Speaker 4: so so I could see Amazon doing that. Microsoft, I

679
00:34:07,720 --> 00:34:10,000
Speaker 4: don't really see them getting into robotics in a significant way.

680
00:34:10,000 --> 00:34:12,360
Speaker 4: They have some R and D efforts Google and we

681
00:34:12,400 --> 00:34:14,400
Speaker 4: talked a little bit about this phenomenon, but they are

682
00:34:14,480 --> 00:34:17,279
Speaker 4: the third or fourth group where we have taken a

683
00:34:17,400 --> 00:34:19,560
Speaker 4: team out of a big tech We did it with

684
00:34:19,680 --> 00:34:23,320
Speaker 4: Google with a company called Osmo to create basically Shazam

685
00:34:23,400 --> 00:34:29,000
Speaker 4: for smell. We did it with a BIOAI company called

686
00:34:29,040 --> 00:34:31,240
Speaker 4: Evolutionary Scale out of meta that's going to be announced

687
00:34:31,239 --> 00:34:33,480
Speaker 4: more publicly soon. And then we did it with this

688
00:34:33,520 --> 00:34:36,040
Speaker 4: team out of Google that became a physical intelligence between

689
00:34:36,040 --> 00:34:38,800
Speaker 4: Google deep Mind and open AI and Stanford and Berkeley.

690
00:34:39,120 --> 00:34:41,080
Speaker 4: So I think that this is going to be more

691
00:34:41,200 --> 00:34:44,120
Speaker 4: the startups. The beneficiaries of the money will still be

692
00:34:44,280 --> 00:34:46,200
Speaker 4: Nvidia and some of the chip players, some of the

693
00:34:46,239 --> 00:34:48,319
Speaker 4: hardware providers, because we need the hardware to be able

694
00:34:48,320 --> 00:34:50,400
Speaker 4: to train the robots. But I really think that this

695
00:34:50,560 --> 00:34:52,719
Speaker 4: is an open field. And again just going back to

696
00:34:52,760 --> 00:34:54,719
Speaker 4: that stat of how many large language model there are

697
00:34:54,800 --> 00:34:57,880
Speaker 4: for chatbots and how few there are for robots. I

698
00:34:57,920 --> 00:35:00,239
Speaker 4: think it's a big opportunity. Now five years from now

699
00:35:00,280 --> 00:35:02,160
Speaker 4: we might have a bubble in robots, but today I

700
00:35:02,200 --> 00:35:05,439
Speaker 4: think it's it's it's a really exciting field, just.

701
00:35:05,360 --> 00:35:09,560
Speaker 2: An existing AI and competitive advantage. So when I first

702
00:35:09,560 --> 00:35:13,640
Speaker 2: in the year two thousand, I think encountered Google, I

703
00:35:13,760 --> 00:35:16,640
Speaker 2: used it. I was like, this is way better than

704
00:35:16,719 --> 00:35:19,239
Speaker 2: Yahoo or anything else. I never like stopped after that.

705
00:35:19,320 --> 00:35:22,040
Speaker 2: Again with chat bots, just going back now, I even

706
00:35:22,040 --> 00:35:24,520
Speaker 2: talking about robotics here. It's like I and I like

707
00:35:24,680 --> 00:35:27,960
Speaker 2: got a chet GPT progount or whatever early on, and

708
00:35:28,000 --> 00:35:30,040
Speaker 2: I thought it was pretty cool and I like use

709
00:35:30,120 --> 00:35:32,640
Speaker 2: it for some stuff. And then like the new version

710
00:35:32,680 --> 00:35:35,040
Speaker 2: of Claude came out from Anthropic and I was like, oh,

711
00:35:35,080 --> 00:35:36,600
Speaker 2: this is actually kind of cool and I kind of

712
00:35:36,640 --> 00:35:38,080
Speaker 2: like it more. I don't really know why I like

713
00:35:38,120 --> 00:35:39,719
Speaker 2: it more, but for some reason, I like it more

714
00:35:39,920 --> 00:35:42,600
Speaker 2: and I had no problem switching over. Could it be

715
00:35:42,719 --> 00:35:45,840
Speaker 2: possible that some of these like core like models do

716
00:35:46,000 --> 00:35:49,000
Speaker 2: not prove to be as sticky or have as deep

717
00:35:49,080 --> 00:35:50,600
Speaker 2: moat as people expect.

718
00:35:50,880 --> 00:35:54,200
Speaker 4: Totally Inflection, which launched there is called PI just wasn't

719
00:35:54,239 --> 00:35:57,560
Speaker 4: that good. They've now gone to Microsoft Anthropic. Yeah, at

720
00:35:57,560 --> 00:36:00,000
Speaker 4: first was a little bit behind. They are the most

721
00:36:00,080 --> 00:36:02,120
Speaker 4: formative model. They're the best performing one. They're the one

722
00:36:02,120 --> 00:36:05,080
Speaker 4: that I also like, you use the most. Why it's fastest,

723
00:36:05,440 --> 00:36:07,440
Speaker 4: it has a little bit more.

724
00:36:08,160 --> 00:36:09,640
Speaker 3: It seems to talk a little bitter.

725
00:36:10,560 --> 00:36:13,240
Speaker 4: Yeah, but here's an interesting thing to your point about

726
00:36:13,239 --> 00:36:16,760
Speaker 4: you know, sort of motes and Google early on most

727
00:36:16,800 --> 00:36:20,080
Speaker 4: searches for Google, and remember Google's two hundred and eighty

728
00:36:20,160 --> 00:36:25,719
Speaker 4: billion dollars add revenue generated, you know, enormous Google. Most

729
00:36:25,719 --> 00:36:28,160
Speaker 4: of those are like five to ten word searches, right,

730
00:36:28,200 --> 00:36:29,600
Speaker 4: so you put something in your like, I don't know,

731
00:36:29,719 --> 00:36:34,759
Speaker 4: restaurant in West Village or you know whatever. Okay, Perplexity,

732
00:36:34,800 --> 00:36:36,839
Speaker 4: which I don't know if you've used, and we met

733
00:36:36,840 --> 00:36:39,560
Speaker 4: the founder early on and we ended up not investing

734
00:36:39,560 --> 00:36:41,279
Speaker 4: and it was probably a miss for us, but the

735
00:36:41,400 --> 00:36:43,160
Speaker 4: founder they are focused. You know what I'm going to

736
00:36:43,239 --> 00:36:45,600
Speaker 4: do the point two percent of searches or the one

737
00:36:45,680 --> 00:36:48,160
Speaker 4: percent of searches that are longer formed where people really

738
00:36:48,160 --> 00:36:50,840
Speaker 4: want to ask a full question. And I see a

739
00:36:50,840 --> 00:36:52,520
Speaker 4: lot of people that are using it and it's not

740
00:36:52,560 --> 00:36:55,759
Speaker 4: coming up with some Wolke answer or some pithy, you know,

741
00:36:55,960 --> 00:37:00,480
Speaker 4: rock like response. It's actually a well researched, footnoted, sources,

742
00:37:00,880 --> 00:37:04,239
Speaker 4: cited response. So to me, the two things that I

743
00:37:04,320 --> 00:37:06,640
Speaker 4: use most on the chap outside right now are clawed

744
00:37:06,880 --> 00:37:10,440
Speaker 4: from Anthropic and Perplexity. But in six months there might

745
00:37:10,480 --> 00:37:15,120
Speaker 4: be something totally different. And the sheer number of models

746
00:37:15,520 --> 00:37:18,239
Speaker 4: that are available on open source. Who knows what Apple

747
00:37:18,320 --> 00:37:21,040
Speaker 4: ends up doing here and that being integrated. You know,

748
00:37:21,120 --> 00:37:25,000
Speaker 4: Series sucked, but so did Alexa. Amazon's making moves Apple

749
00:37:25,000 --> 00:37:28,839
Speaker 4: will too. So yeah, but you know, use you mount

750
00:37:28,840 --> 00:37:31,280
Speaker 4: for a second. All of this stuff and Venture Venture

751
00:37:31,360 --> 00:37:35,080
Speaker 4: is going through this downturn and the one area where

752
00:37:35,160 --> 00:37:38,160
Speaker 4: there are very high valuations and a lot of money

753
00:37:38,160 --> 00:37:40,960
Speaker 4: flowing and a lot of talent is been in AI,

754
00:37:41,160 --> 00:37:43,520
Speaker 4: and therefore the future returns on these things are going

755
00:37:43,560 --> 00:37:45,840
Speaker 4: to be lower. And that's why we at Lucks decided,

756
00:37:45,960 --> 00:37:47,840
Speaker 4: you know, we're going to focus on AI in the

757
00:37:47,840 --> 00:37:50,040
Speaker 4: physical world. Having done all the stuff over the past

758
00:37:50,080 --> 00:37:52,759
Speaker 4: five years, I would say there's this five year psychological bias.

759
00:37:52,760 --> 00:37:54,600
Speaker 4: Everybody wants to be invested today where you should have

760
00:37:54,600 --> 00:37:57,600
Speaker 4: been five years ago, So hugging face and Mosaic and

761
00:37:57,760 --> 00:37:59,400
Speaker 4: which we sold the data bricks. We were in that

762
00:37:59,400 --> 00:38:02,640
Speaker 4: five years ago. Today we're really interested in biology and

763
00:38:02,760 --> 00:38:05,400
Speaker 4: robotics and AIS use in those. And then I have

764
00:38:05,440 --> 00:38:08,319
Speaker 4: a really weird theme which is so sexy because it's

765
00:38:08,440 --> 00:38:11,680
Speaker 4: unsexy to me and to others. You understand accounting, vast

766
00:38:11,719 --> 00:38:14,279
Speaker 4: majority of venture investors and make startup people don't. But

767
00:38:14,560 --> 00:38:17,880
Speaker 4: do you take capex. Capex is made up of two pieces,

768
00:38:18,040 --> 00:38:22,080
Speaker 4: growth and maintenance. And everybody's been funding growth, growth, growth, growth, growth,

769
00:38:22,120 --> 00:38:26,280
Speaker 4: investor growth. So I got interested in maintenance. Why because

770
00:38:26,320 --> 00:38:31,120
Speaker 4: you have trillions of dollars of assets infrastructure, hospital systems,

771
00:38:31,320 --> 00:38:35,480
Speaker 4: energy systems, buildings that need to be maintained and every

772
00:38:35,560 --> 00:38:38,480
Speaker 4: generation and every new startup and every new investor always

773
00:38:38,480 --> 00:38:40,120
Speaker 4: wants to do the new new thing. It's why we

774
00:38:40,120 --> 00:38:41,640
Speaker 4: get new music, and we get new food, and we

775
00:38:41,640 --> 00:38:44,920
Speaker 4: get new fashion. But there's all these neglected assets, and

776
00:38:44,960 --> 00:38:48,800
Speaker 4: I think you can apply new technology to maintaining these systems.

777
00:38:48,840 --> 00:38:52,160
Speaker 4: And so I've become obsessed with this unsexy theme of maintenance,

778
00:38:52,640 --> 00:38:54,160
Speaker 4: which I think is going to become a hot area

779
00:38:54,200 --> 00:38:55,000
Speaker 4: over the next few years.

780
00:38:55,120 --> 00:38:58,400
Speaker 1: Well, you mean maintenance of physical infrastructure. So the idea

781
00:38:58,440 --> 00:38:59,920
Speaker 1: that you could have, I don't know, a little row

782
00:39:00,080 --> 00:39:02,920
Speaker 1: bought that goes around your factory or a bunch of

783
00:39:03,040 --> 00:39:06,680
Speaker 1: highways and sort of surveys it for cracks or things

784
00:39:06,719 --> 00:39:08,839
Speaker 1: that it thinks needs to be fixed.

785
00:39:08,680 --> 00:39:12,880
Speaker 4: Totally, could be infrastructure for transportation, it could be inside

786
00:39:12,920 --> 00:39:15,719
Speaker 4: of hospitals where there are road routine things. And it's

787
00:39:15,760 --> 00:39:18,799
Speaker 4: also oddly, and I know you guys have covered this,

788
00:39:18,920 --> 00:39:22,200
Speaker 4: but the idea that AI is really coming for the

789
00:39:22,239 --> 00:39:25,040
Speaker 4: white collar workers. You know, you joke that you could

790
00:39:25,040 --> 00:39:27,040
Speaker 4: talk about AI and generate a script, you know, based

791
00:39:27,040 --> 00:39:27,520
Speaker 4: on AI.

792
00:39:27,880 --> 00:39:30,280
Speaker 1: But oh no, it wasn't a joke, very serious.

793
00:39:31,360 --> 00:39:34,480
Speaker 4: They always thought that they were relatively insulated and it

794
00:39:34,520 --> 00:39:36,480
Speaker 4: was the blue collar workers. But let me tell you

795
00:39:36,560 --> 00:39:38,480
Speaker 4: the guy that put me in business, Bill Conway, the

796
00:39:38,880 --> 00:39:41,680
Speaker 4: founder of the Carlile Group. He's spending all his philanthropic money,

797
00:39:41,960 --> 00:39:44,640
Speaker 4: or a significant portion of it, funding nursing schools. Why

798
00:39:44,680 --> 00:39:48,440
Speaker 4: because he identified a very high magnitude impact. Because we

799
00:39:48,480 --> 00:39:51,359
Speaker 4: have such a shortage of nurses in this country. That's

800
00:39:51,360 --> 00:39:54,680
Speaker 4: an opportunity for maintenance where robots and technology can play

801
00:39:54,680 --> 00:39:58,040
Speaker 4: a role. How do you augment and help nurses plummers.

802
00:39:58,080 --> 00:40:00,920
Speaker 4: We have a massive shortage of plumbers this country, and

803
00:40:01,000 --> 00:40:03,520
Speaker 4: so I actually think that the blue collar workers, empowered

804
00:40:03,520 --> 00:40:07,000
Speaker 4: by technology and maintaining all of these systems around us,

805
00:40:07,400 --> 00:40:09,360
Speaker 4: are actually going to be a winning combination.

806
00:40:09,520 --> 00:40:12,040
Speaker 2: I want to talk about another aspect of I guess

807
00:40:12,160 --> 00:40:15,640
Speaker 2: AI investing, which is that in this sort of the

808
00:40:15,760 --> 00:40:19,360
Speaker 2: SaaS wave, the twenty tens a decade like compute was

809
00:40:19,520 --> 00:40:22,359
Speaker 2: very cheap, right, and so basically like that part, you're

810
00:40:22,360 --> 00:40:25,320
Speaker 2: like plug into aws and it's sort of yeah, I know,

811
00:40:25,360 --> 00:40:27,640
Speaker 2: it probably costs some money, but is not a big

812
00:40:27,800 --> 00:40:29,520
Speaker 2: lineup or a line item.

813
00:40:29,640 --> 00:40:32,719
Speaker 3: Ultimately for a lot of these companies, How does that change.

814
00:40:32,400 --> 00:40:34,719
Speaker 2: In twenty twenty four when you're dealing with an AI

815
00:40:34,840 --> 00:40:39,319
Speaker 2: company and electricity bills exist or hardware accumulation, depending on

816
00:40:39,560 --> 00:40:42,560
Speaker 2: where they are in the stack. How is an investor

817
00:40:42,880 --> 00:40:45,080
Speaker 2: do you think about like the changing I guess people

818
00:40:45,080 --> 00:40:47,799
Speaker 2: will talk about like you know, shifting, you know, having

819
00:40:47,800 --> 00:40:50,319
Speaker 2: to spend more on capex versus op X relative to

820
00:40:50,360 --> 00:40:53,720
Speaker 2: the prior generation of tech startups. How does that play

821
00:40:53,719 --> 00:40:55,920
Speaker 2: out in the investments you choose?

822
00:40:56,239 --> 00:40:58,480
Speaker 4: It's a great question in the AI world, and then

823
00:40:58,520 --> 00:41:00,799
Speaker 4: I'll give you the biology world. On the AI side,

824
00:41:00,800 --> 00:41:02,920
Speaker 4: you know, take opening Eye. These are all rumored numbers,

825
00:41:02,960 --> 00:41:05,799
Speaker 4: not you know, nothing's fully confirmed, but two billion, maybe

826
00:41:05,840 --> 00:41:08,520
Speaker 4: three billion of revenue. I think about ten million people

827
00:41:08,640 --> 00:41:11,799
Speaker 4: paying twenty bucks a month or thereabout, and probably one

828
00:41:11,840 --> 00:41:13,640
Speaker 4: hundred million users. You know, I don't know how many

829
00:41:13,640 --> 00:41:15,960
Speaker 4: of those are unique, but they're not making money on that.

830
00:41:15,960 --> 00:41:18,440
Speaker 4: They're losing several billion dollars today because you have these

831
00:41:18,520 --> 00:41:21,120
Speaker 4: upfront costs, big cap X, a lot of training, you know,

832
00:41:21,120 --> 00:41:23,280
Speaker 4: and then you try to maybe do some big enterprise deals.

833
00:41:23,719 --> 00:41:26,520
Speaker 4: A company like hugging Face is profitable because they're not

834
00:41:26,560 --> 00:41:28,560
Speaker 4: doing they're just hosting it and you know, letting people

835
00:41:28,640 --> 00:41:30,960
Speaker 4: run in frints and then charging and making margin on

836
00:41:31,000 --> 00:41:34,120
Speaker 4: that kind of stuff. So that to me is interesting

837
00:41:34,160 --> 00:41:36,800
Speaker 4: of the people that spend a ton of money and

838
00:41:36,960 --> 00:41:39,319
Speaker 4: they've got to earn it back. And can you get

839
00:41:39,360 --> 00:41:42,200
Speaker 4: pricing power by going from twenty bucks a month to

840
00:41:42,200 --> 00:41:44,239
Speaker 4: thirty bucks a month? And maybe you get that because

841
00:41:44,280 --> 00:41:47,000
Speaker 4: now you have open AI premium where you have access

842
00:41:47,040 --> 00:41:49,399
Speaker 4: to say Sora for video generation or something like that.

843
00:41:49,640 --> 00:41:51,359
Speaker 4: So that's going to be a big question on are

844
00:41:51,400 --> 00:41:53,719
Speaker 4: these profitable investments? Not? Are they cool? Not? Are they

845
00:41:53,760 --> 00:41:57,240
Speaker 4: world changing? Absolutely, but are they profitable investments? And look,

846
00:41:57,520 --> 00:41:59,520
Speaker 4: the market may not care if they're profitable. Market funds

847
00:41:59,520 --> 00:42:01,279
Speaker 4: all kinds of profitable thing that they believe in the

848
00:42:01,360 --> 00:42:04,360
Speaker 4: narrative in the story. But thinking about fundamental businesses and

849
00:42:04,400 --> 00:42:07,279
Speaker 4: the economic changes between Capeck and op X, I think

850
00:42:07,280 --> 00:42:09,799
Speaker 4: in AI it's very hard if you are building out

851
00:42:09,800 --> 00:42:11,759
Speaker 4: your data centers, trying to do your own training, your

852
00:42:11,800 --> 00:42:15,400
Speaker 4: own inference, hosting these models, it's very hard. Biology we

853
00:42:15,480 --> 00:42:19,200
Speaker 4: will see an AWS moment where instead of you having

854
00:42:19,239 --> 00:42:22,160
Speaker 4: to be a biotech firm that opens your own wet

855
00:42:22,239 --> 00:42:26,239
Speaker 4: lab or moves into Alexandria real estate which is specialized

856
00:42:26,280 --> 00:42:29,560
Speaker 4: in hosting biotech companies in all these different regions, approximate

857
00:42:29,680 --> 00:42:32,799
Speaker 4: to academic research centers. You will be able to just

858
00:42:32,840 --> 00:42:35,719
Speaker 4: take your experiment and upload it to the cloud, where

859
00:42:35,760 --> 00:42:39,920
Speaker 4: there are cloud based robotic labs. We funded some of these.

860
00:42:40,400 --> 00:42:43,120
Speaker 4: There's one company called Stratios. There's a ton that are

861
00:42:43,120 --> 00:42:45,080
Speaker 4: going to come on wave. And this is exciting because

862
00:42:45,120 --> 00:42:47,440
Speaker 4: you can be a scientist on the beach in the Bahamas,

863
00:42:47,480 --> 00:42:50,400
Speaker 4: pull up your iPad, run an experiment. The robots are

864
00:42:50,400 --> 00:42:53,920
Speaker 4: performing ninety percent of the activity of pouring something from

865
00:42:53,960 --> 00:42:57,160
Speaker 4: a beaker into another, running a centrifuge, and then the

866
00:42:57,239 --> 00:42:58,759
Speaker 4: data that comes off of that, and this is the

867
00:42:58,800 --> 00:43:02,040
Speaker 4: really cool part. Then the robot and the machines will

868
00:43:02,040 --> 00:43:04,600
Speaker 4: actually say to you, hey, do you want to run

869
00:43:04,600 --> 00:43:07,680
Speaker 4: this experiment, but change these four parameters or these variables,

870
00:43:07,760 --> 00:43:09,600
Speaker 4: and you just click a button yes, as though it's

871
00:43:09,640 --> 00:43:13,120
Speaker 4: reverse prompting you, and then you run another experiment. So

872
00:43:13,200 --> 00:43:17,920
Speaker 4: the implication here is that the boost in productivity for science,

873
00:43:18,280 --> 00:43:22,120
Speaker 4: for generation of truth, of new information, of new knowledge,

874
00:43:22,360 --> 00:43:24,279
Speaker 4: that to me is the most exciting thing. And the

875
00:43:24,320 --> 00:43:27,680
Speaker 4: companies that capture that, forget about the societal dividend, I think,

876
00:43:27,680 --> 00:43:28,800
Speaker 4: are going to make a lot of money.

877
00:43:29,040 --> 00:43:32,040
Speaker 1: Yeah, this actually reminds me of the conversation that we

878
00:43:32,200 --> 00:43:36,200
Speaker 1: had regarding snack food innovation and this idea that you

879
00:43:36,239 --> 00:43:39,799
Speaker 1: can use a sort of factorio like simulation just to

880
00:43:39,880 --> 00:43:43,680
Speaker 1: run new processes through your factory and see how they

881
00:43:43,680 --> 00:43:46,000
Speaker 1: would actually work out and what the supply chain might

882
00:43:46,040 --> 00:43:48,480
Speaker 1: look like. But not to give in to my five

883
00:43:48,560 --> 00:43:52,719
Speaker 1: year bias too much and overly focus on chat GPT.

884
00:43:53,200 --> 00:43:56,160
Speaker 1: But where are we in terms of context window expansion,

885
00:43:56,200 --> 00:43:58,879
Speaker 1: because this is something we spoke about with you last year,

886
00:43:59,160 --> 00:44:01,120
Speaker 1: and I think for a lot of people it's probably

887
00:44:01,120 --> 00:44:04,680
Speaker 1: one of the overriding annoyances with something like chat GPT,

888
00:44:04,840 --> 00:44:07,440
Speaker 1: the fact that you can't actually copy and paste that

889
00:44:07,560 --> 00:44:10,080
Speaker 1: much text into it, and that you are limited in

890
00:44:10,160 --> 00:44:12,960
Speaker 1: terms of the output that it actually gives you. Have

891
00:44:13,040 --> 00:44:15,680
Speaker 1: there been major advancements since we last spoke to you.

892
00:44:15,960 --> 00:44:18,719
Speaker 4: Well, the CLAW three is one of the largest, and

893
00:44:18,760 --> 00:44:22,040
Speaker 4: then you've got all kinds of interesting collaborations. You've got

894
00:44:22,320 --> 00:44:25,880
Speaker 4: Nvidia and Microsoft died one with a huge number of tokens.

895
00:44:26,120 --> 00:44:29,200
Speaker 4: You've got a one to one labs that has this

896
00:44:29,239 --> 00:44:32,280
Speaker 4: thing called Jurassic Again, a lot of people are making

897
00:44:32,520 --> 00:44:35,400
Speaker 4: headway here, but we are I think a year away

898
00:44:35,920 --> 00:44:40,440
Speaker 4: from you being able to upload hundreds of PDFs thousands

899
00:44:40,440 --> 00:44:44,360
Speaker 4: of books if they're not already immediately referenceable and be

900
00:44:44,480 --> 00:44:50,680
Speaker 4: able to detect pattern change amongst documents, summarize and unearth,

901
00:44:51,160 --> 00:44:54,480
Speaker 4: you know, the entirety of key concepts, and then I

902
00:44:54,480 --> 00:44:57,440
Speaker 4: think the most valuable thing will be it prompting you

903
00:44:57,520 --> 00:45:00,160
Speaker 4: to say, here's a question you didn't ask about all

904
00:45:00,200 --> 00:45:03,000
Speaker 4: these documents that you just uploaded. So yeah, I think

905
00:45:03,320 --> 00:45:07,480
Speaker 4: we're just keep increasing the context window. But that said,

906
00:45:08,200 --> 00:45:11,879
Speaker 4: most of the history of innovation is just like, keep

907
00:45:11,880 --> 00:45:15,400
Speaker 4: increasing this factor, and then somebody else comes along and

908
00:45:15,400 --> 00:45:17,719
Speaker 4: invent something. It's like that factor doesn't matter anymore. You know.

909
00:45:17,760 --> 00:45:21,000
Speaker 4: My favorite iconic example of this is like sailboats. You

910
00:45:21,040 --> 00:45:23,800
Speaker 4: find these sailing ships back in the day, they just

911
00:45:23,840 --> 00:45:25,799
Speaker 4: kept adding more and more sales, Like these things started

912
00:45:25,800 --> 00:45:28,440
Speaker 4: to look ridiculous, you know, and then somebody invents the

913
00:45:28,480 --> 00:45:31,800
Speaker 4: electric motor, and you have a motor both, so I

914
00:45:31,840 --> 00:45:33,719
Speaker 4: think we'll have the same sort of thing. And then

915
00:45:33,719 --> 00:45:37,239
Speaker 4: people figure out, hey, there's a better architecture here than

916
00:45:37,320 --> 00:45:39,880
Speaker 4: just constantly increasing the context window. And some of that

917
00:45:39,960 --> 00:45:43,880
Speaker 4: might be with memory retrieval and being able to reference

918
00:45:43,880 --> 00:45:46,319
Speaker 4: other models and just go into the archive of what

919
00:45:46,360 --> 00:45:48,840
Speaker 4: they have. So yeah, that's going to keep expanding.

920
00:45:48,960 --> 00:45:52,080
Speaker 2: Josh woolf Lux Capital, thank you so much for coming

921
00:45:52,120 --> 00:45:53,160
Speaker 2: back on odd lots.

922
00:45:53,200 --> 00:45:55,719
Speaker 3: Always great to get an update on what you're interested in.

923
00:45:56,040 --> 00:45:57,520
Speaker 4: It was great to be with you guys.

924
00:45:57,680 --> 00:46:02,600
Speaker 2: Yeah good, you're prepared already to ingratiate yourself and to

925
00:46:03,120 --> 00:46:05,160
Speaker 2: blend in with the human art robot.

926
00:46:06,280 --> 00:46:07,920
Speaker 3: Thank you so much. That was fantastic.

927
00:46:22,600 --> 00:46:24,839
Speaker 2: First of all, Tracy, I really like talking to Josh

928
00:46:24,960 --> 00:46:27,680
Speaker 2: and always like getting an update. I really do want

929
00:46:27,680 --> 00:46:30,200
Speaker 2: the folding the clothes folding robot though, Like I actually

930
00:46:30,280 --> 00:46:33,080
Speaker 2: think that's a really big deal and would make almost

931
00:46:33,120 --> 00:46:36,839
Speaker 2: everyone's life better if they didn't have to worry about

932
00:46:36,880 --> 00:46:37,600
Speaker 2: folding clothes.

933
00:46:37,680 --> 00:46:40,200
Speaker 1: I agree, it would be far more useful to have

934
00:46:40,280 --> 00:46:43,799
Speaker 1: something doing physical tasks like folding laundry versus telling you

935
00:46:43,880 --> 00:46:46,799
Speaker 1: where you're lost, Yeah, is in your desk.

936
00:46:47,040 --> 00:46:48,840
Speaker 3: I want I need that folding robot.

937
00:46:48,880 --> 00:46:50,920
Speaker 1: I mean, I will say, I know everyone likes to

938
00:46:51,160 --> 00:46:54,399
Speaker 1: make fun of Alexa as well, but our house we've

939
00:46:54,480 --> 00:46:57,920
Speaker 1: kitted out all the lights on. They're all those smart

940
00:46:57,960 --> 00:47:01,040
Speaker 1: bulbs because we don't have any overhead wiring, so like

941
00:47:01,080 --> 00:47:04,359
Speaker 1: everything has to be lamps. So if you didn't have

942
00:47:04,880 --> 00:47:08,040
Speaker 1: a robot, that was able to turn on all of

943
00:47:08,080 --> 00:47:10,480
Speaker 1: your appliances at once in a room. It would be

944
00:47:10,640 --> 00:47:13,160
Speaker 1: incredibly annoying because you would be going from lamp to

945
00:47:13,239 --> 00:47:15,840
Speaker 1: lamp to lamp. So it does make a difference in

946
00:47:15,880 --> 00:47:18,520
Speaker 1: my daily life at least, I mean, there's so much

947
00:47:18,920 --> 00:47:20,759
Speaker 1: to pull out from that. So one thing that I

948
00:47:20,800 --> 00:47:25,120
Speaker 1: thought was interesting from an industrial policy perspective was Josh's

949
00:47:25,560 --> 00:47:30,319
Speaker 1: discussion of some of the robotic capabilities being developed in

950
00:47:30,480 --> 00:47:33,399
Speaker 1: places like China and the idea that we might have

951
00:47:33,440 --> 00:47:37,080
Speaker 1: another Chips Semiconductor like situation on our hands where we

952
00:47:37,120 --> 00:47:40,400
Speaker 1: wake up in ten years and realize that a primary

953
00:47:40,520 --> 00:47:44,840
Speaker 1: component of robotics is being built much more efficiently and

954
00:47:44,920 --> 00:47:48,160
Speaker 1: cheaply elsewhere outside of the US or the West. And

955
00:47:48,160 --> 00:47:50,000
Speaker 1: then the other thing I thought was interesting was the

956
00:47:50,040 --> 00:47:52,879
Speaker 1: idea of leap frogging, right, So, I think a lot

957
00:47:52,920 --> 00:47:56,480
Speaker 1: of people, myself included, when we think of technological advances,

958
00:47:56,560 --> 00:48:00,239
Speaker 1: it's like, can this do this thing slightly faster? Can

959
00:48:00,280 --> 00:48:02,759
Speaker 1: it do it on a slightly larger scale, to the

960
00:48:02,760 --> 00:48:06,759
Speaker 1: point about the context window and expansion there. But you

961
00:48:06,840 --> 00:48:10,600
Speaker 1: can leap frog in technology, as Josh was saying, and

962
00:48:10,640 --> 00:48:13,520
Speaker 1: you can go from the sailboat to the motor boat,

963
00:48:13,640 --> 00:48:19,000
Speaker 1: or you could bypass a human evolution, for instance, and

964
00:48:19,120 --> 00:48:22,920
Speaker 1: instead of having humanoid robots, you could have a Edward

965
00:48:23,000 --> 00:48:26,200
Speaker 1: scissorhands like thing with a Swiss army knife on the

966
00:48:26,280 --> 00:48:26,799
Speaker 1: end of his arm.

967
00:48:26,960 --> 00:48:27,200
Speaker 4: Yeah.

968
00:48:27,200 --> 00:48:29,239
Speaker 2: That made a ton of sense to me, which is like,

969
00:48:29,280 --> 00:48:32,440
Speaker 2: if you're like starting from scratch, like it's not obvious

970
00:48:32,480 --> 00:48:37,040
Speaker 2: that the human form that was developed over millions of

971
00:48:37,120 --> 00:48:41,040
Speaker 2: years through evolution is necessarily the thing you want to

972
00:48:41,120 --> 00:48:45,400
Speaker 2: create or recreate to do various tasks that you need.

973
00:48:45,640 --> 00:48:47,040
Speaker 2: There was a lot in there that I liked. The

974
00:48:47,160 --> 00:48:48,400
Speaker 2: thing that he would talk about at the end. It

975
00:48:48,480 --> 00:48:52,040
Speaker 2: sort of sounded like cloud kitchens, but for biology les. Yeah,

976
00:48:52,120 --> 00:48:53,560
Speaker 2: so if you just have all the robots do it

977
00:48:53,600 --> 00:48:56,640
Speaker 2: and then they can prompt you for other ideas, that's interesting.

978
00:48:56,800 --> 00:49:00,480
Speaker 3: It does seem exciting. The idea of ways to.

979
00:49:00,480 --> 00:49:03,799
Speaker 2: Accumulate training data for these sort of like you know

980
00:49:04,040 --> 00:49:07,160
Speaker 2: that you could maybe solve the mechanical engineering, but without

981
00:49:07,280 --> 00:49:09,560
Speaker 2: you know, there's no equivalent of like all of the

982
00:49:09,600 --> 00:49:12,839
Speaker 2: text on Reddit or Wikipedia or whatever that change or

983
00:49:12,880 --> 00:49:16,319
Speaker 2: you know, Google books or YouTube. So like having to

984
00:49:16,360 --> 00:49:20,640
Speaker 2: recreate that as a bottleneck for building robots was really interesting.

985
00:49:21,000 --> 00:49:24,319
Speaker 2: I love the term he used I think was ignorance arbitrage, Yeah,

986
00:49:24,360 --> 00:49:26,399
Speaker 2: which is a really great term. So it's like, yeah,

987
00:49:26,480 --> 00:49:29,640
Speaker 2: like in a lot of like pure science spaces, you're

988
00:49:29,680 --> 00:49:32,279
Speaker 2: going to get investors who are willing to throw money

989
00:49:32,320 --> 00:49:34,320
Speaker 2: at someone who just like has a really good idea

990
00:49:34,360 --> 00:49:35,880
Speaker 2: on paper because that person is smart.

991
00:49:35,920 --> 00:49:38,359
Speaker 1: Well, I think this is also the really unusual thing

992
00:49:38,480 --> 00:49:41,520
Speaker 1: about this particular cycle, which is the dominance of the

993
00:49:41,560 --> 00:49:44,640
Speaker 1: incumbents and the fact that on the one hand, you

994
00:49:44,719 --> 00:49:47,239
Speaker 1: do have a bunch of open source software and to

995
00:49:47,280 --> 00:49:51,440
Speaker 1: some extent you can take something off of a repository

996
00:49:51,600 --> 00:49:53,879
Speaker 1: and you can pitch it to investors and say this

997
00:49:53,920 --> 00:49:56,319
Speaker 1: is the next big thing, and they might not have

998
00:49:56,360 --> 00:50:01,200
Speaker 1: the technological expertise to actually evaluate that. But when it

999
00:50:01,239 --> 00:50:05,759
Speaker 1: comes to making you know, actual advancements in something like robotics,

1000
00:50:05,800 --> 00:50:07,880
Speaker 1: it does feel like you have to have an edge

1001
00:50:07,920 --> 00:50:09,799
Speaker 1: in one respect or another. You either have to have

1002
00:50:09,920 --> 00:50:12,840
Speaker 1: the capital to deploy or you have to have access

1003
00:50:13,080 --> 00:50:13,840
Speaker 1: to that data.

1004
00:50:14,160 --> 00:50:14,840
Speaker 4: So I don't know.

1005
00:50:14,880 --> 00:50:16,080
Speaker 1: I guess we'll see how it shakes out.

1006
00:50:16,200 --> 00:50:19,080
Speaker 2: I guess we'll have Josh back next year, yeah, next, next,

1007
00:50:19,400 --> 00:50:21,640
Speaker 2: next Springer summer to see what the next big thing is.

1008
00:50:21,640 --> 00:50:24,400
Speaker 1: Then hopefully he can bring a robot with him of some.

1009
00:50:24,320 --> 00:50:25,600
Speaker 3: Sort or folding robot.

1010
00:50:25,719 --> 00:50:27,080
Speaker 1: Yeah, all right, shall we leave it there.

1011
00:50:27,160 --> 00:50:27,879
Speaker 3: Let's leave it there.

1012
00:50:28,040 --> 00:50:30,840
Speaker 1: This has been another episode of the All Thoughts podcast.

1013
00:50:30,920 --> 00:50:34,200
Speaker 1: I'm Tracy Alloway. You can follow me at Tracy Alloway and.

1014
00:50:34,160 --> 00:50:36,640
Speaker 2: I'm joll Wisenthal. You can follow me at the Stalwart.

1015
00:50:36,840 --> 00:50:40,240
Speaker 2: Follow our guest Josh Wolf. He's at Wolf Josh. Follow

1016
00:50:40,280 --> 00:50:44,040
Speaker 2: our producers Carmen Rodriguez at Carman Erman dash, Ol Bennett

1017
00:50:44,040 --> 00:50:46,960
Speaker 2: at Dashbot and kel Brooks at Kelbrooks. Thank you to

1018
00:50:47,000 --> 00:50:49,760
Speaker 2: our producer Moses on Them. For more odd Lags content,

1019
00:50:49,840 --> 00:50:51,920
Speaker 2: go to Bloomberg dot com slash odd Lots, where we

1020
00:50:51,960 --> 00:50:55,080
Speaker 2: have transcripts of blog and a newsletter. And if you

1021
00:50:55,080 --> 00:50:57,479
Speaker 2: want to chat about all of these topics, including AI

1022
00:50:57,560 --> 00:51:00,279
Speaker 2: and robotics, there's a room in that in the lot

1023
00:51:00,360 --> 00:51:03,800
Speaker 2: Discord chatroom Discord dot gg odd lots.

1024
00:51:03,880 --> 00:51:04,600
Speaker 3: Go check it out.

1025
00:51:04,880 --> 00:51:07,359
Speaker 1: And if you enjoy odd Thoughts, if you want us

1026
00:51:07,400 --> 00:51:11,520
Speaker 1: to crowdsource buying a unitary humanoid robot or.

1027
00:51:11,480 --> 00:51:13,040
Speaker 3: Something similar, I'm Ellie Express.

1028
00:51:13,160 --> 00:51:16,239
Speaker 1: That's right, then please leave us a positive review on

1029
00:51:16,320 --> 00:51:20,000
Speaker 1: your favorite podcast platform. And remember, if you are a

1030
00:51:20,040 --> 00:51:23,120
Speaker 1: Bloomberg subscriber, you can listen to all of our episodes

1031
00:51:23,239 --> 00:51:26,080
Speaker 1: absolutely ad free. All you need to do is connect

1032
00:51:26,120 --> 00:51:30,120
Speaker 1: your Bloomberg subscription with Apple Podcasts. Thanks for listening

1033
00:52:00,040 --> 00:52:01,719
Speaker 2: And