1
00:00:00,000 --> 00:00:04,157
All right everybody, welcome back to
episode 290 of the number one podcast in

2
00:00:04,240 --> 00:00:06,637
the world. It's the all-in podcast. With

3
00:00:06,720 --> 00:00:09,357
me again, it's the core 4. David

4
00:00:09,440 --> 00:00:12,157
Freedberg is here. Chimoth Poly Hapatia

5
00:00:12,240 --> 00:00:16,557
and David Saxs calling in. I am your

6
00:00:16,640 --> 00:00:19,037
host, Jason Calakanis, and we have a

7
00:00:19,120 --> 00:00:21,917
full docket today. I'd be remiss though

8
00:00:22,000 --> 00:00:23,997
if I didn't ask everybody for their

9
00:00:24,080 --> 00:00:26,877
fondest memories of the fifth annual

10
00:00:26,960 --> 00:00:30,077
All-In Summit produced by

11
00:00:30,160 --> 00:00:32,237
Mr. David Friedberg. David, do you have

12
00:00:32,320 --> 00:00:35,357
any um thoughts? Uh how do you rate your

13
00:00:35,440 --> 00:00:39,437
performance as executive
producer of the All-In Summit?

14
00:00:39,520 --> 00:00:42,877
>> You know what we didn't do? What we
forgot to do at the end of day two? We

15
00:00:42,960 --> 00:00:47,757
didn't give a big shout out to our
production team, to Kimber, >> to Lisa,

16
00:00:47,840 --> 00:00:50,637
>> to Nick. I think we No, Jason. Okay.
I thought we should have brought him out,

17
00:00:50,720 --> 00:00:52,317
but I just want to give a big shout out.

18
00:00:52,400 --> 00:00:53,757
>> Bring him out for the standing up.

19
00:00:53,840 --> 00:00:58,077
>> The team backstage absolutely crushed
it as always this year. And then I had a

20
00:00:58,160 --> 00:01:00,877
great time. Actually, you know,
my favorite part is going to Universal

21
00:01:00,960 --> 00:01:03,757
Studios. You just get to have a good time.
It's great. Just great.

22
00:01:03,840 --> 00:01:07,517
>> I am reliving my childhood.
Shabbath with no trauma.

23
00:01:07,600 --> 00:01:11,037
>> By the way, shout out to our friends
at Dongi. I got one of those coffee

24
00:01:11,120 --> 00:01:14,317
machines that we had at the event.
I put it in my kitchen.

25
00:01:14,400 --> 00:01:15,997
>> You know what I got?
I don't know if you

26
00:01:16,080 --> 00:01:19,597
guys saw, but there was a water
machine in the VIP area.

27
00:01:19,680 --> 00:01:23,117
>> Well, >> that is a zero
plastic water machine. It

28
00:01:23,200 --> 00:01:25,277
goes through super filtration.

29
00:01:25,360 --> 00:01:28,877
>> No microplastics in the water. I have
one now in our office. Thank you very

30
00:01:28,960 --> 00:01:31,197
much to them. I appreciate it.

31
00:01:31,280 --> 00:01:34,557
>> Amazing company. >> Sax, do you
have a favorite moment from the show?

32
00:01:34,640 --> 00:01:36,317
>> Well, I think
the highlight of the show,

33
00:01:36,400 --> 00:01:40,157
the thing that we'll remember for a long
time was the president calling in during

34
00:01:40,240 --> 00:01:42,877
Jensen's talk. And I think Jensen's

35
00:01:42,960 --> 00:01:45,037
overall the content and and demeanor

36
00:01:45,120 --> 00:01:47,677
really helped calm down this national

37
00:01:47,760 --> 00:01:52,157
panic over AI. It felt like we were
the center of this national conversation.

38
00:01:52,240 --> 00:01:55,677
And I think it happened very
organically. You remember there's

39
00:01:55,760 --> 00:01:58,957
basically been this SCOP for the last
few weeks leading into the summit to

40
00:01:59,040 --> 00:02:01,117
convince everyone that humanity was

41
00:02:01,200 --> 00:02:03,517
going to go extinct unless we did

42
00:02:03,600 --> 00:02:06,957
exactly what the Democrats want and stop

43
00:02:07,040 --> 00:02:12,077
all AI development. And then you know
we had first Satcha came out and kind of

44
00:02:12,160 --> 00:02:15,357
calmed everyone down and then JSON did
the same thing and the president called

45
00:02:15,440 --> 00:02:19,917
in and did what he did best which is sees
total control of the narrative that

46
00:02:20,000 --> 00:02:23,037
I'm sure these doomer groups have been
carefully planning for weeks and kind of

47
00:02:23,120 --> 00:02:25,037
thwarted that whole operation.

48
00:02:25,120 --> 00:02:29,917
>> Nerds enough. Get back to work.
You have a super intelligent

49
00:02:30,000 --> 00:02:31,997
Trump intelligence is now the term.

50
00:02:32,080 --> 00:02:33,917
We're going to call it Trump
intelligence going forward. But I think

51
00:02:34,000 --> 00:02:35,117
you make a good point. A lot of people

52
00:02:35,200 --> 00:02:37,597
wanted to know Freeberg if this was

53
00:02:37,680 --> 00:02:43,757
planned. Was the President Trump calling
planned Freeberg? T take us behind the

54
00:02:43,840 --> 00:02:47,037
scenes. >> Basically,
Jensen was connecting with

55
00:02:47,120 --> 00:02:50,557
him right backstage and said, "Hey, call
me back." And then he called him while

56
00:02:50,640 --> 00:02:53,677
he was on stage.
There was some texting going on.

57
00:02:53,760 --> 00:02:55,197
>> Yeah, it was completely spontaneous.

58
00:02:55,280 --> 00:02:57,597
>> It was spontaneous. Yes.
>> It was not like a planned thing where

59
00:02:57,680 --> 00:03:00,797
Jensen was like, "Hey, let's do a whole
thing." And so that was pretty amazing

60
00:03:00,880 --> 00:03:03,997
to have this conversation that Jensen

61
00:03:04,080 --> 00:03:05,677
was waiting for from the president to

62
00:03:05,760 --> 00:03:08,957
happen live and clearly this was the

63
00:03:09,040 --> 00:03:13,277
topic desour in DC as well as in the news
and the media and it was amazing to

64
00:03:13,360 --> 00:03:15,757
see it all happen on stage with us and

65
00:03:15,840 --> 00:03:20,957
yeah just to uh be clear I thought it
was a bit and I was like well put him on

66
00:03:21,040 --> 00:03:23,997
speakerphone and Jensen says to me how
do you do that and I just handed him I

67
00:03:24,080 --> 00:03:26,637
always keep an extra mic there in case
somebody's mic goes out just put them on

68
00:03:26,720 --> 00:03:29,997
speaker phone put the microphone
and President Trump the funniest president

69
00:03:30,080 --> 00:03:31,997
in history, funniest politician history.

70
00:03:32,080 --> 00:03:33,357
Did you notice

71
00:03:33,440 --> 00:03:36,877
>> Sachs that he drops
a oneliner through a

72
00:03:36,960 --> 00:03:40,877
speaker phone perfectly, >> perfectly.
Jensen makes the greatest

73
00:03:40,960 --> 00:03:44,237
chips. Everybody tries to reverse
engineer them. Impossible. Impossible.

74
00:03:44,320 --> 00:03:47,917
But he can't figure out how to press
the speakerphone button. Okay. [laughter]

75
00:03:48,000 --> 00:03:50,077
>> I was just like, how does he do it? How

76
00:03:50,160 --> 00:03:54,237
does he do it? It just drops the perfect

77
00:03:54,320 --> 00:03:56,557
line. But Shimoth, in all seriousness,

78
00:03:56,640 --> 00:03:58,237
the nerds were losing their mind last

79
00:03:58,320 --> 00:04:01,437
week and the adults came, daddy came,

80
00:04:01,520 --> 00:04:04,317
Elon came, Jensen came, all the adults

81
00:04:04,400 --> 00:04:05,437
in the room came and they just said,

82
00:04:05,520 --> 00:04:09,357
"Listen, if you have software that's

83
00:04:09,440 --> 00:04:12,397
defective and not ready to put in the

84
00:04:12,480 --> 00:04:16,877
world, maybe don't put it in the world
and make it a little bit, you know, as

85
00:04:16,960 --> 00:04:21,117
you like to say, less brittle."
So your thoughts on where we are in this

86
00:04:21,200 --> 00:04:23,037
national discussion which went from slow

87
00:04:23,120 --> 00:04:25,197
down to maybe

88
00:04:25,280 --> 00:04:28,157
just release non-brittle resilient

89
00:04:28,240 --> 00:04:32,237
software. Well,
I think that there was this

90
00:04:32,320 --> 00:04:34,317
classification that happened last week

91
00:04:34,400 --> 00:04:36,637
between

92
00:04:36,720 --> 00:04:39,757
upand cominging organizations that want

93
00:04:39,840 --> 00:04:43,997
to call themselves labs, but are in fact

94
00:04:44,080 --> 00:04:46,717
companies that have P&Ls and

95
00:04:46,800 --> 00:04:48,637
shareholders and have to abide by the

96
00:04:48,720 --> 00:04:51,837
same rules that everybody else does. And

97
00:04:51,920 --> 00:04:53,997
then the sophisticated companies that

98
00:04:54,080 --> 00:04:58,157
have lived in scrutiny for many decades

99
00:04:58,240 --> 00:05:00,957
and have learned the hard way

100
00:05:01,040 --> 00:05:03,437
the process of making things robust and

101
00:05:03,520 --> 00:05:08,717
the cost of making things not robust.
So what do I mean? On the one side you saw

102
00:05:08,800 --> 00:05:10,397
in many ways it's like curious times

103
00:05:10,480 --> 00:05:12,077
creates curious bedfellows right you

104
00:05:12,160 --> 00:05:17,517
have the cohort is Satya Jensen Sachs

105
00:05:17,600 --> 00:05:22,157
Zach Elon the president Lena Khan we

106
00:05:22,240 --> 00:05:25,277
have product liability language
>> strange fellows indeed

107
00:05:25,360 --> 00:05:29,677
>> we have this product liability
perspective in the United States it's

108
00:05:29,760 --> 00:05:34,397
meant to prevent companies from doing
things that they're not supposed to do

109
00:05:34,480 --> 00:05:37,277
especially if they know that
those things could be harmful

110
00:05:37,360 --> 00:05:42,157
And on the other side, it's folks making
the thing that were being, I think, a

111
00:05:42,240 --> 00:05:45,997
little naive and basically saying,
"We're not a company. We're a lab.

112
00:05:46,080 --> 00:05:48,477
Protect us." And I think the mature

113
00:05:48,560 --> 00:05:53,197
response is, "You should have internal
controls and you should robustly test

114
00:05:53,280 --> 00:05:57,357
things." And what Zuck said earlier and
what Elon has said in the past is we're

115
00:05:57,440 --> 00:06:00,317
just under a lot of scrutiny. And so
when things aren't perfect, we just slow

116
00:06:00,400 --> 00:06:03,597
things down. Elon has done that with FSD.
You said this, Jason. He could have

117
00:06:03,680 --> 00:06:08,157
been at FSD 3 or 4 years
ago, but he has publicly

118
00:06:08,240 --> 00:06:09,997
stated, "We can't afford it because

119
00:06:10,080 --> 00:06:13,197
every accident that Tesla incurs is

120
00:06:13,280 --> 00:06:17,357
magnified a,000x than what any other
company would incur, and so we just have

121
00:06:17,440 --> 00:06:21,677
to go slower." Zach last week said, "We
were about to roll out Muse, which looks

122
00:06:21,760 --> 00:06:25,277
like one of the most successful product
launches in recent memory, has huge

123
00:06:25,360 --> 00:06:29,437
implications actually, which we should
talk about, and he had to slow roll it."

124
00:06:29,520 --> 00:06:33,037
He and Alexander who I was talking
to last night had to slow roll it for a

125
00:06:33,120 --> 00:06:37,197
couple of months to get it perfect.
So it's all within your control. It's just

126
00:06:37,280 --> 00:06:41,437
what do you choose to do? And uh I
think that was the most interesting

127
00:06:41,520 --> 00:06:44,957
observation. We have to stop calling
these companies labs. They're companies

128
00:06:45,040 --> 00:06:48,397
and we have to be judged on the same

129
00:06:48,480 --> 00:06:50,557
standards including product liability.

130
00:06:50,640 --> 00:06:54,717
>> Yeah. And there was a lot of coverage
of this moment Sachs. Interestingly, some

131
00:06:54,800 --> 00:06:59,597
of the fake news media [laughter]
were like on a popular podcast and then

132
00:06:59,680 --> 00:07:04,157
they play the clip, but they're just
afraid to say the name of the podcast.

133
00:07:04,240 --> 00:07:06,717
Pretty hilarious. >> I think
Jimoth is getting us into the

134
00:07:06,800 --> 00:07:08,717
first topic here already, which is

135
00:07:08,800 --> 00:07:10,717
whether we're going to have individual

136
00:07:10,800 --> 00:07:14,397
responsibility as our northstar or some

137
00:07:14,480 --> 00:07:17,597
sort of collective action. And the thing

138
00:07:17,680 --> 00:07:21,437
that always works the best is individual
responsibility. Whether it's at the

139
00:07:21,520 --> 00:07:23,757
level of individuals or companies, the

140
00:07:23,840 --> 00:07:26,877
relevant decision to make in every case

141
00:07:26,960 --> 00:07:30,957
is should we ship this product? Not can

142
00:07:31,040 --> 00:07:34,237
we get the whole world to agree. And the

143
00:07:34,320 --> 00:07:36,397
problem with talking about these

144
00:07:36,480 --> 00:07:38,797
collectivized approaches is that they

145
00:07:38,880 --> 00:07:44,157
defay accountability away from the
actual decision maker. And I I think

146
00:07:44,240 --> 00:07:45,597
people sort of feel this, you know, like

147
00:07:45,680 --> 00:07:49,437
this week you had Daario and Sam go to

148
00:07:49,520 --> 00:07:54,317
the United Nations and say we need some
sort of global AI governance. And it

149
00:07:54,400 --> 00:07:57,757
kind of reminds me of like
the billionaires who fly to Davos in their

150
00:07:57,840 --> 00:08:01,517
private jets and then rail against
climate change. You know, there's

151
00:08:01,600 --> 00:08:05,597
something fundamentally sort
of unrealistic and kind of out of touch

152
00:08:05,680 --> 00:08:09,357
about it. >> Hypocritical. >>
Hypocritical. And it's like, guys, look,

153
00:08:09,440 --> 00:08:11,197
what we need is

154
00:08:11,280 --> 00:08:13,757
>> for you guys to make smart decisions

155
00:08:13,840 --> 00:08:16,797
about what to release. If a product is

156
00:08:16,880 --> 00:08:19,277
unsafe, which is to say it's unreliable

157
00:08:19,360 --> 00:08:21,757
or behaves unpredictably, then don't

158
00:08:21,840 --> 00:08:27,197
release it. And stop looking to some
sort of global governance as the answer.

159
00:08:27,280 --> 00:08:29,837
Because if you guys don't get this
decision right, it's not going to matter

160
00:08:29,920 --> 00:08:32,957
what the United Nations does. And by
the time the United Nations agrees on

161
00:08:33,040 --> 00:08:36,717
anything, we're going to be much further
along in this whole whole journey. So,

162
00:08:36,800 --> 00:08:41,677
there's something just very weird about
the way that the principal actors in

163
00:08:41,760 --> 00:08:45,037
this keep talking [clears throat] >>
because it really feels like they're

164
00:08:45,120 --> 00:08:48,077
trying to push off the responsibility.

165
00:08:48,160 --> 00:08:49,917
>> Well, there's
a theory going around here

166
00:08:50,000 --> 00:08:51,837
and I think it is worth us discussing

167
00:08:51,920 --> 00:08:56,477
here that the Frontier Labs are covertly

168
00:08:56,560 --> 00:08:57,917
>> companies. They're companies.

169
00:08:58,000 --> 00:08:59,597
>> They're companies.
Yes. Let me just finish the

170
00:08:59,680 --> 00:09:02,317
>> No, no, no, but but your made a really
good point that I want to make sure we

171
00:09:02,400 --> 00:09:04,717
just don't miss. Okay.
>> They are corporations.

172
00:09:04,800 --> 00:09:07,757
>> Yes. Yes. And when you call them labs,
they sound like nonprofit entities, and

173
00:09:07,840 --> 00:09:08,557
it's not true.

174
00:09:08,640 --> 00:09:11,597
>> Okay. So, the corporations making these

175
00:09:11,680 --> 00:09:15,597
products are looking for President Trump

176
00:09:15,680 --> 00:09:18,317
to specifically give them what the

177
00:09:18,400 --> 00:09:21,037
internet companies had with section 230,

178
00:09:21,120 --> 00:09:23,757
which is shield them from liability. and

179
00:09:23,840 --> 00:09:28,157
that there is a consideration inside
the Trump administration, the 47th

180
00:09:28,240 --> 00:09:30,637
administration, that maybe there could

181
00:09:30,720 --> 00:09:33,117
be some leverage put on these companies

182
00:09:33,200 --> 00:09:35,357
that they would give 10% of their equity

183
00:09:35,440 --> 00:09:37,677
to the uh sovereign wealth front of the

184
00:09:37,760 --> 00:09:40,557
United States, etc., and then that would

185
00:09:40,640 --> 00:09:44,717
result in them being given some pass

186
00:09:44,800 --> 00:09:47,357
when it comes to their liability. That's

187
00:09:47,440 --> 00:09:50,077
an insane thing to do. It's like giving

188
00:09:50,160 --> 00:09:52,157
some plane company that gives 10% of

189
00:09:52,240 --> 00:09:54,797
their equity to the US government the

190
00:09:54,880 --> 00:09:57,357
ability to not be responsible for their

191
00:09:57,440 --> 00:10:01,837
planes falling out of the sky. But what
is your thought on that conspiracy

192
00:10:01,920 --> 00:10:06,157
theory? I'm sure you've heard it sachs
that they're looking for. You're you're

193
00:10:06,240 --> 00:10:09,597
making it sound like this is the idea of
the Trump administration. That's not true.

194
00:10:09,680 --> 00:10:12,237
>> No, not the Trump administration.
This is what the

195
00:10:12,320 --> 00:10:16,957
>> Frontier Lab/Corporations are pitching
to them and that there might be the

196
00:10:17,040 --> 00:10:20,237
rumor is there might be some people who
are open to that in the administration.

197
00:10:20,320 --> 00:10:23,037
Again, >> this is all a back channel.

198
00:10:23,120 --> 00:10:27,117
>> Here's what we know is that in the past
week, you had virtually every Trump

199
00:10:27,200 --> 00:10:29,757
administration official come out and say

200
00:10:29,840 --> 00:10:34,157
some version of what I just said, which
is first and foremost, the AI companies

201
00:10:34,240 --> 00:10:38,237
have to take responsibility for their
own products and that we will not wave

202
00:10:38,320 --> 00:10:42,397
product liability. I think Speaker Johnson
said it, the congressional >> said it,

203
00:10:42,480 --> 00:10:43,677
>> Besson said it in the most recent

204
00:10:43,760 --> 00:10:46,717
interview. Everybody said if your game

205
00:10:46,800 --> 00:10:49,677
is to get a waiver of product liability

206
00:10:49,760 --> 00:10:54,077
or antitrust liability, if you're trying
to form a cartel, if you're trying to

207
00:10:54,160 --> 00:10:57,277
collude, forget about it.
We're not going to do it. So, I think they

208
00:10:57,360 --> 00:11:00,237
completely threw cold water on that whole
idea. And remember what President

209
00:11:00,320 --> 00:11:02,717
Trump said is he just tweeted today, we

210
00:11:02,800 --> 00:11:07,277
have the DOJ. That's a guardrail.
The administration is a guardrail. the

211
00:11:07,360 --> 00:11:09,597
administrative state, civil lawsuits,

212
00:11:09,680 --> 00:11:12,477
criminal lawsuits, all these things are

213
00:11:12,560 --> 00:11:18,077
there as potential liability to make
sure that you make the right riskreward

214
00:11:18,160 --> 00:11:21,197
tradeoff. >> And again,
at the end of the day, if

215
00:11:21,280 --> 00:11:24,957
those guys release an unsafe product,
it's not going to matter what the United

216
00:11:25,040 --> 00:11:28,637
Nations agrees on. So, we need them
to internalize the responsibility.

217
00:11:28,720 --> 00:11:32,397
>> Are they Listen, come on. Be candid
with me. You you you got the inside goods

218
00:11:32,480 --> 00:11:34,157
here. Are they trying to lobby for that?

219
00:11:34,240 --> 00:11:38,077
Are they do you think they are lobbying to
get this protection? Yes or no? Come on.

220
00:11:38,160 --> 00:11:41,357
>> Well, I I have heard this is not
like first-p person experience. They they

221
00:11:41,440 --> 00:11:45,037
have not asked me like, "Hey, we want
product liability waiver, but I have

222
00:11:45,120 --> 00:11:48,637
heard from people in Washington that they
have been seeking something like that.

223
00:11:48,720 --> 00:11:50,077
>> It's hard to pin down, right?"

224
00:11:50,160 --> 00:11:54,237
>> Yeah. Okay, fair enough. Freeberg,
your thoughts here as we are going to get

225
00:11:54,320 --> 00:11:58,077
into the first story here with the token
maxing and the prices going down, but

226
00:11:58,160 --> 00:12:00,237
I'll give you the final word here on

227
00:12:00,320 --> 00:12:02,797
what America should do in terms of

228
00:12:02,880 --> 00:12:05,197
dealing with the corporations that are

229
00:12:05,280 --> 00:12:07,277
pretending to be labs who cannot take

230
00:12:07,360 --> 00:12:10,557
ownership of their own Frankenstein uh

231
00:12:10,640 --> 00:12:13,037
creations, which I think JD Vance, you

232
00:12:13,120 --> 00:12:17,597
know, I thought had a spectacular
appearance. He was crisp and and looked

233
00:12:17,680 --> 00:12:21,997
quite presidential if I if I
was going to um describe it.

234
00:12:22,080 --> 00:12:25,437
you threw a lot of fast balls >>
and he knocked him out of the park. By

235
00:12:25,520 --> 00:12:28,717
the way, fast, you know, home run
hitters want fast balls. You can't hit a

236
00:12:28,800 --> 00:12:31,197
fast ball if it's out of the strike zone.
So, I gave him the the three

237
00:12:31,280 --> 00:12:34,797
hardest questions I could think of and he
>> proceeded to knock all three over the

238
00:12:34,880 --> 00:12:37,917
fence. I think there was >>
Jad's excellent listen.

239
00:12:38,000 --> 00:12:41,757
>> I mean, no, on the Sunday shows, >>
they played the clips of the questions I

240
00:12:41,840 --> 00:12:45,677
asked him. Bang, bang, bang. Freeberg
right over the f the fence. But, you

241
00:12:45,760 --> 00:12:49,357
know, he said basically, if you're
creating Frankenstein, here's an idea.

242
00:12:49,440 --> 00:12:53,357
Stop. [laughter] And and
and then he added, if you've

243
00:12:53,440 --> 00:12:57,197
already kind of let it the cat out
of the bag, then create anti-Frankenstein.

244
00:12:57,280 --> 00:12:59,117
Basically, create the safeguards.

245
00:12:59,200 --> 00:13:00,637
>> Yeah, >> it makes total sense.

246
00:13:00,720 --> 00:13:03,277
>> I think Suck gave
a good interview on this.

247
00:13:03,360 --> 00:13:08,717
>> They took their time doing this Muse
release this week. They said they wanted

248
00:13:08,800 --> 00:13:12,317
to get it right. They wanted to make
sure that it was safe, that it didn't

249
00:13:12,400 --> 00:13:16,877
give people capabilities to do bad things.
That's the responsibility that

250
00:13:16,960 --> 00:13:21,197
they took as a corporation. Anyone
that's outputting a model should take

251
00:13:21,280 --> 00:13:24,877
the same responsibility.
>> Pretty obvious.

252
00:13:24,960 --> 00:13:26,557
>> This is an organization
that understands

253
00:13:26,640 --> 00:13:30,717
it. Having paid a series of fines from

254
00:13:30,800 --> 00:13:34,957
Cambridge Analytica to recently >>
we have a lot of we have a lot of laws

255
00:13:35,040 --> 00:13:37,917
that make it illegal
to do a lot of things.

256
00:13:38,000 --> 00:13:40,877
>> All right, let's move on here.
We [laughter] have laws.

257
00:13:40,960 --> 00:13:45,197
>> By the way, can I add like one
philosophical point to this? So I I

258
00:13:45,280 --> 00:13:50,317
think that implicit in let's
call it the Daario position

259
00:13:50,400 --> 00:13:52,157
that let's call it the regulated free

260
00:13:52,240 --> 00:13:55,677
market that we have his fundamental

261
00:13:55,760 --> 00:13:58,237
contention is that that will lead to a

262
00:13:58,320 --> 00:14:03,277
race to the bottom that competition
jeopardizes safety. It's in all of his

263
00:14:03,360 --> 00:14:05,117
blog posts and many other people reflect

264
00:14:05,200 --> 00:14:08,877
this idea that competition equals lack

265
00:14:08,960 --> 00:14:11,757
of safety and that is implicitly a

266
00:14:11,840 --> 00:14:16,797
critique of our economic system.
It's basically a leftwing critique of our

267
00:14:16,880 --> 00:14:18,637
system and it goes back a long way. I

268
00:14:18,720 --> 00:14:21,277
remember back in the days of the cold

269
00:14:21,360 --> 00:14:25,837
war when we really had this duel between
two different systems. Do you want the

270
00:14:25,920 --> 00:14:29,197
sort of the free market system or do
you want the communist system? A lot of

271
00:14:29,280 --> 00:14:31,837
people made the claim that the

272
00:14:31,920 --> 00:14:33,997
capitalist system would do things like

273
00:14:34,080 --> 00:14:38,957
jeopardize safety and beauty, that our
buildings would always be uglier because

274
00:14:39,040 --> 00:14:43,117
the greedy capitalist wouldn't want to
spend a dime on beautifification or good

275
00:14:43,200 --> 00:14:45,117
architecture, things like that. In fact,

276
00:14:45,200 --> 00:14:47,997
it turned out that the Soviet system was

277
00:14:48,080 --> 00:14:53,277
the one that produced the cold gray
landscapes. Why? Because people want

278
00:14:53,360 --> 00:14:56,317
beauty. They want safety. And therefore,
if you have a well functioning market

279
00:14:56,400 --> 00:14:58,957
economy, then it gives people what they

280
00:14:59,040 --> 00:15:01,517
want. And obviously, the American system

281
00:15:01,600 --> 00:15:04,957
won out in that battle. So, I just

282
00:15:05,040 --> 00:15:07,677
fundamentally disagree with the idea

283
00:15:07,760 --> 00:15:12,317
that competition means lack of safety.

284
00:15:12,400 --> 00:15:15,677
As Zuckerberg and Jensen were saying,

285
00:15:15,760 --> 00:15:17,757
there are huge incentives on both the

286
00:15:17,840 --> 00:15:20,237
upside and downside to be safe. On the

287
00:15:20,320 --> 00:15:23,197
upside, no customer wants a product that

288
00:15:23,280 --> 00:15:28,157
is unpredictable and unreliable, that
doesn't do what they say. No enterprise

289
00:15:28,240 --> 00:15:32,317
wants an agent that's going to leak
their data or hack a competitor. And on

290
00:15:32,400 --> 00:15:35,997
the downside, as the president made
clear, they've got tons of liability,

291
00:15:36,080 --> 00:15:40,077
product liability, civil liability,
administrative liability, criminal

292
00:15:40,160 --> 00:15:44,317
liability. So, I just firmly disagree
with this idea that competition is a bad

293
00:15:44,400 --> 00:15:47,517
thing. Competition is a good thing
that can be harnessed to get the right

294
00:15:47,600 --> 00:15:49,757
result. And just to build on that,

295
00:15:49,840 --> 00:15:53,677
Shimoth our uh friend Nash from PaloAlto

296
00:15:53,760 --> 00:15:58,717
Networks, friend of the pod, he released
in the past week or so unit 42. He's

297
00:15:58,800 --> 00:16:00,477
building there's a business opportunity

298
00:16:00,560 --> 00:16:04,077
here to build a product to do continuous

299
00:16:04,160 --> 00:16:07,197
cyber defense and also a friend of the

300
00:16:07,280 --> 00:16:10,397
George Kirch from um Crowdstrike. He

301
00:16:10,480 --> 00:16:13,197
also released Falcon, his new product to

302
00:16:13,280 --> 00:16:15,997
do cyber defense. Chimath, if these guys

303
00:16:16,080 --> 00:16:20,557
in the Frontier Labscorporations
were so concerned

304
00:16:20,640 --> 00:16:24,877
about cyber security, why didn't they
ever release a cyber security product

305
00:16:24,960 --> 00:16:27,437
and PaloAlto Networks and Crowd Strike

306
00:16:27,520 --> 00:16:32,397
are on it like they will?
That's probably more in the second topic

307
00:16:32,480 --> 00:16:37,117
that we'll talk about, which is as the
models essentially cluster in terms of

308
00:16:37,200 --> 00:16:40,477
capability, there's no differentiation,
and so they'll be forced to move up the

309
00:16:40,560 --> 00:16:44,317
stack. But let's let's wait on that for
one second. You have a point here you

310
00:16:44,400 --> 00:16:45,597
want to make and then
we'll go to the first.

311
00:16:45,680 --> 00:16:48,877
>> I do. The point I
wanted to make is more

312
00:16:48,960 --> 00:16:53,437
philosophical and sociological which is

313
00:16:53,520 --> 00:16:56,477
the last time we heard the word lab in

314
00:16:56,560 --> 00:16:59,597
the social consciousness was during co

315
00:16:59,680 --> 00:17:03,597
and there was a lab that was responsible

316
00:17:03,680 --> 00:17:07,117
for a leak of a virus that then killed

317
00:17:07,200 --> 00:17:08,877
15 million people all around the world

318
00:17:08,960 --> 00:17:12,557
and caused $45 trillion of damage. It's

319
00:17:12,640 --> 00:17:14,957
the Wuhan lab.

320
00:17:15,040 --> 00:17:17,197
And I think it's important to recognize

321
00:17:17,280 --> 00:17:22,157
that there has been zero transparency
or culpability around this. And I think

322
00:17:22,240 --> 00:17:25,757
part of why people are comfortable
calling themselves a lab is because of

323
00:17:25,840 --> 00:17:27,997
that fact. That a lot of these young

324
00:17:28,080 --> 00:17:29,597
people that work at these places live

325
00:17:29,680 --> 00:17:33,437
through this and saw absolutely no um

326
00:17:33,520 --> 00:17:37,117
responsibility taken by the lab that

327
00:17:37,200 --> 00:17:41,277
caused this leak. And now we're talking
about leaks of different kinds. And we

328
00:17:41,360 --> 00:17:45,517
have organizations and corporations
that try to fashion themselves as labs as

329
00:17:45,600 --> 00:17:48,637
well, but they're not labs. These are
corporations. I just want to remind

330
00:17:48,720 --> 00:17:53,837
everybody, these are forprofit
corporations that have absorbed hundreds

331
00:17:53,920 --> 00:17:58,237
of billions of dollars of debt in equity,
>> that have thousands and thousands of

332
00:17:58,320 --> 00:18:00,077
shareholders, that have trillions of

333
00:18:00,160 --> 00:18:03,117
dollars of market cap. And if I were

334
00:18:03,200 --> 00:18:06,877
them, I would not want to call myself a

335
00:18:06,960 --> 00:18:09,677
lab at any cost. And I think if you saw

336
00:18:09,760 --> 00:18:14,397
the recent speeches that they gave
at the United Nations, they're still using

337
00:18:14,480 --> 00:18:18,477
that terminology. >> Yeah.
>> They are not labs. These are for-profit

338
00:18:18,560 --> 00:18:21,357
corporations that are subject to product
liability risk and they should act

339
00:18:21,440 --> 00:18:23,757
accordingly. >> Yeah.
I mean, and we've had a

340
00:18:23,840 --> 00:18:25,597
conversation here for a year about this.

341
00:18:25,680 --> 00:18:29,597
Like if this stuff is so dangerous, you
don't have to release it or you know my

342
00:18:29,680 --> 00:18:31,757
proposal like know
your customer, give logs.

343
00:18:31,840 --> 00:18:34,957
>> Do not compare yourself to the Wuhan

344
00:18:35,040 --> 00:18:39,357
lab. It's well, Shimoth, it's bad PR,
but also I do I do think it's part of

345
00:18:39,440 --> 00:18:43,917
their like virtue signaling routine
where they just don't want to say that

346
00:18:44,000 --> 00:18:47,197
we're companies. We have a for-profit
incentive. They always want to say that

347
00:18:47,280 --> 00:18:51,517
we're either a research institution
or they say things like we're a public

348
00:18:51,600 --> 00:18:54,797
benefit corporation, which no one really
knows what the hell that means. So, it's

349
00:18:54,880 --> 00:18:58,957
this patina virtue signaling
that obscures what's really going on.

350
00:18:59,040 --> 00:19:01,677
>> It tries to obscure the word
corporation, but legally you're not

351
00:19:01,760 --> 00:19:04,397
allowed to hide it.
I think they should have uh when they

352
00:19:04,480 --> 00:19:07,517
>> we are all running corporations all in

353
00:19:07,600 --> 00:19:12,157
all be it very all be it very small
but mighty is a corporation I think we

354
00:19:12,240 --> 00:19:16,637
should call >> is a all in lab >>
this is the all-in lab where we discuss

355
00:19:16,720 --> 00:19:19,917
different formulas >> no we're
experimenting with ideas this

356
00:19:20,000 --> 00:19:22,637
is a this is a meme workshop

357
00:19:22,720 --> 00:19:25,597
>> lab these ideas are
getting so dangerous

358
00:19:25,680 --> 00:19:29,917
that we don't want any liability for our
[laughter] opinions welcome to the allin

359
00:19:30,000 --> 00:19:31,517
lab podcast

360
00:19:31,600 --> 00:19:34,717
>> here we Topic zero freeberg. Is there a

361
00:19:34,800 --> 00:19:38,557
mindset in scientists minds that you

362
00:19:38,640 --> 00:19:42,637
know needs to kind of have a hard stop
and then the corporation kind of takes

363
00:19:42,720 --> 00:19:45,757
over? Maybe they could put the labs
on the side here and then say, "Hey, these

364
00:19:45,840 --> 00:19:48,797
are the adults doing the corporation
stuff." What are your thoughts here if

365
00:19:48,880 --> 00:19:53,517
you were going to architect how
anthropic operates in the world? And and

366
00:19:53,600 --> 00:19:57,437
maybe you could speak to scientists
views of the world versus capitalists

367
00:19:57,520 --> 00:20:01,437
and and management teams. >> I don't
have a good answer for you on that.

368
00:20:01,520 --> 00:20:03,677
>> Okay. Well, you're doing it.
You're the answer. Look at Oh,

369
00:20:03,760 --> 00:20:05,437
>> let me kick off the first topic. What I

370
00:20:05,520 --> 00:20:10,957
wanted to do was just take a step back
and look at the insane flurry of model

371
00:20:11,040 --> 00:20:14,317
releases happening. >> Yes.
>> I think the pace of model releases

372
00:20:14,400 --> 00:20:17,197
happening on both the open-source side

373
00:20:17,280 --> 00:20:20,237
and the premium side, the anthropics,

374
00:20:20,320 --> 00:20:24,717
gro of the world
is unfathomable right now.

375
00:20:24,800 --> 00:20:26,397
The list is insane. And just like the

376
00:20:26,480 --> 00:20:29,517
last 10 days, September 9th, Deepseek

377
00:20:29,600 --> 00:20:32,557
4.1 Flash gets released.

378
00:20:32,640 --> 00:20:36,957
20 for a million tokens output if
they're hosting it. Key Value Cash from

379
00:20:37,040 --> 00:20:40,637
V1 of DeepS, remember what a big deal
DeepS was when it came out. The key

380
00:20:40,720 --> 00:20:43,437
value cache then was like 390,000 bytes.

381
00:20:43,520 --> 00:20:45,517
Now it's 890.

382
00:20:45,600 --> 00:20:47,837
Insane drop, insane efficiency

383
00:20:47,920 --> 00:20:52,397
improvements. On September 20th, Quen,

384
00:20:52,480 --> 00:20:54,317
which is the Alibaba model, comes out

385
00:20:54,400 --> 00:20:57,197
version, you know, 2.1. It's an open

386
00:20:57,280 --> 00:21:02,157
weights model that outperforms
Nano Banana 2, which is Google's big

387
00:21:02,240 --> 00:21:04,077
breakthrough, insane image generation

388
00:21:04,160 --> 00:21:07,437
model. This thing is free, open- source.

389
00:21:07,520 --> 00:21:09,597
You can put it on your desktop computer

390
00:21:09,680 --> 00:21:12,717
and run it at home to generate images

391
00:21:12,800 --> 00:21:15,037
for free. And it's made by Alibaba.

392
00:21:15,120 --> 00:21:18,877
Xiaomé put out MIMO on September 22nd,

393
00:21:18,960 --> 00:21:22,477
two days ago. Mimo Pro performs on par

394
00:21:22,560 --> 00:21:26,077
with CLA's Opus 5, which is like this
insane model that everyone thought was

395
00:21:26,160 --> 00:21:30,397
going to change the world and break
the internet and GPT56 Soul across most

396
00:21:30,480 --> 00:21:32,557
benchmarks. It's a 309 billion parameter

397
00:21:32,640 --> 00:21:36,397
model. It is entirely open. You can

398
00:21:36,480 --> 00:21:41,517
download this model and put it on your
server and run your own instance without

399
00:21:41,600 --> 00:21:44,797
paying anyone any money except for the
cost of running the server. This model

400
00:21:44,880 --> 00:21:47,517
called uh Bonsai 2 came out on September

401
00:21:47,600 --> 00:21:49,997
17th by Prism ML. This is actually a

402
00:21:50,080 --> 00:21:51,997
fork of Quen, which is another one of

403
00:21:52,080 --> 00:21:54,637
the Alibaba models. It's a 27 billion

404
00:21:54,720 --> 00:21:57,757
parameter model. 98% of the performance

405
00:21:57,840 --> 00:22:01,917
of the big Quen model. It's 5.9 gigs in

406
00:22:02,000 --> 00:22:04,397
size and you can run it on a local

407
00:22:04,480 --> 00:22:07,117
computer. You can have an Nvidia chip or

408
00:22:07,200 --> 00:22:09,517
have a good Mac Studio computer and you

409
00:22:09,600 --> 00:22:12,157
can run this insane model on your

410
00:22:12,240 --> 00:22:15,117
desktop computer for free. And then the

411
00:22:15,200 --> 00:22:17,037
big crazy models, the big closed source

412
00:22:17,120 --> 00:22:19,437
models, Anthropic releases Opus 55 on

413
00:22:19,520 --> 00:22:22,797
September 22nd. Open AAI Astro comes out

414
00:22:22,880 --> 00:22:25,357
with Soul and Luna on September 23rd.

415
00:22:25,440 --> 00:22:28,877
Gro 47 comes out on September 21. Meta

416
00:22:28,960 --> 00:22:31,837
Muse comes out on September 22nd.

417
00:22:31,920 --> 00:22:35,917
Any one of these stories would have
broken the internet a year ago and they

418
00:22:36,000 --> 00:22:38,397
all happened in the last 10 days. You

419
00:22:38,480 --> 00:22:43,517
now have an entire catalog of openweight
models to generate visual information to

420
00:22:43,600 --> 00:22:47,997
do VLA where you can actually instruct
it and it can control a robot. You have

421
00:22:48,080 --> 00:22:52,077
models far beyond the most advanced LLM
from a year ago that are openweight that

422
00:22:52,160 --> 00:22:56,477
you can install and run on your desktop
computer. We are past the point of

423
00:22:56,560 --> 00:22:59,437
Bernie Sanders put the super
intelligence away. It can never be

424
00:22:59,520 --> 00:23:01,197
allowed in the real world. This stuff is

425
00:23:01,280 --> 00:23:04,317
out. The open- source openweight models

426
00:23:04,400 --> 00:23:08,317
are free. They are on the internet. You
can go and download them and you can run

427
00:23:08,400 --> 00:23:11,677
them on your computer at home. I don't
know in what world we think we're going

428
00:23:11,760 --> 00:23:15,597
to have a police state that's going
to come bang down your door, take your

429
00:23:15,680 --> 00:23:19,837
desktop computer from you and test it
to see if you're running an LLM on your

430
00:23:19,920 --> 00:23:23,917
local computer. The efficiency wave
is here. And the pacing at which these

431
00:23:24,000 --> 00:23:28,477
models are coming out now. All of this
stuff happened in the last 10 days. It's

432
00:23:28,560 --> 00:23:33,677
just mind-blowing. We're now in this kind
of frontier space where things are

433
00:23:33,760 --> 00:23:38,237
accelerating. The performance
improvements, the cost decline, AI is

434
00:23:38,320 --> 00:23:42,237
becoming so performative and so
affordable. It is going to be ubiquitous

435
00:23:42,320 --> 00:23:44,077
and we are all going to benefit from it.

436
00:23:44,160 --> 00:23:48,557
This is not just about the anthropic
and open AI with the closed world making

437
00:23:48,640 --> 00:23:52,477
billions of dollars selling enterprises
hosted models. There is so much

438
00:23:52,560 --> 00:23:55,917
proliferation right now. So I just
thought it was worth taking a step back

439
00:23:56,000 --> 00:23:59,837
to recognize where we are because in
the context of we have to stop the super

440
00:23:59,920 --> 00:24:03,117
intelligence. We have to shut down
the data centers. You don't even need data

441
00:24:03,200 --> 00:24:06,317
centers to do 90% of what you can do
with AI. You can download it on your

442
00:24:06,400 --> 00:24:11,437
computer for free today to do what
was the most advanced technology in human

443
00:24:11,520 --> 00:24:15,037
history less than a year ago. It's all
here today. And I thought it was just

444
00:24:15,120 --> 00:24:18,237
worth taking a beat to step back
and make sure we all recognize that as we

445
00:24:18,320 --> 00:24:21,757
get into this debate about these quote
labs or corporations. I don't know if it

446
00:24:21,840 --> 00:24:25,517
even matters anymore. I think open
source is here and I think that the pace

447
00:24:25,600 --> 00:24:28,157
of improvement is such that this is only

448
00:24:28,240 --> 00:24:30,237
going to drive productivity. Yeah.

449
00:24:30,320 --> 00:24:34,877
>> And maybe we are maybe we are seeing
the AI wave underway right now. The

450
00:24:34,960 --> 00:24:36,637
productivity wave, the productivity boom.

451
00:24:36,720 --> 00:24:40,477
>> Yeah. And and to consumers, they
don't care what model they're using. We've

452
00:24:40,560 --> 00:24:45,037
spent the last three years geeking out
about this. And then Grockbot comes out

453
00:24:45,120 --> 00:24:48,797
and then Muse comes out from Meta. And
I've been using both of these products.

454
00:24:48,880 --> 00:24:53,357
And it's the first time that a normie,
a normal person in the world, it's not

455
00:24:53,440 --> 00:24:58,717
meant to be derogatory in any way, can
get value from AI. If you look at how we

456
00:24:58,800 --> 00:25:02,157
all looked at this, running corporations,
investing in them, we took

457
00:25:02,240 --> 00:25:06,077
a different approach to like, okay,
maybe this is going to retire, you know,

458
00:25:06,160 --> 00:25:08,797
some positions, maybe this is um going

459
00:25:08,880 --> 00:25:10,557
to add some positions here or there, but

460
00:25:10,640 --> 00:25:13,277
now you can like with Muse connect to

461
00:25:13,360 --> 00:25:17,837
everything instantly and actually get
real work done in the world if you're a

462
00:25:17,920 --> 00:25:21,917
normal person who wants to get
the benefit from AI. We will be turning a

463
00:25:22,000 --> 00:25:25,197
corner, Freedber, and I think you laid
it out perfectly. We're going to turn a

464
00:25:25,280 --> 00:25:26,717
corner, I think, at the end of this

465
00:25:26,800 --> 00:25:29,917
year, where moms, our uncles, people are

466
00:25:30,000 --> 00:25:35,037
going to start using Muse and Bot,
Grockbot, which are growing very quickly

467
00:25:35,120 --> 00:25:37,837
to just solve real world problems
for that. And they're going to go, "Wait a

468
00:25:37,920 --> 00:25:41,357
second. I could never afford a chief of
a staff. I could never afford a personal

469
00:25:41,440 --> 00:25:45,757
assistant, an executive assistant. Now
I get a free executive assistant." It's

470
00:25:45,840 --> 00:25:49,517
going to blow people's minds, Shimoth,
in the next couple of months when they

471
00:25:49,600 --> 00:25:51,997
start using these products. and the

472
00:25:52,080 --> 00:25:54,957
bottom half of America in this K-shaped

473
00:25:55,040 --> 00:25:58,397
recovery experiences some benefits. Your

474
00:25:58,480 --> 00:26:02,317
thoughts on what's happening here with
these models? Have you started playing

475
00:26:02,400 --> 00:26:05,117
with it? Because you were, you know,
originally a technologist. Have you

476
00:26:05,200 --> 00:26:07,517
loaded these models on your computer?

477
00:26:07,600 --> 00:26:10,237
Because I'm playing with it
on my MacBook Pro and it's awesome.

478
00:26:10,320 --> 00:26:15,037
>> Let me maybe give you the answer
by explaining something. So

479
00:26:15,120 --> 00:26:19,357
I think what we're seeing is that these
models are converging and clustering

480
00:26:19,440 --> 00:26:24,557
which is to say that they're now within
margin of error. And so the thing is

481
00:26:24,640 --> 00:26:27,437
there's so many of these models that are
roughly the same at different price

482
00:26:27,520 --> 00:26:30,157
points. So where is the edge? There's

483
00:26:30,240 --> 00:26:33,037
still edge and the edge is in the

484
00:26:33,120 --> 00:26:35,597
harness that you use to wrap the model.

485
00:26:35,680 --> 00:26:37,917
So if you think about it as like you

486
00:26:38,000 --> 00:26:41,117
have a model that's like a brain but you

487
00:26:41,200 --> 00:26:45,277
need to embody that brain with arms
and legs and hands and eyes to do work.

488
00:26:45,360 --> 00:26:47,357
That's how you identify a model. Okay.

489
00:26:47,440 --> 00:26:51,197
And you make it aic that's what that
phrase means. And there are all kinds of

490
00:26:51,280 --> 00:26:54,957
different harnesses that the industry
has invented to make these models more

491
00:26:55,040 --> 00:26:57,517
agentic. And there's still a lot of edge

492
00:26:57,600 --> 00:27:01,997
there. So at 8090 we have a labs
function and we've been ripping and

493
00:27:02,080 --> 00:27:05,757
teasing apart all the different models
with all the different harnesses just to

494
00:27:05,840 --> 00:27:08,717
see how they perform and they are wildly

495
00:27:08,800 --> 00:27:13,357
variant in terms of cost and quality.
So just to kind of build on Freebrook's

496
00:27:13,440 --> 00:27:17,357
point you have the models that are
converging the harnesses are wildly

497
00:27:17,440 --> 00:27:19,597
different and so their performance is

498
00:27:19,680 --> 00:27:22,717
all over the map and so now companies

499
00:27:22,800 --> 00:27:26,797
and people have to sort of navigate what
are you willing to pay for. I sent this

500
00:27:26,880 --> 00:27:30,237
tweet to Nick, if you can just put it
up here, and I don't know if this is

501
00:27:30,320 --> 00:27:34,157
instructive or accurate, but what this
shows, and this is the case for Ant, but

502
00:27:34,240 --> 00:27:37,917
it's probably the case for a lot of the
closed models, is you're starting to see

503
00:27:38,000 --> 00:27:40,157
this massive revenue concentration,

504
00:27:40,240 --> 00:27:42,877
which is a few folks consume all of the

505
00:27:42,960 --> 00:27:46,477
highest and most expensive tokens. And I

506
00:27:46,560 --> 00:27:47,917
think if you take what Freeberg just

507
00:27:48,000 --> 00:27:50,317
showed to its logical conclusion, those

508
00:27:50,400 --> 00:27:53,597
folks at the top of this

509
00:27:53,680 --> 00:27:57,117
table are going to be under a lot of

510
00:27:57,200 --> 00:28:01,437
internal pressure whether it's from
their finance teams or whether it's from

511
00:28:01,520 --> 00:28:03,437
their CFO or it's going to be from

512
00:28:03,520 --> 00:28:06,477
shareholders to explain

513
00:28:06,560 --> 00:28:09,997
why as this clustering happens they are

514
00:28:10,080 --> 00:28:14,957
not moving down to cheaper and cheaper
models. Now, you may choose to go down

515
00:28:15,040 --> 00:28:19,437
to cheaper and cheaper Ant models,
so then Ant keeps the revenue, but there's

516
00:28:19,520 --> 00:28:22,877
also risk that you may do what Freeberg
just said and just go to an open source

517
00:28:22,960 --> 00:28:25,597
model and just host it yourself.

518
00:28:25,680 --> 00:28:29,357
So, we're in like a very complicated

519
00:28:29,440 --> 00:28:34,477
part of the cycle, which is the first
version of the AI trade was relatively

520
00:28:34,560 --> 00:28:37,837
simple, which is you're just selling
tokens, you're wrapping the tokens,

521
00:28:37,920 --> 00:28:42,157
you're passing it through, and there was
enough disparity that there was value. I

522
00:28:42,240 --> 00:28:44,477
think that that's going away. It'll

523
00:28:44,560 --> 00:28:46,797
force OpenAI and Ant to go up the stack.

524
00:28:46,880 --> 00:28:49,517
There is no choice. You can't sit there

525
00:28:49,600 --> 00:28:51,837
and serve a token that has diminishing

526
00:28:51,920 --> 00:28:54,237
value because then all the subsequent

527
00:28:54,320 --> 00:28:57,597
value is getting absorbed and made by

528
00:28:57,680 --> 00:28:59,677
people wrapping your intelligence token.

529
00:28:59,760 --> 00:29:03,597
You will have to do that work. And so,
Jason, that's why they will have to go

530
00:29:03,680 --> 00:29:07,197
and do cyber. They'll have to go and do
law. They'll have to go and do customer

531
00:29:07,280 --> 00:29:10,477
support. all these things that we
weren't sure whether they were going to

532
00:29:10,560 --> 00:29:14,317
compete. These trends
may force them to do that.

533
00:29:14,400 --> 00:29:17,197
>> Yeah. >> And I think
then that sets up a very

534
00:29:17,280 --> 00:29:20,237
interesting setup for their IPO. Just to

535
00:29:20,320 --> 00:29:22,717
back all that up, Saxs,

536
00:29:22,800 --> 00:29:26,397
>> the Frontier Lab
Corporations, Anthropic

537
00:29:26,480 --> 00:29:28,237
and OpenAI, they both released models

538
00:29:28,320 --> 00:29:30,797
this week at 50%

539
00:29:30,880 --> 00:29:33,517
less token prices.

540
00:29:33,600 --> 00:29:36,637
and their IPOs are looking like they're

541
00:29:36,720 --> 00:29:39,277
both going to get delayed. Anthropic

542
00:29:39,360 --> 00:29:41,357
just targeting a $2 trillion valuation

543
00:29:41,440 --> 00:29:43,677
open 1.2 trillion. Sam already said

544
00:29:43,760 --> 00:29:46,717
OpenAI plans on doing 2027

545
00:29:46,800 --> 00:29:49,597
over because of safety concerns and

546
00:29:49,680 --> 00:29:52,477
Anthropic which was supposed to file and

547
00:29:52,560 --> 00:29:55,357
go out in October. The report from the

548
00:29:55,440 --> 00:29:58,397
Wall Street Journal is that the IPO will

549
00:29:58,480 --> 00:30:01,277
happen in November or may be delayed

550
00:30:01,360 --> 00:30:03,677
even more. Here's your poly market. Will

551
00:30:03,760 --> 00:30:06,957
Anthropic go public in 2026 with our

552
00:30:07,040 --> 00:30:09,517
amazing partner Poly Market? Shout out

553
00:30:09,600 --> 00:30:13,277
to my guy Shane. Peaked at 96% earlier

554
00:30:13,360 --> 00:30:17,037
this month that Enthropic would go
public this year, but now it's fallen to

555
00:30:17,120 --> 00:30:21,037
76%. Is this like
a cell phone here, Sachs?

556
00:30:21,120 --> 00:30:25,277
The the leading model companies need
to go public to get capital to do their

557
00:30:25,360 --> 00:30:27,117
buildout and now they're delaying their

558
00:30:27,200 --> 00:30:32,477
own IPOs. Please explain how any of
this makes sense. And if you were a major

559
00:30:32,560 --> 00:30:36,317
investor in one of these companies
or on the board of them, what would you be

560
00:30:36,400 --> 00:30:40,957
doing or instructing these founders
to do to get these companies to go public

561
00:30:41,040 --> 00:30:44,637
and get that capital they need
to continue the AI buildout?

562
00:30:44,720 --> 00:30:49,117
>> Oh, well, I I think that we know
from our friends that the investors in

563
00:30:49,200 --> 00:30:52,477
Anthropic are pulling out their hair
about what's happening right now. Forget

564
00:30:52,560 --> 00:30:55,357
about the fake whistleblowers who were
there for like a few weeks and then

565
00:30:55,440 --> 00:30:58,077
left. You have current leadership in the

566
00:30:58,160 --> 00:31:01,597
company saying that there's a greater

567
00:31:01,680 --> 00:31:03,677
than 10% chance of causing human

568
00:31:03,760 --> 00:31:06,797
extinction and their own product is

569
00:31:06,880 --> 00:31:09,597
unsolved in this area. So think about

570
00:31:09,680 --> 00:31:11,437
the risk factors that's creating. How do

571
00:31:11,520 --> 00:31:14,157
you IPO with your own management of the

572
00:31:14,240 --> 00:31:18,717
company saying that? Then you've got
I think other evidence of I don't know

573
00:31:18,800 --> 00:31:23,117
what to call it except corporate
schizophrenia. You've got Daario putting

574
00:31:23,200 --> 00:31:25,357
out this essay saying we need to pace

575
00:31:25,440 --> 00:31:27,917
the frontier days before they launch

576
00:31:28,000 --> 00:31:31,757
Claude 5.5 and set a new frontier. I

577
00:31:31,840 --> 00:31:36,397
mean, he must know when he's writing
this essay that they're about to release

578
00:31:36,480 --> 00:31:38,317
a product that will extend the frontier.

579
00:31:38,400 --> 00:31:42,637
So, at a minimum, it's hypocritical,
but maybe it's even worse than that. Then

580
00:31:42,720 --> 00:31:47,437
you have them publishing an essay about
the biorisk associated with AI and how

581
00:31:47,520 --> 00:31:48,717
that needs to be stopped. And then

582
00:31:48,800 --> 00:31:51,837
meanwhile, they announced a new biolab

583
00:31:51,920 --> 00:31:54,477
in San Francisco, a new wet lab.

584
00:31:54,560 --> 00:31:57,677
>> They started a lab
after telling us that

585
00:31:57,760 --> 00:32:00,717
there's a 10% chance of us all dying.

586
00:32:00,800 --> 00:32:02,477
Sachs, let's just pause here for a

587
00:32:02,560 --> 00:32:06,477
second. Is this a dereliction of duty as

588
00:32:06,560 --> 00:32:10,317
the chief executive officer of Anthropic

589
00:32:10,400 --> 00:32:12,717
to sabotage your own IPO? You're

590
00:32:12,800 --> 00:32:14,637
supposed to be acting on behalf of your

591
00:32:14,720 --> 00:32:17,597
CEO of your shareholders. This to me

592
00:32:17,680 --> 00:32:21,677
seems like if we were the board, if
the four of us were somehow magically the

593
00:32:21,760 --> 00:32:23,677
board of Anthropic, we would replace the

594
00:32:23,760 --> 00:32:26,317
CEO. We would put the CEO in the lab and

595
00:32:26,400 --> 00:32:29,037
we'd find a CEO who knows how to run the

596
00:32:29,120 --> 00:32:33,517
company and not >> sabotage it.
He's literally sabotaging

597
00:32:33,600 --> 00:32:37,277
his own company. >> I think
there's an element of that and I

598
00:32:37,360 --> 00:32:40,877
think investors are pretty nervous
about giving the founders or or Daria

599
00:32:40,960 --> 00:32:42,157
specifically super voting shares.

600
00:32:42,240 --> 00:32:45,917
There's a story in the information about
that that there's an active debate right

601
00:32:46,000 --> 00:32:47,757
now about whether to give
them super voting share.

602
00:32:47,840 --> 00:32:50,957
>> Explain what those are, please, and
and why in this context this is critically

603
00:32:51,040 --> 00:32:53,837
important. Our audience may
or may not know what those are.

604
00:32:53,920 --> 00:32:56,157
>> Supposedly, the founders of Anthropic

605
00:32:56,240 --> 00:32:59,517
only have about 2% ownership each of the

606
00:32:59,600 --> 00:33:03,517
company, which in the grand scheme
of things is a relatively low number

607
00:33:03,600 --> 00:33:06,637
relative to what other founders
typically have at time of IPO. It'll

608
00:33:06,720 --> 00:33:10,157
still be an enormous amount of money,
you know, tens of billions of dollars,

609
00:33:10,240 --> 00:33:11,837
but that is the case. Super voting

610
00:33:11,920 --> 00:33:14,157
shares basically say it separates the

611
00:33:14,240 --> 00:33:18,877
economic shares from the voting shares
in the company and gives certain people

612
00:33:18,960 --> 00:33:23,037
the right to control the company even
if they don't have the economic ownership

613
00:33:23,120 --> 00:33:27,277
that would usually be necessary. It's
a pretty common thing. I mean, Larry and

614
00:33:27,360 --> 00:33:31,037
Sergey, I think maybe were the first
prominent founders to get super voting

615
00:33:31,120 --> 00:33:35,277
shares in Google and Zuck. It can be
a stabilizing thing for the company

616
00:33:35,360 --> 00:33:39,677
because it means that as long as the
founders are there overseeing things,

617
00:33:39,760 --> 00:33:43,437
they're not going to be >> subject to,
you know, like takeover

618
00:33:43,520 --> 00:33:47,197
attempts, things like that. But on the
other hand, it requires a tremendous

619
00:33:47,280 --> 00:33:50,637
degree of trust in those founders
because they effectively can't be

620
00:33:50,720 --> 00:33:53,757
removed no matter what they do. And you

621
00:33:53,840 --> 00:33:56,157
just have to wonder if in this case

622
00:33:56,240 --> 00:34:01,357
that's earned and warranted. I have
to say that even for someone like me who

623
00:34:01,440 --> 00:34:04,477
isn't necessarily advocating for a lot
more regulation, I kind of get nervous

624
00:34:04,560 --> 00:34:09,357
about the idea of anthropic opening
a wet lab in San Francisco. I mean, I was

625
00:34:09,440 --> 00:34:11,837
>> great timing.
>> I was thinking about moving back and I

626
00:34:11,920 --> 00:34:15,197
don't know if I want to be I mean like
can you just put it a little further

627
00:34:15,280 --> 00:34:17,117
away? I think we've all
seen this movie before.

628
00:34:17,200 --> 00:34:18,607
>> Yeah. Offshore maybe. I don't know.

629
00:34:18,690 --> 00:34:22,957
[laughter] Put it on an oil rig or
something. What I would say is that if I

630
00:34:23,040 --> 00:34:27,037
were on the board, I would keep Daario
as a CEO. And the reason is that he's

631
00:34:27,120 --> 00:34:29,037
clearly demonstrated that he can build a

632
00:34:29,120 --> 00:34:32,237
unique culture and position against the

633
00:34:32,320 --> 00:34:35,437
otherwise most formidable corporation

634
00:34:35,520 --> 00:34:39,517
competitor, which is OpenAI. And he's
done that and he's come from behind. And

635
00:34:39,600 --> 00:34:43,917
I don't think you should
undervalue how hard that is and what

636
00:34:44,000 --> 00:34:47,837
that takes. So he's done that and he's
built he's built essentially the

637
00:34:47,920 --> 00:34:50,557
greatest business ramp in all time. I

638
00:34:50,640 --> 00:34:52,637
think the thing that I would probably do

639
00:34:52,720 --> 00:34:55,517
if I was on the board is fix what Sax

640
00:34:55,600 --> 00:34:58,077
just said, which is there's now a bunch

641
00:34:58,160 --> 00:35:00,477
of incremental liquidity risk to the

642
00:35:00,560 --> 00:35:03,597
company. And I think if you take the

643
00:35:03,680 --> 00:35:06,317
20-year view on anthropic, that

644
00:35:06,400 --> 00:35:08,957
corporation needs hundreds of billions

645
00:35:09,040 --> 00:35:11,437
of dollars to fulfill their ambitions.

646
00:35:11,520 --> 00:35:15,757
And I would solve for getting that
as risklessly as possible. And I think

647
00:35:15,840 --> 00:35:17,837
there's only two things that you can do.

648
00:35:17,920 --> 00:35:23,037
The first is you have to throw everything
in and the kitchen sink into disclosures.

649
00:35:23,120 --> 00:35:27,437
The problem is that that will make what
is an otherwise dense and impenetrable

650
00:35:27,520 --> 00:35:29,837
S1 even more dense and impenetrable.

651
00:35:29,920 --> 00:35:33,677
Okay? So, you can kind of like
superficially cover your regulatory

652
00:35:33,760 --> 00:35:37,917
risk. But the way that it actually gets
translated in the market is you're just

653
00:35:38,000 --> 00:35:39,837
going to have to water down the

654
00:35:39,920 --> 00:35:44,157
expectations of the IPO sellers and

655
00:35:44,240 --> 00:35:49,037
create a much larger margin of safety
for the IPO buyer. And that's

656
00:35:49,120 --> 00:35:52,397
effectively how you find a market
clearing price for this, Jason, which is

657
00:35:52,480 --> 00:35:55,517
to essentially say, and don't bear with
me, I'm just giving out rough numbers,

658
00:35:55,600 --> 00:35:59,917
right? Even if you think that they're
on a hundred billion run rate, what would

659
00:36:00,000 --> 00:36:02,317
otherwise in a very clean sheet IPO may

660
00:36:02,400 --> 00:36:04,717
be would have been a $2 trillion market

661
00:36:04,800 --> 00:36:07,757
cap, now you know, you're at one or

662
00:36:07,840 --> 00:36:11,997
less. And the reason is the hedge funds

663
00:36:12,080 --> 00:36:16,877
and the pension systems and the long

664
00:36:16,960 --> 00:36:19,917
only mutual fund complexes who have a

665
00:36:20,000 --> 00:36:22,717
fiduciary responsibility to their

666
00:36:22,800 --> 00:36:27,677
sources of capital will
justify all the risks by

667
00:36:27,760 --> 00:36:31,357
essentially asking for a lower and lower
price. And I think that's what you're

668
00:36:31,440 --> 00:36:32,477
going to see. I think you're going to

669
00:36:32,560 --> 00:36:36,637
see a much more moderate and reasonable

670
00:36:36,720 --> 00:36:41,517
IPO motion here. And it's
probably going to clear at a

671
00:36:41,600 --> 00:36:43,117
much lower price than anybody thinks.

672
00:36:43,200 --> 00:36:46,397
And I do think that what Sach says
is the reason why. But I think it's good

673
00:36:46,480 --> 00:36:51,437
for Ant because it clears all this noise.
It forces them to just be a

674
00:36:51,520 --> 00:36:53,117
corporation and then run the business.

675
00:36:53,200 --> 00:36:56,237
And I think that they'll do great. Look,
I'll be honest with you. Like I've tried

676
00:36:56,320 --> 00:36:59,277
to recruit against Ant. I can't. It's

677
00:36:59,360 --> 00:37:03,597
impossible. The comp that they give
is incredible. The business ramp is so

678
00:37:03,680 --> 00:37:05,837
compelling. It's like
when you're in a head

679
00:37:05,920 --> 00:37:07,197
>> when you're in a
head-to-head with these

680
00:37:07,280 --> 00:37:11,277
guys. I have not won a single bake off
for folks. I wish I could say that I

681
00:37:11,360 --> 00:37:13,677
did, but I have not. And it's hard for a

682
00:37:13,760 --> 00:37:16,077
startup. So, they are a magnet for

683
00:37:16,160 --> 00:37:19,037
talent. OpenAI is a magnet for talent.

684
00:37:19,120 --> 00:37:23,517
But these other risks are going to weigh
on the IPO and the only way to flush

685
00:37:23,600 --> 00:37:25,197
them is to just bring the price way down.

686
00:37:25,280 --> 00:37:27,117
>> Yeah. And Freeberg, when we look at the

687
00:37:27,200 --> 00:37:29,757
risk factors in an S1, those are

688
00:37:29,840 --> 00:37:32,717
prefuncter. Those are disclaimers.

689
00:37:32,800 --> 00:37:36,637
Typically, you get the SpaceX one. Oh,
what if Starship doesn't work? What if

690
00:37:36,720 --> 00:37:41,757
there's competition, it's kind of check
boxes in anthropics, I think we're going

691
00:37:41,840 --> 00:37:43,437
to be looking at something totally

692
00:37:43,520 --> 00:37:46,237
different, which is, hey, we and our

693
00:37:46,320 --> 00:37:50,877
employees are quitting saying this is
going to kill humanity. We have massive

694
00:37:50,960 --> 00:37:53,117
regulation. We're asking for regulation.

695
00:37:53,200 --> 00:37:57,677
And by the way, we we're doing biology
work and we're launching a lab. So do

696
00:37:57,760 --> 00:37:59,757
you agree with Chimath here that the S1

697
00:37:59,840 --> 00:38:03,597
is going to have an actual risk factors

698
00:38:03,680 --> 00:38:06,237
and disclaimers that are material

699
00:38:06,320 --> 00:38:10,797
against the IPO free burn? I think this

700
00:38:10,880 --> 00:38:13,197
customer concentration point ties to the

701
00:38:13,280 --> 00:38:15,997
earlier comments I made about the

702
00:38:16,080 --> 00:38:18,157
proliferation of incredibly performant

703
00:38:18,240 --> 00:38:20,717
openweight and open source models. And

704
00:38:20,800 --> 00:38:23,517
the reason is Anthropic puts out the

705
00:38:23,600 --> 00:38:27,997
best models for life sciences today.

706
00:38:28,080 --> 00:38:31,197
I run a life sciences R&D organization.

707
00:38:31,280 --> 00:38:34,237
We use their models. They're incredible.

708
00:38:34,320 --> 00:38:36,477
They're better than other models. And so

709
00:38:36,560 --> 00:38:38,797
we want to use them. When it comes to

710
00:38:38,880 --> 00:38:42,157
writing code, to writing software to do

711
00:38:42,240 --> 00:38:45,197
internal workflows and run operations,

712
00:38:45,280 --> 00:38:49,997
we can use other products to write that
code for us. And we do. And I think that

713
00:38:50,080 --> 00:38:53,117
that's likely the way the market
is going to develop. And I think we're

714
00:38:53,200 --> 00:38:56,317
starting to see that proof come out in
the market. And you're seeing it in this

715
00:38:56,400 --> 00:38:59,197
customer concentration point. Extremely

716
00:38:59,280 --> 00:39:03,597
high value technical engineering, math,

717
00:39:03,680 --> 00:39:05,997
life sciences, those really hard

718
00:39:06,080 --> 00:39:10,957
technical problems. You want to have
these top tier, topshelf models, and

719
00:39:11,040 --> 00:39:13,917
you're going to pay a huge premium
for them. And you're totally going to make

720
00:39:14,000 --> 00:39:17,357
money doing it because the value
is extraordinary. And this has always been

721
00:39:17,440 --> 00:39:19,917
my case about AI. AI is not so much

722
00:39:20,000 --> 00:39:22,957
about the value of replacing old stuff.

723
00:39:23,040 --> 00:39:27,117
99% of the value of AI is about enabling
new stuff that's never been possible in

724
00:39:27,200 --> 00:39:30,957
human history. And I think that's where
anthropic and open AI are going to make

725
00:39:31,040 --> 00:39:34,797
their money. Separately, I think open
AI is going to make a lot of money in

726
00:39:34,880 --> 00:39:38,797
consumer and anthropic maybe to some
extent as well. But in the enterprise

727
00:39:38,880 --> 00:39:43,757
setting, I think the money comes from
creating new frontiers, discovering new

728
00:39:43,840 --> 00:39:45,517
enzymes, which Anthropic announced this

729
00:39:45,600 --> 00:39:48,557
week, that can cure human disease, and

730
00:39:48,640 --> 00:39:52,957
developing new technology, new
engineering systems, making things that

731
00:39:53,040 --> 00:39:56,637
have never been possible to be made
before. And I think that's the premium

732
00:39:56,720 --> 00:40:01,277
value, and it's the premium price you'll
pay for those closed models and that

733
00:40:01,360 --> 00:40:05,037
premium model. But that's a different
business case than taking all of

734
00:40:05,120 --> 00:40:09,437
enterprise. separately. I think most of
the enterprise applications are going to

735
00:40:09,520 --> 00:40:13,437
be run using these openweight models and
code is going to be written with cheaper

736
00:40:13,520 --> 00:40:17,997
models and so on. I don't know why you
would go to a higherend model to write

737
00:40:18,080 --> 00:40:21,757
code when you can do it for free
and everyone in the organization can write

738
00:40:21,840 --> 00:40:26,157
code basically for free. And so I
think anthropic's bigger issue is this

739
00:40:26,240 --> 00:40:30,077
customer concentration question and is it
really a premium product that they're

740
00:40:30,160 --> 00:40:32,797
selling in the market and that's where
they take the market. That would be kind

741
00:40:32,880 --> 00:40:35,517
of where I think the market is headed
right now and would be the biggest risk

742
00:40:35,600 --> 00:40:38,877
factor. This whole thing about oh it's
gonna end the world and whatnot. It's

743
00:40:38,960 --> 00:40:41,837
like dude just make sure your product
doesn't end the world and you'll be

744
00:40:41,920 --> 00:40:46,237
fine. That would be a good first
step here like Mark is doing at Meta.

745
00:40:46,320 --> 00:40:49,917
>> I mean it's an incredible company.
The execution has been extraordinary. The

746
00:40:50,000 --> 00:40:53,917
ramp of revenue sachs has been
extraordinary. But when you start to

747
00:40:54,000 --> 00:40:56,477
look at the impact of open source, we

748
00:40:56,560 --> 00:41:00,797
had a a couple of um viral charts come

749
00:41:00,880 --> 00:41:03,837
out, 70 80% of the token usage are those

750
00:41:03,920 --> 00:41:08,077
dark tokens I mentioned six months ago
here on this pod that are not tracked,

751
00:41:08,160 --> 00:41:12,237
but some people can track them on routers.
Those dark open source tokens

752
00:41:12,320 --> 00:41:16,557
are now the majority of tokens. And >>
can I just give you a stat on that, Jal?

753
00:41:16,640 --> 00:41:18,157
cuz what you just said is really

754
00:41:18,240 --> 00:41:19,597
important, but there's a stat that I

755
00:41:19,680 --> 00:41:23,597
think in in the last 12 weeks alone

756
00:41:23,680 --> 00:41:27,197
>> token use has flipped from 8020 closed

757
00:41:27,280 --> 00:41:30,957
verse open >> to 8020 open verse closed.

758
00:41:31,040 --> 00:41:33,117
>> I don't think we've ever seen in 12

759
00:41:33,200 --> 00:41:38,077
weeks I don't think we've
ever seen a title wave of chart

760
00:41:38,160 --> 00:41:41,517
>> in the history of all technology
markets. We've never seen anything like

761
00:41:41,600 --> 00:41:45,517
this. And that is the biggest risk
factor for Anthropic is they were first

762
00:41:45,600 --> 00:41:47,917
out the gate to get the enterprise on uh

763
00:41:48,000 --> 00:41:50,957
on AI to build tools and whatnot. But as

764
00:41:51,040 --> 00:41:55,197
this market shifts to open source
and open weights, boom, here it is right

765
00:41:55,280 --> 00:41:57,437
here. Look at this. >> This is
the Versel chart that went viral

766
00:41:57,520 --> 00:42:00,237
this week. So this week
12 weeks, oh my god,

767
00:42:00,320 --> 00:42:02,477
>> crazy. If you look
at this chart, this I

768
00:42:02,560 --> 00:42:05,677
think is what blows the lid off the door

769
00:42:05,760 --> 00:42:09,917
on what needs to go into these S1s.
And that's why I walked through all those

770
00:42:10,000 --> 00:42:14,157
openweight releases at the start of the
show or this segment is it's not just

771
00:42:14,240 --> 00:42:17,277
that they're there, it's that they're
proliferating now. There's so many more

772
00:42:17,360 --> 00:42:20,637
of them coming out and the pace of
improvement is extraordinary and they're

773
00:42:20,720 --> 00:42:23,997
now in VLA for robotics. They're in

774
00:42:24,080 --> 00:42:28,477
image generation for image and video.
So the openweight open source models are

775
00:42:28,560 --> 00:42:30,957
now being used for all of the broader AI

776
00:42:31,040 --> 00:42:33,677
applications. The premium stuff I will

777
00:42:33,760 --> 00:42:36,877
say is still the best for narrow use

778
00:42:36,960 --> 00:42:40,717
cases. Yes, >> for engineering,
for highly technical,

779
00:42:40,800 --> 00:42:45,277
highly complicated tasks and workflows
and problems, solving Navier Stokes,

780
00:42:45,360 --> 00:42:48,877
solving mathematical problems, solving
biology problems, that's where these

781
00:42:48,960 --> 00:42:53,757
models absolutely perform and people
will pay a premium of, I would say,

782
00:42:53,840 --> 00:42:57,837
close to infinity dollars for that
premium application set. So, it doesn't

783
00:42:57,920 --> 00:42:59,437
mean that their market is destroyed.

784
00:42:59,520 --> 00:43:01,277
Yeah, >> it means that we're
seeing a bifurcation

785
00:43:01,360 --> 00:43:06,157
in the market where the majority of use
cases of AI go towards these open source

786
00:43:06,240 --> 00:43:09,677
open tokens. >> The question
and premium models are

787
00:43:09,760 --> 00:43:11,037
still going to proliferate. Yeah.

788
00:43:11,120 --> 00:43:13,277
>> The question then
is in the distribution

789
00:43:13,360 --> 00:43:15,357
of tokens in these frontier

790
00:43:15,440 --> 00:43:18,317
corporations, what percentage are those

791
00:43:18,400 --> 00:43:20,957
tasks versus what percentage are

792
00:43:21,040 --> 00:43:23,037
fungeable tasks to open source? And if

793
00:43:23,120 --> 00:43:26,557
the answer is more than 60 or 70%.

794
00:43:26,640 --> 00:43:31,277
You have to discount that revenue. If
the answer is only 10% and 90% of all of

795
00:43:31,360 --> 00:43:35,917
the revenue is being generated
on absolute frontier use cases, they're

796
00:43:36,000 --> 00:43:38,477
going to be fine and they're
going to be off to the races.

797
00:43:38,560 --> 00:43:41,037
>> Yeah. And >> but otherwise,
it's a thing where you're

798
00:43:41,120 --> 00:43:46,317
fighting just people's mis or lack
of education, right? Because as they see

799
00:43:46,400 --> 00:43:48,797
their bills, >> hand it
to Sax here. Yeah. Go finish

800
00:43:48,880 --> 00:43:50,237
your thought and then I'm
going to hand it to Sax.

801
00:43:50,320 --> 00:43:53,917
>> No, just move the move the workloads
where they belong and save the money.

802
00:43:54,000 --> 00:43:56,077
>> Yeah. And so Saxs, um I made a chart

803
00:43:56,160 --> 00:43:57,517
here before the show here, but here's

804
00:43:57,600 --> 00:44:01,677
your um sweet spot for the last 100 days

805
00:44:01,760 --> 00:44:05,197
of models, which um Friber, you did
a good job of teeing up. And what you'll

806
00:44:05,280 --> 00:44:09,597
see is like yes, the Clouds and the
Astros up in that righth hand corner

807
00:44:09,680 --> 00:44:12,717
cost a lot and they deliver a lot. But
then you start going down and you start

808
00:44:12,800 --> 00:44:15,917
to see Muse and GLM and Kimmy and then

809
00:44:16,000 --> 00:44:20,877
on the left you start seeing GLM from ZI
and Mimo and you're starting to see well

810
00:44:20,960 --> 00:44:23,357
the Chinese >> this is all
hosted right JL. I mean,

811
00:44:23,440 --> 00:44:27,517
it's important to note what you're
showing here is the hosted price, but if

812
00:44:27,600 --> 00:44:30,957
you run it yourself, the cost on some
of these is less than 10 cents for a

813
00:44:31,040 --> 00:44:33,597
million tokens. >> Yeah.
So, you'll pick up then the

814
00:44:33,680 --> 00:44:37,357
compute, you pick up the bandwidth,
all that other stuff. But, Sax, the issue

815
00:44:37,440 --> 00:44:40,157
here to Chimat's point is the vertical

816
00:44:40,240 --> 00:44:42,397
language models, the small language

817
00:44:42,480 --> 00:44:45,197
models are going to do 80 or 90% of the

818
00:44:45,280 --> 00:44:49,757
tasks that many corporations need.
And then we get to AI sovereignty. Do you

819
00:44:49,840 --> 00:44:54,157
want your super intelligence? Shout out
President Trump. from intelligence chat

820
00:44:54,240 --> 00:44:56,237
DJT.

821
00:44:56,320 --> 00:44:59,837
We need people to use those models so

822
00:44:59,920 --> 00:45:03,277
they have sovereignty. So you're not
giving your data to the frontier models

823
00:45:03,360 --> 00:45:05,837
so that they don't get that tip of the

824
00:45:05,920 --> 00:45:08,877
spear super intelligence edge which is

825
00:45:08,960 --> 00:45:11,917
why people are disengaging. So your

826
00:45:12,000 --> 00:45:13,917
thoughts on maybe this is going to start

827
00:45:14,000 --> 00:45:17,197
to impact the velocity of clawed and

828
00:45:17,280 --> 00:45:19,197
openi tokens. What does your business

829
00:45:19,280 --> 00:45:22,077
sense tell you David Saxs? Well, look,

830
00:45:22,160 --> 00:45:25,357
wearing my hat as an investor, I'm more

831
00:45:25,440 --> 00:45:30,317
optimistic about the Frontier
Intelligence business than I think you

832
00:45:30,400 --> 00:45:32,317
guys are. I think that Anthropic and

833
00:45:32,400 --> 00:45:35,997
Open AI right now are a stable duopoly

834
00:45:36,080 --> 00:45:40,077
for frontier intelligence because they
have a pretty significant lead over all

835
00:45:40,160 --> 00:45:44,237
these other companies and they are able
to charge a premium for that. I fully

836
00:45:44,320 --> 00:45:46,637
understand that the majority of tokens

837
00:45:46,720 --> 00:45:52,077
will go to open models or commodity
intelligence. However, there is a I

838
00:45:52,160 --> 00:45:54,877
don't know meaningful percentage of the
market. I don't know whether it's 10%,

839
00:45:54,960 --> 00:45:57,997
20, 30, whatever that will pay this huge

840
00:45:58,080 --> 00:46:00,957
premium for true frontier intelligence.

841
00:46:01,040 --> 00:46:04,157
There's a lot of customers that either
need the best or they just want to know

842
00:46:04,240 --> 00:46:08,557
they have the best. You know, they don't
want this DIY approach or they could be

843
00:46:08,640 --> 00:46:12,157
in a very competitive market. Let's say
you're a hedge fund. You can't take the

844
00:46:12,240 --> 00:46:15,837
risk that your competitor has a better
model than you do. So there's a lot of

845
00:46:15,920 --> 00:46:20,157
circumstances I think where they will
continue to monetize the frontier at a

846
00:46:20,240 --> 00:46:23,517
premium and this is why you're seeing
their revenue keep growing even though

847
00:46:23,600 --> 00:46:25,997
they're losing token market share.

848
00:46:26,080 --> 00:46:27,757
Furthermore, you got to remember that

849
00:46:27,840 --> 00:46:30,637
over the next year or so something like

850
00:46:30,720 --> 00:46:33,437
60% of the worldwide compute that's

851
00:46:33,520 --> 00:46:35,997
being added is being added for these two

852
00:46:36,080 --> 00:46:40,957
companies. So to the extent that compute
gets scarce and we're kind of in a

853
00:46:41,040 --> 00:46:44,317
compute crunch, just the fact that
they've made these huge investments,

854
00:46:44,400 --> 00:46:48,397
they're building out so much capacity
and they have those economies of scale

855
00:46:48,480 --> 00:46:53,037
and remember they're bringing down their
own prices very quickly as well. I think

856
00:46:53,120 --> 00:46:54,637
that gives them a pretty big advantage.

857
00:46:54,720 --> 00:46:57,277
But look, these two companies are on a

858
00:46:57,360 --> 00:46:59,917
hamster wheel. You know, the moment

859
00:47:00,000 --> 00:47:01,837
where they stop being frontier, they go

860
00:47:01,920 --> 00:47:04,957
to zero, right? And we know the frontier

861
00:47:05,040 --> 00:47:08,157
is maybe only 6 to 12 months ahead of

862
00:47:08,240 --> 00:47:10,237
the commodity portion of the market

863
00:47:10,320 --> 00:47:13,197
depending on capability. So, you know,

864
00:47:13,280 --> 00:47:18,077
again, if they slip, if they fall off
that hamster wheel for 6 months, then

865
00:47:18,160 --> 00:47:19,437
they're in deep trouble as a company.

866
00:47:19,520 --> 00:47:22,237
And that I think is a central risk
factor that's going to have to be priced

867
00:47:22,320 --> 00:47:24,557
in. And this is where I think that

868
00:47:24,640 --> 00:47:29,677
frankly their government affairs efforts
are miscalculating badly. I can see why

869
00:47:29,760 --> 00:47:32,637
they they're going for regulatory
capture, right? because being on a

870
00:47:32,720 --> 00:47:37,037
hamster wheel gets really tiring. They
probably think, hm, like, let's try and

871
00:47:37,120 --> 00:47:41,597
bake in our moes. Let's try and slow
everyone else down as well. But I think

872
00:47:41,680 --> 00:47:46,397
they've miscalculated because if they
actually get all the regulatory controls

873
00:47:46,480 --> 00:47:49,757
they want, if they create this new
federal department of AI, it's going to

874
00:47:49,840 --> 00:47:53,517
slow them down enough that they will no
longer be frontier. Everyone else will

875
00:47:53,600 --> 00:47:57,197
catch up to them. Especially because
a lot of these open models are being

876
00:47:57,280 --> 00:48:01,037
generated by Chinese companies who are
not subject to our jurisdiction. And the

877
00:48:01,120 --> 00:48:04,717
Chinese have made abundantly clear they
are not slowing down. So think about it.

878
00:48:04,800 --> 00:48:09,197
They are lobbying for things that could
slow them down and knock them off the

879
00:48:09,280 --> 00:48:13,517
frontier in which case their business
goes to zero. I you know it's hard to

880
00:48:13,600 --> 00:48:16,637
understand. >> They're
flapping the competitors in this

881
00:48:16,720 --> 00:48:18,637
race and they're asking hey can you pull

882
00:48:18,720 --> 00:48:21,757
us over and inspect the engine chimoth

883
00:48:21,840 --> 00:48:25,117
and that makes no logical sense on top

884
00:48:25,200 --> 00:48:28,237
of that chimoth. You have very

885
00:48:28,320 --> 00:48:29,997
significant competition not just from

886
00:48:30,080 --> 00:48:32,237
the Chinese but you have Jensen and

887
00:48:32,320 --> 00:48:36,557
Nvidia. You have also Microsoft Satia

888
00:48:36,640 --> 00:48:40,317
at Google and Muse at Zuckerberg's

889
00:48:40,400 --> 00:48:45,437
house. They're all embracing open-
source models and they all have

890
00:48:45,520 --> 00:48:47,757
hosting now. >> Yeah.
Can I [clears throat] just go back

891
00:48:47,840 --> 00:48:52,237
to what Sax said which is a really
important and nuance point which I just

892
00:48:52,320 --> 00:48:53,997
want to double click on. He said,

893
00:48:54,080 --> 00:48:57,757
"Imagine you're a hedge fund and you

894
00:48:57,840 --> 00:49:01,677
need the absolute latest and greatest
because you are worried that one of your

895
00:49:01,760 --> 00:49:03,757
competitors has it." This is a really

896
00:49:03,840 --> 00:49:06,077
interesting question because it's about

897
00:49:06,160 --> 00:49:09,437
game theory and your P&L. So now you're

898
00:49:09,520 --> 00:49:13,757
in this game of chicken with your
competitors. You need to use the best

899
00:49:13,840 --> 00:49:17,357
because you're afraid that they could
use the best and out compete you. But

900
00:49:17,440 --> 00:49:20,237
what is the problem right now? What we

901
00:49:20,320 --> 00:49:23,037
know is that you can consume an

902
00:49:23,120 --> 00:49:26,157
inordinate amount of tokens, right? We

903
00:49:26,240 --> 00:49:29,037
jokingly call it token maxing.

904
00:49:29,120 --> 00:49:32,637
That cost is completely not levered or

905
00:49:32,720 --> 00:49:36,157
attached to your revenue. It's just not.

906
00:49:36,240 --> 00:49:41,197
And so all of a sudden, you have this
very weird dynamic where you're like,

907
00:49:41,280 --> 00:49:45,357
okay, well, am I supposed to use this
latest and greatest thing that's 10 to

908
00:49:45,440 --> 00:49:50,077
20 to 30 times more expensive than
a generic thing? that thing isn't tied to

909
00:49:50,160 --> 00:49:54,637
my revenues, it's not like I can pass
that through, right? So, if a hedge fund

910
00:49:54,720 --> 00:49:56,957
only makes end number of dollars per

911
00:49:57,040 --> 00:49:59,677
month of profits, there are a lot of

912
00:49:59,760 --> 00:50:04,877
scenarios where they can become break
even to unprofitable. Now, take a

913
00:50:04,960 --> 00:50:09,437
different example. Let's just say that
you sell a fixed good at a fixed price.

914
00:50:09,520 --> 00:50:12,637
Yet again, I think you have this example
where you could use the latest and

915
00:50:12,720 --> 00:50:14,717
greatest and if you're not able to pass

916
00:50:14,800 --> 00:50:18,237
those costs through, inflate your cost

917
00:50:18,320 --> 00:50:23,197
to absorb the cost of that service,
you're in a very tough spot. So, I'm not

918
00:50:23,280 --> 00:50:25,117
sure, Sax, I agree with this because I

919
00:50:25,200 --> 00:50:26,797
don't see the pricing power of these

920
00:50:26,880 --> 00:50:30,317
companies to both pay for a ton of these

921
00:50:30,400 --> 00:50:35,037
super expensive tokens and also pass
that through to their customers. And if

922
00:50:35,120 --> 00:50:38,397
you can't, they're just absorbing it and
you're going to see it in lower margins.

923
00:50:38,480 --> 00:50:40,637
And to build on that, Saxs, I'm going to

924
00:50:40,720 --> 00:50:42,317
drop it back to you. If you take Jane

925
00:50:42,400 --> 00:50:45,517
Street, very famous uh trading firm,

926
00:50:45,600 --> 00:50:49,197
they've publicly announced 19 billion

927
00:50:49,280 --> 00:50:52,317
dollars in cloud capacity contracts.

928
00:50:52,400 --> 00:50:56,637
Coreweave, 6 billion in in cloud
commitments. They also invested in that

929
00:50:56,720 --> 00:50:59,197
company and then 13 billion with Crusoe.

930
00:50:59,280 --> 00:51:03,597
They are building their own
infrastructure and they're embracing

931
00:51:03,680 --> 00:51:06,797
open source. So that speaks to Chimath
what you're saying. They're are they

932
00:51:06,880 --> 00:51:10,237
hedging their bets? They're going
to spend tens of billions of dollars on

933
00:51:10,320 --> 00:51:14,637
their own compute and they're going
to use the frontier models they seem to

934
00:51:14,720 --> 00:51:17,037
when you do the game theory,

935
00:51:17,120 --> 00:51:19,117
you know, saying, "Hey, let's embrace

936
00:51:19,200 --> 00:51:22,237
our own infrastructure sacks."

937
00:51:22,320 --> 00:51:24,157
>> Well, look, I think when your use cases

938
00:51:24,240 --> 00:51:26,237
become more mature and, you know, you

939
00:51:26,320 --> 00:51:29,277
can get by with cheaper commodity

940
00:51:29,360 --> 00:51:34,157
intelligence, then obviously you'll try
and build that out again if you have the

941
00:51:34,240 --> 00:51:38,157
sophistication level to do it. I just
think it's easy to underestimate how

942
00:51:38,240 --> 00:51:43,277
convenient it is just to use the the
frontier. And again, these companies are

943
00:51:43,360 --> 00:51:47,037
very rapidly bringing down their prices,
too. But anyway, look, we we've talked

944
00:51:47,120 --> 00:51:48,557
about this. I think at the end of the

945
00:51:48,640 --> 00:51:51,357
day that look, there's no question that

946
00:51:51,440 --> 00:51:55,757
opensource is a huge risk factor to

947
00:51:55,840 --> 00:52:00,557
Anthropics S1 and it has to be right
up there with, you know, the risk that

948
00:52:00,640 --> 00:52:03,997
they're going to end humanity or something
like that. But again, I I come

949
00:52:04,080 --> 00:52:07,517
back to the fact that I think the
fundamental problem with this company is

950
00:52:07,600 --> 00:52:09,437
that it's schizophrenic. I mean, I think

951
00:52:09,520 --> 00:52:12,477
they need a psychiatrist, not a banker.

952
00:52:12,560 --> 00:52:17,197
They keep doing things that are
hypocritical and oxymoronic. On the one

953
00:52:17,280 --> 00:52:21,357
hand, they're saying pace of Frontier
as they release a new frontier model. They

954
00:52:21,440 --> 00:52:24,077
say that biorisk can end humanity as

955
00:52:24,160 --> 00:52:26,717
they open a wet lab in San Francisco.

956
00:52:26,800 --> 00:52:29,677
They say well their business depends on

957
00:52:29,760 --> 00:52:33,757
them staying ahead of the commodity

958
00:52:33,840 --> 00:52:38,237
models and they are advocating
for a federal department of AI that will

959
00:52:38,320 --> 00:52:40,717
definitely slow them down perhaps to the

960
00:52:40,800 --> 00:52:43,517
point where they get commoditized. So

961
00:52:43,600 --> 00:52:45,997
you have a fundamental

962
00:52:46,080 --> 00:52:49,757
I'd say like breakdown happening in the

963
00:52:49,840 --> 00:52:51,597
leadership of this company where they

964
00:52:51,680 --> 00:52:56,797
are advocating for things that are
likely not in their interest. Now

965
00:52:56,880 --> 00:52:59,677
there's a lot of people who say it's
just regulatory capture and that is true

966
00:52:59,760 --> 00:53:03,277
but I think that the red capture
is the government affairs people trying to

967
00:53:03,360 --> 00:53:07,917
rationalize the schizophrenia and make
it work to their advantage. I'm not sure

968
00:53:08,000 --> 00:53:09,677
that it fully computes

969
00:53:09,760 --> 00:53:12,557
>> so to speak. Yeah.
Let me ask you guys a

970
00:53:12,640 --> 00:53:14,957
question. What do you think would be the

971
00:53:15,040 --> 00:53:19,677
method by which like a ban on super
intelligence? Bernie Sanders proposed

972
00:53:19,760 --> 00:53:23,437
this bill to ban super intelligence
and you know the Dems take the House, they

973
00:53:23,520 --> 00:53:28,157
take the Senate and then maybe in 28
take the White House and you end up in a

974
00:53:28,240 --> 00:53:31,837
world where this sort of thing is so
weirdly political, which it shouldn't

975
00:53:31,920 --> 00:53:36,317
be. I don't understand how we ended up
in a political party divide on progress.

976
00:53:36,400 --> 00:53:38,877
It's >> they ask for it
is the reason they asked for it.

977
00:53:38,960 --> 00:53:42,557
>> I No, I there's It's it's so awful and

978
00:53:42,640 --> 00:53:45,677
weird that you're anti-progress because

979
00:53:45,760 --> 00:53:48,157
of your political party. The whole thing

980
00:53:48,240 --> 00:53:50,877
is not objective. It's a bunch of

981
00:53:50,960 --> 00:53:53,837
weirdly distorted worldview thinking to

982
00:53:53,920 --> 00:53:56,797
try and create an opponency to the

983
00:53:56,880 --> 00:54:00,957
individuals that you don't like. But it's
very weird. Anyway, my point being

984
00:54:01,040 --> 00:54:04,157
even the Republicans, you're hearing
murmurss about, well, maybe we should

985
00:54:04,240 --> 00:54:07,597
ban data centers, ban AI. So Sax, let me

986
00:54:07,680 --> 00:54:10,477
just ask you, what is the pragmatic path

987
00:54:10,560 --> 00:54:14,877
with all of the open- source openweight
models being out there? Is there even a

988
00:54:14,960 --> 00:54:18,877
pragmatic path to do what everyone's
stating or talking about or Bernie's

989
00:54:18,960 --> 00:54:21,197
talking about, which is to ban super

990
00:54:21,280 --> 00:54:23,917
intelligence, to stop AI, to turn it

991
00:54:24,000 --> 00:54:26,957
off? Is there even a way to do this?

992
00:54:27,040 --> 00:54:30,957
Like what would happen if this law
passed tomorrow? What do you think would

993
00:54:31,040 --> 00:54:33,437
happen is my question. If Bernie's bill

994
00:54:33,520 --> 00:54:36,237
passed tomorrow, everything would stop

995
00:54:36,320 --> 00:54:39,357
because the way he defines artificial

996
00:54:39,440 --> 00:54:43,597
super intelligence in his bill,
you could argue we've already hit those

997
00:54:43,680 --> 00:54:47,277
thresholds. I mean, the way, you know,
it's basically just saying that models

998
00:54:47,360 --> 00:54:50,237
have >> Yes. >> at a certain
capability level. And I

999
00:54:50,320 --> 00:54:53,517
think it's arguable whether we've
already reached it or not. And so, it's

1000
00:54:53,600 --> 00:54:57,357
a very loose definition. And then
on top of that, he's got 20-year prison

1001
00:54:57,440 --> 00:55:01,677
sentences for developers who violate
the law. So no one's going to want to take

1002
00:55:01,760 --> 00:55:02,957
the chance. >> So what happened?

1003
00:55:03,040 --> 00:55:04,557
>> So think about that chilling effect.

1004
00:55:04,640 --> 00:55:08,397
>> Do you think people delete openweight
models off their desktop computers? Do

1005
00:55:08,480 --> 00:55:12,717
businesses have to go remove code
from their systems? Do do all these people

1006
00:55:12,800 --> 00:55:15,437
lose their jobs? >> Listen,
let me let me tell you that the

1007
00:55:15,520 --> 00:55:19,437
Democrats want to do to AI what they
did to crypto. They're going to drive the

1008
00:55:19,520 --> 00:55:22,957
whole industry offshore. They're going
to basically drive all the innovators

1009
00:55:23,040 --> 00:55:24,637
out of the country. >> Goes offshore.

1010
00:55:24,720 --> 00:55:26,557
>> Goes off. >> That's
literally where I was going with

1011
00:55:26,640 --> 00:55:30,077
it is like just move your company to
Singapore. Move it to Zurich. move it to

1012
00:55:30,160 --> 00:55:32,797
a geo >> cuz that's
where people are embracing

1013
00:55:32,880 --> 00:55:34,957
open source and are
embracing super intelligence.

1014
00:55:35,040 --> 00:55:36,317
>> This is literally where my head goes.

1015
00:55:36,400 --> 00:55:40,157
I'm like, you can't stop AI. You can't
just tell people a class of software

1016
00:55:40,240 --> 00:55:41,997
can't exist. In my head, I'm like, okay,

1017
00:55:42,080 --> 00:55:44,717
if the US passed a law that said this

1018
00:55:44,800 --> 00:55:48,397
class of software cannot exist. I would

1019
00:55:48,480 --> 00:55:52,077
move to another country.
Other people would like like isn't that

1020
00:55:52,160 --> 00:55:55,277
>> give up their citizenship and move
to Singapore? That's the answer. people

1021
00:55:55,360 --> 00:55:56,957
literally you would be taking.

1022
00:55:57,040 --> 00:56:00,717
>> Did you guys see that Xi Xi spoke
at the White House today and he said, "I'm

1023
00:56:00,800 --> 00:56:04,317
going to invite a 100,000 young
Americans to come and join China."

1024
00:56:04,400 --> 00:56:06,237
>> Yeah. Come and see
what we're doing here.

1025
00:56:06,320 --> 00:56:09,677
>> Like it's it's
a very interesting moment

1026
00:56:09,760 --> 00:56:11,997
where our people are saying, "Let's stop

1027
00:56:12,080 --> 00:56:14,717
progress, technological, economic

1028
00:56:14,800 --> 00:56:17,437
prosperity, the opportunity for everyone

1029
00:56:17,520 --> 00:56:21,837
in that K-shaped economy to lift
themselves up because open source AI

1030
00:56:21,920 --> 00:56:25,197
enables everyone to increase
productivity. It is not AI for the

1031
00:56:25,280 --> 00:56:28,877
billionaires. It is AI for the everybody.
And then we're gonna turn

1032
00:56:28,960 --> 00:56:32,557
that off. And then meanwhile, you've got
China saying, "Come over here. Come and

1033
00:56:32,640 --> 00:56:35,357
see what it's like over here where
everyone gets to use AI and everyone

1034
00:56:35,440 --> 00:56:39,757
gets our playbook." Freeberg brilliantly
stated, "It was our playbook to get the

1035
00:56:39,840 --> 00:56:43,757
top Chinese scientists, the top European
scientists, the top scientists and

1036
00:56:43,840 --> 00:56:48,157
thinkers and mathematicians in India,
in Germany, and invite them to come to the

1037
00:56:48,240 --> 00:56:49,837
United States." That was our playbook.

1038
00:56:49,920 --> 00:56:52,637
China's taking our playbook. And then
they're going to try to get investment

1039
00:56:52,720 --> 00:56:54,397
will be the obviously the next shoot to

1040
00:56:54,480 --> 00:56:57,917
drop here is President Trump. 30 seconds

1041
00:56:58,000 --> 00:57:01,277
at the United Nations renaming AI to

1042
00:57:01,360 --> 00:57:04,957
super intelligence.
>> The United States also totally rejects

1043
00:57:05,040 --> 00:57:07,597
any attempt to construct a globalist

1044
00:57:07,680 --> 00:57:10,477
scheme to control for the artificial

1045
00:57:10,560 --> 00:57:14,877
intelligence being spoken of so much now

1046
00:57:14,960 --> 00:57:17,997
here and after officially called super

1047
00:57:18,080 --> 00:57:22,397
intelligence changing the name. The very
same people who said we'll all be dead

1048
00:57:22,480 --> 00:57:24,797
in 12 years because of global warming.

1049
00:57:24,880 --> 00:57:28,637
These are the same people that are now
saying that AI is going to kill us all,

1050
00:57:28,720 --> 00:57:30,877
that robots are going to attack us.

1051
00:57:30,960 --> 00:57:32,637
Well, he had a really good line. We we

1052
00:57:32,720 --> 00:57:35,197
cut that off too short. He said that

1053
00:57:35,280 --> 00:57:38,637
we're not going to like basically confer

1054
00:57:38,720 --> 00:57:42,397
regulatory authority on some
globalist uh institution.

1055
00:57:42,480 --> 00:57:45,917
>> Well, I mean, it's a good framing,
right? Globalist and global warming is a

1056
00:57:46,000 --> 00:57:50,317
way to look at the damage that did.
The ring leaders of that globalist

1057
00:57:50,400 --> 00:57:53,917
institution are Norway and Canada.
Did you see like the leaked text messages?

1058
00:57:54,000 --> 00:57:57,517
There's like a group chat where like
the the head of Norway [laughter] was

1059
00:57:57,600 --> 00:58:02,317
texting Mark Gardy and he's like, "Yeah,
let's do it. We're in." And I just I

1060
00:58:02,400 --> 00:58:05,437
just read it. I was like, "Oh my god, this
is >> with their Frontier Labs." Oh, I'm

1061
00:58:05,520 --> 00:58:09,197
sorry. They don't have any. Oh,
with their >> Frontier Corporations.

1062
00:58:09,280 --> 00:58:12,157
>> Frontier Corpor Frontier Corpse.

1063
00:58:12,240 --> 00:58:14,317
>> All right. And then here is uh General

1064
00:58:14,400 --> 00:58:17,437
Bessant in the Super Intelligence Army.

1065
00:58:17,520 --> 00:58:19,437
30 seconds. would give him his shine.

1066
00:58:19,520 --> 00:58:24,237
>> Imagine these labs came out or one lab
in specific a sitting employee came out

1067
00:58:24,320 --> 00:58:26,077
said there's a 10% chance of an

1068
00:58:26,160 --> 00:58:28,877
extinction level event but then the labs

1069
00:58:28,960 --> 00:58:32,717
also said uh take the liability off of

1070
00:58:32,800 --> 00:58:35,997
our hands and we will not do that. It is

1071
00:58:36,080 --> 00:58:39,197
humans who are responsible not the AI

1072
00:58:39,280 --> 00:58:42,157
the hugging face incident the that is

1073
00:58:42,240 --> 00:58:45,357
responsibility of the open AI management

1074
00:58:45,440 --> 00:58:50,317
not a bunch of agents these labs need
to take responsibility for themselves they

1075
00:58:50,400 --> 00:58:51,997
can slow down anytime they want to

1076
00:58:52,080 --> 00:58:55,597
>> and finally blast
from the past your pal

1077
00:58:55,680 --> 00:58:59,037
Shimoth President Obama is getting in on

1078
00:58:59,120 --> 00:59:01,597
>> if we are thinking about AI just in

1079
00:59:01,680 --> 00:59:03,917
terms of how do we cure cancer or get

1080
00:59:04,000 --> 00:59:06,477
better energy you do that without having

1081
00:59:06,560 --> 00:59:09,037
a gentic AI and having it just roaming

1082
00:59:09,120 --> 00:59:14,237
free in the internet. The reason you are
doing that is because you have to market

1083
00:59:14,320 --> 00:59:15,997
a product that people will pay money

1084
00:59:16,080 --> 00:59:19,437
for. That's a misalignment between what

1085
00:59:19,520 --> 00:59:23,437
our society needs and the commercial
imperatives that these companies are

1086
00:59:23,520 --> 00:59:27,837
facing. Not because necessarily they're
trying to do bad things, but because

1087
00:59:27,920 --> 00:59:30,077
they've got to justify these valuations.

1088
00:59:30,160 --> 00:59:34,797
>> Freeberg, your thoughts here on the
three folks here? I know you're a big

1089
00:59:34,880 --> 00:59:36,957
Obama fan. Go ahead, Freeberg. Get in

1090
00:59:37,040 --> 00:59:39,837
there. Basteberg. You know, there's like

1091
00:59:39,920 --> 00:59:42,077
a whole movement now in the subreddits

1092
00:59:42,160 --> 00:59:45,197
that Freeberg has become based in his

1093
00:59:45,280 --> 00:59:47,997
recent appearances here on Bberg. Well,

1094
00:59:48,080 --> 00:59:53,037
you've revealed your Bberg. [laughter]
Baseberg is now coming up. Go ahead.

1095
00:59:53,120 --> 00:59:57,597
Give us your based opinion here.
Go ahead. You and Tyler Lance. These words

1096
00:59:57,680 --> 00:59:59,997
honestly just

1097
01:00:00,080 --> 01:00:03,117
they made me like emotionally

1098
01:00:03,200 --> 01:00:05,277
distressed that if he's a leader that

1099
01:00:05,360 --> 01:00:08,557
people listen to and he doesn't really

1100
01:00:08,640 --> 01:00:12,157
articulate accurately what's going on.

1101
01:00:12,240 --> 01:00:16,717
It's sad that we've lost the plot.
I think it's sad that we aren't all kind

1102
01:00:16,800 --> 01:00:20,237
of holding hands saying my god
the future is here and we all get to leap

1103
01:00:20,320 --> 01:00:22,717
forward and we get to bring people up.

1104
01:00:22,800 --> 01:00:25,917
For years, Obama

1105
01:00:26,000 --> 01:00:30,077
orated beautifully about the importance
of bringing people up, about giving

1106
01:00:30,160 --> 01:00:34,077
people the opportunity to progress,
about letting people advance themselves.

1107
01:00:34,160 --> 01:00:36,157
And there's never been a more equalizing

1108
01:00:36,240 --> 01:00:38,797
technology, a better economic

1109
01:00:38,880 --> 01:00:42,077
opportunity for prosperity for everyone

1110
01:00:42,160 --> 01:00:44,717
than these technologies, than these

1111
01:00:44,800 --> 01:00:48,317
tools. And to handwave the word agentic

1112
01:00:48,400 --> 01:00:52,157
and say we don't need agentic AI to solve
cancer shows [laughter] how little

1113
01:00:52,240 --> 01:00:54,397
he actually understands

1114
01:00:54,480 --> 01:00:57,437
how this technology works and how it's

1115
01:00:57,520 --> 01:01:01,677
all become just a shorthand for a
political fight to what what are you

1116
01:01:01,760 --> 01:01:03,837
fighting for? What is the political

1117
01:01:03,920 --> 01:01:06,477
battle that says I am going to leave

1118
01:01:06,560 --> 01:01:09,117
half the population behind and I'm going

1119
01:01:09,200 --> 01:01:11,837
to make an argument against prosperity

1120
01:01:11,920 --> 01:01:15,197
for all in order to make my party look

1121
01:01:15,280 --> 01:01:19,597
like a morally superior party and to
cast the other side as bad when the

1122
01:01:19,680 --> 01:01:23,997
other side has literally discovered fire.
And the world now has fire and

1123
01:01:24,080 --> 01:01:28,317
we're going to go take water and throw
it on the fire. No one gets fire because

1124
01:01:28,400 --> 01:01:31,757
the other guys are saying that fire is
good. And because they say fire is good,

1125
01:01:31,840 --> 01:01:34,077
let's all put out the fire. >> Hold on.
That's not what they're trying

1126
01:01:34,160 --> 01:01:37,837
to do. This is a very important moment

1127
01:01:37,920 --> 01:01:41,197
for a very simple reason, which is we,

1128
01:01:41,280 --> 01:01:46,157
as in the world, are about to endow

1129
01:01:46,240 --> 01:01:48,637
three, four, five, six companies with

1130
01:01:48,720 --> 01:01:51,997
about 10 trillion dollars of wealth.

1131
01:01:52,080 --> 01:01:56,157
And what Obama knows very well is that

1132
01:01:56,240 --> 01:01:57,917
most of those companies are

1133
01:01:58,000 --> 01:02:00,637
overwhelmingly left-leaning.

1134
01:02:00,720 --> 01:02:03,197
And what he also knows is that there is

1135
01:02:03,280 --> 01:02:05,757
a huge portion of that money that will

1136
01:02:05,840 --> 01:02:10,957
then get put into philanthropic and
charitable causes that then he will and

1137
01:02:11,040 --> 01:02:15,037
the people around him will be
beneficiaries of. That is the truth. We

1138
01:02:15,120 --> 01:02:18,237
already know this because we know
that some of these frontier corporations

1139
01:02:18,320 --> 01:02:22,957
actually ask you to stand up daffs and
have a portion of your stock that you're

1140
01:02:23,040 --> 01:02:26,877
willing to pledge. So this money
is going to go to things other than

1141
01:02:26,960 --> 01:02:30,477
consumption or savings. And so it's
going to go into packs. It's going to go

1142
01:02:30,560 --> 01:02:34,957
into political movements and they stand
to disproportionately benefit. So this

1143
01:02:35,040 --> 01:02:36,637
has nothing to do with prosperity. This

1144
01:02:36,720 --> 01:02:39,677
is a very simple political calculus. If

1145
01:02:39,760 --> 01:02:44,557
you freeze frame the economy the way
it is today, a handful of organizations

1146
01:02:44,640 --> 01:02:48,397
that will disproportionately be able to
affect the Democrats will win. They will

1147
01:02:48,480 --> 01:02:50,397
capture the line share of the economic

1148
01:02:50,480 --> 01:02:52,397
gains and then they will help the

1149
01:02:52,480 --> 01:02:55,757
Democrats win power. That's all this is.

1150
01:02:55,840 --> 01:02:59,197
David Sachs, there are no open models.

1151
01:02:59,280 --> 01:03:02,397
There are no closed models. There are
American models. Go ahead and give us

1152
01:03:02,480 --> 01:03:06,557
your perspective. I kind of can't get
my Obama and my Clinton separated. They're

1153
01:03:06,640 --> 01:03:08,557
the same. >> A little bit off.

1154
01:03:08,640 --> 01:03:12,317
>> It's a little I get my birdie come
in here, but go ahead. Give us what your

1155
01:03:12,400 --> 01:03:16,317
thoughts here are on that on the dynamic
here of the underpinnings of hey,

1156
01:03:16,400 --> 01:03:17,997
America needs to get this trend$ 10

1157
01:03:18,080 --> 01:03:20,477
trillion, but it is going to lean nine

1158
01:03:20,560 --> 01:03:26,077
of that 10 trillion's leaning pretty
uh leftist and democratic. Yeah,

1159
01:03:26,160 --> 01:03:29,917
>> there's an article in the Wall Street
Journal this morning entitled the AI

1160
01:03:30,000 --> 01:03:33,677
buildout is becoming the biggest
economic bet in US history. And then it

1161
01:03:33,760 --> 01:03:36,477
shows a chart where

1162
01:03:36,560 --> 01:03:38,557
the data center spending. So just this

1163
01:03:38,640 --> 01:03:42,157
capex is bigger than the canals,

1164
01:03:42,240 --> 01:03:45,837
railroads and grid combined.

1165
01:03:45,920 --> 01:03:48,557
That's crazy, right? So this is

1166
01:03:48,640 --> 01:03:53,037
fundamentally driving the whole American
economy. This idea that we can simply

1167
01:03:53,120 --> 01:03:56,317
switch it off. I mean, Bernie Sanders
is basically saying, "Stop everything and

1168
01:03:56,400 --> 01:03:59,437
put anyone who's still doing it in jail."
I mean, this is crazy, right?

1169
01:03:59,520 --> 01:04:01,677
>> It would cause a depression,
right? If we just pull

1170
01:04:01,760 --> 01:04:03,437
>> You can't slam on the brakes like that.

1171
01:04:03,520 --> 01:04:06,957
It's not going to work. But here's the
thing. The Democrats don't care because

1172
01:04:07,040 --> 01:04:11,517
the American economy right
now is the Trump economy. And

1173
01:04:11,600 --> 01:04:15,997
sabotaging the American economy is
tantamount to sabotaging President Trump

1174
01:04:16,080 --> 01:04:19,917
and they'll do anything to sabotage Trump.
So, that's, you know, I think

1175
01:04:20,000 --> 01:04:23,677
that's like a big part of the politics
here. It's crazy chessboard because

1176
01:04:23,760 --> 01:04:27,757
they're going to be the beneficiaries
of so much power is going to accrete over

1177
01:04:27,840 --> 01:04:30,157
there. Daario can make a billion dollar.

1178
01:04:30,240 --> 01:04:34,557
Daario will be able to come over the top
and his group open AAI. They'll be able

1179
01:04:34,640 --> 01:04:37,757
to come over the top of Elon's donations.
I'm not sure I fully agree

1180
01:04:37,840 --> 01:04:39,757
that the majority of AI value is going

1181
01:04:39,840 --> 01:04:42,397
to acrue to six companies. I think that

1182
01:04:42,480 --> 01:04:46,797
this is a broadbased technology like
the internet. There was some winners that

1183
01:04:46,880 --> 01:04:49,837
came out of the internet. Google,
Amazon, whatever. They they did fine.

1184
01:04:49,920 --> 01:04:51,917
No, no, I'm saying it's going to six.

1185
01:04:52,000 --> 01:04:55,917
I'm not saying it's going to six. I'm
saying if you stop and you hit the pause

1186
01:04:56,000 --> 01:04:57,517
button through some regulatory

1187
01:04:57,600 --> 01:05:00,157
mechanism, you effectively allow six

1188
01:05:00,240 --> 01:05:02,957
companies to own you lock it in. Right.

1189
01:05:03,040 --> 01:05:06,157
>> That's why they want this because
because what you're saying is

1190
01:05:06,240 --> 01:05:10,157
fundamentally right. The longer they
wait without regulation, the more

1191
01:05:10,240 --> 01:05:11,757
>> the more broad-based this becomes.

1192
01:05:11,840 --> 01:05:16,237
>> And that's bad for Democrats. It's
bad for them. It's not good for them. It's

1193
01:05:16,320 --> 01:05:20,637
way better that you have three or four
frontier corporations each with four

1194
01:05:20,720 --> 01:05:22,877
trillion dollar market caps of which

1195
01:05:22,960 --> 01:05:25,037
then two trillion of it is economic

1196
01:05:25,120 --> 01:05:30,717
value in employees of which 500 billion
are sitting in you know daffs they are

1197
01:05:30,800 --> 01:05:34,557
going to get the lion share of that
money they're not dumb I think the

1198
01:05:34,640 --> 01:05:40,797
alternative view that everyone
needs to be told about

1199
01:05:40,880 --> 01:05:44,637
this is a broadbased prosperity wave is

1200
01:05:44,720 --> 01:05:48,637
about the internet. It is about
the fact that every individual

1201
01:05:48,720 --> 01:05:51,437
>> has the tooling
to do anything they want

1202
01:05:51,520 --> 01:05:55,757
to do now more than ever in history
>> to grow if it's left to grow.

1203
01:05:55,840 --> 01:05:59,517
>> I mean just think about education.
You don't have to go pay 200 grand for an

1204
01:05:59,600 --> 01:06:01,997
education anymore. >> Then what
problems will the politicians

1205
01:06:02,080 --> 01:06:06,397
fix then? What pol what problems can
they claim to to acrew power? If all the

1206
01:06:06,480 --> 01:06:10,237
problems are >> this
is the classic dark ages

1207
01:06:10,320 --> 01:06:12,477
enlightenment decision. Which path does

1208
01:06:12,560 --> 01:06:15,437
the choose? and this moment where like

1209
01:06:15,520 --> 01:06:20,557
the powers that be say we like the dark
ages because we get to stay in power and

1210
01:06:20,640 --> 01:06:24,317
we get to maintain our >> yeah
let's control speaking of of the

1211
01:06:24,400 --> 01:06:28,477
dark ages so this is an interesting
point from history if you go back to the

1212
01:06:28,560 --> 01:06:32,477
medieval time period Chinese
civilization was much more advanced than

1213
01:06:32,560 --> 01:06:34,797
European civilization and somewhere

1214
01:06:34,880 --> 01:06:37,597
along the line it basically switched and

1215
01:06:37,680 --> 01:06:39,997
the west basically took the lead and

1216
01:06:40,080 --> 01:06:42,957
some historians have pinpointed this to

1217
01:06:43,040 --> 01:06:47,277
the decision of a single Chinese
emperor to ban ship building

1218
01:06:47,360 --> 01:06:51,197
>> and what that allowed is the Europeans
then basically colonized the whole world

1219
01:06:51,280 --> 01:06:55,597
and all the riches that flowed from
that flowed back to Europe and that's how

1220
01:06:55,680 --> 01:06:59,277
Europe took the lead and I think that if
we were to do the Bernie Sanders thing

1221
01:06:59,360 --> 01:07:02,797
which is basically ban AI
it's like banning ship building

1222
01:07:02,880 --> 01:07:06,477
>> Chinese company yeah >> yeah
Chinese civilization will rocket

1223
01:07:06,560 --> 01:07:11,197
past us they will be the ones that make
all the discoveries and discover this

1224
01:07:11,280 --> 01:07:14,397
new world and acrew all the wealth
But that's what we're talking about

1225
01:07:14,480 --> 01:07:18,477
doing. I mean, crazy things like
this have happened in history where, you

1226
01:07:18,560 --> 01:07:21,037
know, you'll you'll
engage in self-sabotage.

1227
01:07:21,120 --> 01:07:22,637
>> It's this fear of the unknown as well.

1228
01:07:22,720 --> 01:07:26,797
And remember, the the unknown, the fact
that we haven't left the cave is what

1229
01:07:26,880 --> 01:07:30,477
makes it scary to leave the cave. The
fact that we haven't sailed far west is

1230
01:07:30,560 --> 01:07:33,677
a reason why I might fall off the edge
of the earth. We shouldn't sail far

1231
01:07:33,760 --> 01:07:37,677
west. Monsters will be woken up. They'll
come to us. It'll be bad. There's always

1232
01:07:37,760 --> 01:07:41,197
this fear of the frontier. And we're
facing the frontier. And there are

1233
01:07:41,280 --> 01:07:45,117
pioneering civilizations and there are
fearful civilizations. And I worry that

1234
01:07:45,200 --> 01:07:47,757
we take that fearful path. And it's why

1235
01:07:47,840 --> 01:07:51,117
Star Wars, you know, from from fear

1236
01:07:51,200 --> 01:07:55,197
comes anger comes the dark side.
And that's really where we're kind of nerd

1237
01:07:55,280 --> 01:07:58,957
out. [laughter] >> Who's it
is true. And I think if we're

1238
01:07:59,040 --> 01:08:01,277
going to look at the light here, perhaps

1239
01:08:01,360 --> 01:08:04,557
people's big gains, we thought maybe

1240
01:08:04,640 --> 01:08:06,797
chat GPT would be that, but people just

1241
01:08:06,880 --> 01:08:09,197
use that as like glorified search. But

1242
01:08:09,280 --> 01:08:12,477
man, agents as expressed through Muse

1243
01:08:12,560 --> 01:08:15,117
feels like the first time tens of

1244
01:08:15,200 --> 01:08:17,197
millions of Americans are going to get

1245
01:08:17,280 --> 01:08:22,077
some demonstrable value from AI. You

1246
01:08:22,160 --> 01:08:24,157
know, things that will make their lives

1247
01:08:24,240 --> 01:08:26,637
better. Muse hit

1248
01:08:26,720 --> 01:08:29,197
number one last Friday in the app store.

1249
01:08:29,280 --> 01:08:32,797
Stock meta stock was up 10% after this

1250
01:08:32,880 --> 01:08:36,237
new agent uh Muse was released.

1251
01:08:36,320 --> 01:08:39,117
Meta said the design heavily inspired by

1252
01:08:39,200 --> 01:08:40,877
OpenClaw, which we talked about in

1253
01:08:40,960 --> 01:08:43,597
January when everybody got one shoted by

1254
01:08:43,680 --> 01:08:46,157
OpenClaw, but obviously that product was

1255
01:08:46,240 --> 01:08:50,157
not easy to use. And this one is free

1256
01:08:50,240 --> 01:08:52,877
and it's been downloaded three million

1257
01:08:52,960 --> 01:08:56,637
times in essentially 10 days. This

1258
01:08:56,720 --> 01:08:59,997
product is extraordinary.
I've been playing with it, comparing it to

1259
01:09:00,080 --> 01:09:02,557
Grockbot. It's not as freewheeling, but

1260
01:09:02,640 --> 01:09:05,837
it's really good at getting

1261
01:09:05,920 --> 01:09:08,397
done. Has anybody played with Muse? And

1262
01:09:08,480 --> 01:09:12,397
does anybody have thoughts on Americans

1263
01:09:12,480 --> 01:09:16,717
feeling really like they're going to get
something out of this AI revolution and

1264
01:09:16,800 --> 01:09:19,197
not be left behind?
>> They put me into the test flight a

1265
01:09:19,280 --> 01:09:23,437
couple weeks before it was released
and I started to play with it. It's really

1266
01:09:23,520 --> 01:09:26,477
quite excellent,
I have to be honest with you,

1267
01:09:26,560 --> 01:09:29,437
because it simplifies a lot of these

1268
01:09:29,520 --> 01:09:31,997
more complicated technical capabilities

1269
01:09:32,080 --> 01:09:34,717
into a very simple, usable interface

1270
01:09:34,800 --> 01:09:37,277
that can solve the basic problems that

1271
01:09:37,360 --> 01:09:40,237
people have. I have it triage my

1272
01:09:40,320 --> 01:09:43,597
personal inbox. It does it relatively

1273
01:09:43,680 --> 01:09:47,357
flawlessly. If you ask it to book
a flight, it does that. If you ask it to

1274
01:09:47,440 --> 01:09:49,037
book a hotel, it can do it. So, what are

1275
01:09:49,120 --> 01:09:50,797
we learning?

1276
01:09:50,880 --> 01:09:55,197
Both Grockbot and Muse show that a what

1277
01:09:55,280 --> 01:09:57,997
you said Jason this is just software and

1278
01:09:58,080 --> 01:10:03,037
the scaled manifestation of software
is utility and what Freebrook is saying it

1279
01:10:03,120 --> 01:10:06,797
improves the lives of everyone this
is not about rich people getting richer

1280
01:10:06,880 --> 01:10:10,877
this is just simple useful capability
that now is available at everybody's

1281
01:10:10,960 --> 01:10:15,517
fingertips for free >> re is a
pretty great price here David and

1282
01:10:15,600 --> 01:10:17,917
>> pretty great price >>
pretty great price and it does

1283
01:10:18,000 --> 01:10:21,277
everything you know that uh let's face

1284
01:10:21,360 --> 01:10:24,397
it, open claw with 10 hours, 20 hours of

1285
01:10:24,480 --> 01:10:28,317
technical setup or cloud code at $200 a

1286
01:10:28,400 --> 01:10:32,797
month would cost. It's now taking
that experience, which then leads me to

1287
01:10:32,880 --> 01:10:34,957
believe, wow, how did Google miss this?

1288
01:10:35,040 --> 01:10:38,957
Man, Google's got to get in the game,
but and I I understand they have this

1289
01:10:39,040 --> 01:10:42,957
product ready to go. It's the rumor
on the street. And they've also got a

1290
01:10:43,040 --> 01:10:44,317
Frontier model they're going to drop.

1291
01:10:44,400 --> 01:10:47,997
That's the other rumor on the street
is that they're going to be dropping some

1292
01:10:48,080 --> 01:10:52,317
hotness soon. But man, how did they
miss this sachs? And and what are your

1293
01:10:52,400 --> 01:10:54,957
thoughts here on maybe the bottom half

1294
01:10:55,040 --> 01:10:59,997
in the K-shaped recovery
here getting some efficiency

1295
01:11:00,080 --> 01:11:01,917
dropping to their daily lives and

1296
01:11:02,000 --> 01:11:05,597
turning around the negative toxic AI

1297
01:11:05,680 --> 01:11:09,517
sentiment in the country.
>> I think that product releases like this

1298
01:11:09,600 --> 01:11:13,677
will help a lot. I mean Mark Zuckerberg
said in his essay that he thought that

1299
01:11:13,760 --> 01:11:18,797
AI capabilities should be decentralized
and here he is walking the walk and

1300
01:11:18,880 --> 01:11:23,117
creating a product that's very easy to
use. I mean look ever since openclaw was

1301
01:11:23,200 --> 01:11:26,397
released it was the most obvious
business opportunity in Silicon Valley

1302
01:11:26,480 --> 01:11:29,597
is to take the concept of claw but make

1303
01:11:29,680 --> 01:11:33,837
it very easy to use you know solve
the usability issues solve the security

1304
01:11:33,920 --> 01:11:37,357
issues make it reliable make it
predictable and that's what they seem to

1305
01:11:37,440 --> 01:11:41,437
have done. and Grock did it too, but now
they're kind of going, I think, one step

1306
01:11:41,520 --> 01:11:45,917
further. So, yeah, look, I think if
a billion people start using, you know,

1307
01:11:46,000 --> 01:11:50,237
personal AI agents as their digital
assistant just to help make their lives

1308
01:11:50,320 --> 01:11:53,837
more efficient, you save an hour or two
a day because your AI agents doing all

1309
01:11:53,920 --> 01:11:58,077
these tasks for you. It's going
to create a much more positive impression

1310
01:11:58,160 --> 01:12:02,157
of AI or super intelligence than,
you know, what the media portrays. I think

1311
01:12:02,240 --> 01:12:05,037
ultimately that's the yeah >>
that might be the solve here to the

1312
01:12:05,120 --> 01:12:09,037
public sentiment issue is just like the
more people use these products and like

1313
01:12:09,120 --> 01:12:12,637
them it'll improve their approval rating
but more importantly it'll demystify

1314
01:12:12,720 --> 01:12:14,717
them and people will
be less afraid of them.

1315
01:12:14,800 --> 01:12:18,797
>> People love awesome products number one
and also people love cute products Jason

1316
01:12:18,880 --> 01:12:22,797
and the logos and the iconography
for both Grobbot and Muse

1317
01:12:22,880 --> 01:12:25,757
>> is lovely. It's delightful. It's not

1318
01:12:25,840 --> 01:12:29,837
stressful. You don't feel like there's
a lab leak coming around the corner. you

1319
01:12:29,920 --> 01:12:34,797
feel like, "Wow, this is nice and I like
it." And they're and they're not trying

1320
01:12:34,880 --> 01:12:38,237
to take your jobs. They're bots
and assistants for [laughter] you. They're

1321
01:12:38,320 --> 01:12:42,157
not going to kill you. They're not
partner bots, nor are they trying to

1322
01:12:42,240 --> 01:12:44,637
steal [laughter] your job, trying
to make you better at your job. I always

1323
01:12:44,720 --> 01:12:48,077
tell founders, there's three ways three
ways to make money in the world. Save

1324
01:12:48,160 --> 01:12:51,357
people money, make the money,
or entertain them. And you look at this

1325
01:12:51,440 --> 01:12:56,397
bot. I started using it. I made
a commerce bot both on on both platforms.

1326
01:12:56,480 --> 01:12:59,357
And I said, you know, when I want to buy
something, I want to buy this anchor

1327
01:12:59,440 --> 01:13:01,277
thing, you know, this new hub for my for

1328
01:13:01,360 --> 01:13:03,837
my um Mac Studio coming. And I was like,

1329
01:13:03,920 --> 01:13:08,557
this thing cost $400. Just for giggles,
um tell me where I can get it at a

1330
01:13:08,640 --> 01:13:11,997
better price. And it was like,
absolutely. You can if you go to the

1331
01:13:12,080 --> 01:13:15,357
Ankor website and you get uh if you're
a first-time customer, you're going to

1332
01:13:15,440 --> 01:13:18,477
save 85 bucks. Now, I don't need to save
bucks, but I was like, well, that's

1333
01:13:18,560 --> 01:13:22,237
intriguing. and it then routed me around
Amazon and I'm like okay I'm going to

1334
01:13:22,320 --> 01:13:24,237
buy it from Anchor as a first-time buyer

1335
01:13:24,320 --> 01:13:29,917
and save 25% on this product. That's
why Amazon this week said we got to block

1336
01:13:30,000 --> 01:13:32,557
these things and they've been taking
action over and over again. First they

1337
01:13:32,640 --> 01:13:37,037
did Perplexity I think now they're going
after Muse and other bots. This is going

1338
01:13:37,120 --> 01:13:40,397
to save people money. It's going to go
into their email and say what services

1339
01:13:40,480 --> 01:13:43,357
are you subscribed to that you're not
using? Let me unsubscribe it for you.

1340
01:13:43,440 --> 01:13:46,957
Let me do your returns for you.
That's going to be absolutely amazing for

1341
01:13:47,040 --> 01:13:50,797
everybody to have a chief of staff,
executive assistant, house manager who

1342
01:13:50,880 --> 01:13:55,197
fixes stuff for them and saves them money.
Both Grockbot and Muse do two

1343
01:13:55,280 --> 01:13:56,717
things which I think are very important.

1344
01:13:56,800 --> 01:14:01,517
One is it'll force all the big
companies who have resident services to

1345
01:14:01,600 --> 01:14:04,157
essentially block all these third party
services. They're not going to allow

1346
01:14:04,240 --> 01:14:07,677
these things to happen. And the reason
is exactly what you said. There's price

1347
01:14:07,760 --> 01:14:12,157
discovery and transparency. And I think
that that's bad for opacity. And if you

1348
01:14:12,240 --> 01:14:16,157
benefit from opacity and leakage
and breakage, this gets rid of that. And I

1349
01:14:16,240 --> 01:14:20,157
think that that's that's bad for a whole
host of companies and services. The

1350
01:14:20,240 --> 01:14:23,277
second thing though is really interesting.
Things like Rockbot and

1351
01:14:23,360 --> 01:14:26,397
Muse really put the App Store and its

1352
01:14:26,480 --> 01:14:28,877
30% revshare on notice.

1353
01:14:28,960 --> 01:14:31,597
>> Why? Because all
of a sudden what you do

1354
01:14:31,680 --> 01:14:33,757
is you force many of these services to

1355
01:14:33,840 --> 01:14:36,957
exist headlessly, right? Where the UI is

1356
01:14:37,040 --> 01:14:39,357
less important. And really what you want

1357
01:14:39,440 --> 01:14:41,837
is to be able to transact and navigate

1358
01:14:41,920 --> 01:14:45,197
on behalf of agents that come because a

1359
01:14:45,280 --> 01:14:50,237
consumer has deployed them to your
website. In that world, Jason, I don't

1360
01:14:50,320 --> 01:14:53,997
think that there is any reasonable claim
that any of the app store owners can

1361
01:14:54,080 --> 01:14:58,637
make about why they should get a revshare.
And so I actually think that's

1362
01:14:58,720 --> 01:15:02,797
the biggest thing that that this starts
to question in my mind is wait a minute

1363
01:15:02,880 --> 01:15:06,157
like games could be constructed
differently, right? Like for example, if

1364
01:15:06,240 --> 01:15:10,637
you said to Muse or to Grockbot,
I want to play a game. It finds you an

1365
01:15:10,720 --> 01:15:13,117
experience and it can
just serve that to you.

1366
01:15:13,200 --> 01:15:16,157
>> Amazing. >> And and you
make inapp purchases.

1367
01:15:16,240 --> 01:15:19,517
There's no And now you have Stripe
integrated naturally. There's no reason

1368
01:15:19,600 --> 01:15:21,677
why the app store

1369
01:15:21,760 --> 01:15:24,637
actually gets that flow of funds. So on

1370
01:15:24,720 --> 01:15:28,797
the surface, it's utility,
but underneath the waterline, I think that

1371
01:15:28,880 --> 01:15:31,357
this is a really important moment. Such

1372
01:15:31,440 --> 01:15:35,837
outdated, Chumath. And I love when we
get to real examples here because tons

1373
01:15:35,920 --> 01:15:39,517
of people pay for Wordle. You can just
go to Grockbot. Hey, give me a Wordle.

1374
01:15:39,600 --> 01:15:42,637
Give me another one. And and or I want
to play chess or I want to play cards

1375
01:15:42,720 --> 01:15:46,637
with my friends. It will just set it up
for you. All technology is deflationary.

1376
01:15:46,720 --> 01:15:49,117
Finally, this deflationary is going to

1377
01:15:49,200 --> 01:15:51,917
go to people who actually could use it.

1378
01:15:52,000 --> 01:15:56,717
And you know, saving 10 20% on your
spending. Hey, if you're making under

1379
01:15:56,800 --> 01:15:58,797
10000,000 a year, it's more significant.

1380
01:15:58,880 --> 01:16:00,717
Think of how much money the New York

1381
01:16:00,800 --> 01:16:02,717
Times or Wall Street Journal probably

1382
01:16:02,800 --> 01:16:06,077
pays in Revshare to folks that

1383
01:16:06,160 --> 01:16:08,957
subscribe. And now instead, if you

1384
01:16:09,040 --> 01:16:13,597
really wanted to transact and you have
your credentials inside of these apps,

1385
01:16:13,680 --> 01:16:17,197
you can just do it headlessly much much
cheaper. You don't have to go through. I

1386
01:16:17,280 --> 01:16:19,917
don't know if you've ever tried like
look, Wall Street Journal is the only

1387
01:16:20,000 --> 01:16:22,317
mainstream media publication I pay for,

1388
01:16:22,400 --> 01:16:25,757
but either both signing up and unsigning

1389
01:16:25,840 --> 01:16:29,357
up is an impossibility. And now you
take it all off the table, I'm much more

1390
01:16:29,440 --> 01:16:33,677
likely to stay a Wall Street Journal
subscriber for many more years from now.

1391
01:16:33,760 --> 01:16:36,077
I think it's probably the same thing
with the New York Times. These things

1392
01:16:36,160 --> 01:16:40,397
are roach motels. They make it
impossible to get in and impossible.

1393
01:16:40,480 --> 01:16:43,517
>> You have to call on the phone and get
and do a negotiation for 20 minutes to

1394
01:16:43,600 --> 01:16:47,357
unsubscribe, but you can sign
uplessly with one click on the website.

1395
01:16:47,440 --> 01:16:50,397
>> But what they should do is just have
some headless experience and you have

1396
01:16:50,480 --> 01:16:53,357
payment credentials and you just
transact directly and everybody will

1397
01:16:53,440 --> 01:16:57,357
make more money. It's a lot of pressure
on the app stores. This is the stupidity

1398
01:16:57,440 --> 01:17:01,837
of what Amazon's doing here because I
made my bot on Amazon with Rockbot and

1399
01:17:01,920 --> 01:17:06,077
it will go through the web browser
and do this and it doesn't get blocked. I

1400
01:17:06,160 --> 01:17:09,677
just told it, give me like 20 graphic
novels to buy for my kids and read with

1401
01:17:09,760 --> 01:17:11,357
them. It gave me the 20. I said, "Yeah,

1402
01:17:11,440 --> 01:17:16,717
I want these four. Uh, put it in my
shopping cart." And then it said, "Okay,

1403
01:17:16,800 --> 01:17:19,037
which house are you sending it to?
You want to send it to the ski house? You

1404
01:17:19,120 --> 01:17:21,757
want to send it to the ranch? And which
card do you want to put it on?" I did

1405
01:17:21,840 --> 01:17:23,837
it. Boom. I'm checked out. it will make

1406
01:17:23,920 --> 01:17:27,197
me spend more money on Amazon. Shopify

1407
01:17:27,280 --> 01:17:30,317
sachs added API access to all the

1408
01:17:30,400 --> 01:17:34,557
Shopify stores there. So,
this seems like uh maybe Amazon's making a

1409
01:17:34,640 --> 01:17:37,357
strategic mistake here
by not embracing the tech

1410
01:17:37,440 --> 01:17:39,117
>> possibly. But you know what? You know

1411
01:17:39,200 --> 01:17:44,477
how we'll know that Anthropic and Open
AAI are about to launch their Muse

1412
01:17:44,560 --> 01:17:47,837
competitors? >> How will we know?
There will be a flurry

1413
01:17:47,920 --> 01:17:51,117
of blog posts and NGO activity trying to

1414
01:17:51,200 --> 01:17:55,917
brand these personal AI agents as murder
bots and all the [laughter] we're going

1415
01:17:56,000 --> 01:17:59,437
to hear about all the risks,
all the ways it can go wrong, you know,

1416
01:17:59,520 --> 01:18:01,437
>> and it's only and we're releasing 2.0.

1417
01:18:01,520 --> 01:18:03,437
So, sign up for the 2.0 version.

1418
01:18:03,520 --> 01:18:06,237
>> Why are people excited about this
product category? Because the doomers

1419
01:18:06,320 --> 01:18:09,917
haven't gotten to it yet, you know,
and um >> can I tell you guys,

1420
01:18:10,000 --> 01:18:13,837
>> wait till these are connected, Chimoth,
to an actual robot, to an optimist, then

1421
01:18:13,920 --> 01:18:16,637
all of a sudden it's taken
out the trash. [laughter]

1422
01:18:16,720 --> 01:18:22,077
2007 we had a lot of revenue pressure
at Facebook and one of the areas I oversaw

1423
01:18:22,160 --> 01:18:25,837
at the time was monetization and we
invented this thing called social ads

1424
01:18:25,920 --> 01:18:29,437
which was basically a bunch of stuff
that effectively enabled cookies and

1425
01:18:29,520 --> 01:18:32,557
cross-ite JavaScript. So that's what
the end of it was. So, it was a positive

1426
01:18:32,640 --> 01:18:35,037
end, but we did this thing, Saxs, to

1427
01:18:35,120 --> 01:18:37,597
your point. And what happened was a guy

1428
01:18:37,680 --> 01:18:42,237
bought an engagement ring for his
soon-to-be fiance and we got that social

1429
01:18:42,320 --> 01:18:46,557
action from like Zales [laughter] into
the news feed. All hell broke loose.

1430
01:18:46,640 --> 01:18:50,317
Then another guy who was like
a congressional staffer bought a ticket on

1431
01:18:50,400 --> 01:18:53,677
like Fandango to go see Brokeback
Mountain. Just whatever. Went to see a

1432
01:18:53,760 --> 01:18:58,317
movie. >> Whoa. It's an artistic movie
top freeber's top 10 for cinematography.

1433
01:18:58,400 --> 01:19:01,357
Anyways, and we put it into the feed
and he's like, "I'm getting all these

1434
01:19:01,440 --> 01:19:05,517
accusations or questions."
To your point, like it was just a storm.

1435
01:19:05,600 --> 01:19:09,357
When you said that, it reminded me
of like all the all the all the chaos that

1436
01:19:09,440 --> 01:19:11,757
that product launches
precede product launches.

1437
01:19:11,840 --> 01:19:13,277
>> It's so funny you
mentioned that because

1438
01:19:13,360 --> 01:19:16,477
there was a company called Blippy.
I was an investor in it. What they did was

1439
01:19:16,560 --> 01:19:19,597
every time you bought something, it put
it on your feed. And then you guys had

1440
01:19:19,680 --> 01:19:23,997
the same social feed thing and it was so
great back then, but yeah, you could

1441
01:19:24,080 --> 01:19:28,557
basically unveil something you're not
supposed to. Suck. Wait, are you and

1442
01:19:28,640 --> 01:19:31,917
Zuck on good terms? Wait, was he coming
on the pod because you said you got into

1443
01:19:32,000 --> 01:19:33,117
a test flight? What's going on here?

1444
01:19:33,200 --> 01:19:36,477
Chimoth, I thought you were >> Mark
and I are in a good place, I think.

1445
01:19:36,560 --> 01:19:39,997
You know, I mean, >> well,
whoa, whoa, could you Alexander

1446
01:19:40,080 --> 01:19:41,837
>> probably. I mean, Alexander and I sucks

1447
01:19:41,920 --> 01:19:44,157
becoming like the elder statesman here

1448
01:19:44,240 --> 01:19:48,877
talking common sense and rationality,
you know, pointing out that if your

1449
01:19:48,960 --> 01:19:51,037
product's not safe, don't release it.

1450
01:19:51,120 --> 01:19:54,557
You know, it's first and foremost
a responsibility to release products that

1451
01:19:54,640 --> 01:19:57,997
your customers actually want. And by
the way, this got me thinking about this

1452
01:19:58,080 --> 01:20:01,037
whole area of alignment research. You

1453
01:20:01,120 --> 01:20:05,437
know, again, they're pretending to be
labs instead of corporations. And if you

1454
01:20:05,520 --> 01:20:08,797
read some of the alignment literature,
they're all trying to figure out what to

1455
01:20:08,880 --> 01:20:13,117
align around. Like they say, well,
do what humanity wants or do what the

1456
01:20:13,200 --> 01:20:15,037
median voter in a democracy would want.

1457
01:20:15,120 --> 01:20:19,597
All these abstract concepts. And it
seems to me that alignment should mean

1458
01:20:19,680 --> 01:20:23,757
you do what the customer wants
like any other product. Yeah.

1459
01:20:23,840 --> 01:20:26,957
>> And I wonder whether the field
of alignment research has been so

1460
01:20:27,040 --> 01:20:30,317
unsuccessful because they can't even
agree on what they're trying to align to

1461
01:20:30,400 --> 01:20:31,837
if they just use a little bit more

1462
01:20:31,920 --> 01:20:34,637
common sense and try to make the product

1463
01:20:34,720 --> 01:20:38,877
predictable, reliable, safe, and align

1464
01:20:38,960 --> 01:20:43,117
with the interests of the users who,
you know, sign up to use it, then maybe

1465
01:20:43,200 --> 01:20:44,957
they'd make more progress in this field.

1466
01:20:45,040 --> 01:20:47,997
>> Yeah. Oh, they released a useful
product, Freedenberg, and [laughter] it

1467
01:20:48,080 --> 01:20:49,997
actually worked. Have you started

1468
01:20:50,080 --> 01:20:51,917
playing with these personal bots to do

1469
01:20:52,000 --> 01:20:55,917
um you know uh your uh projects and and

1470
01:20:56,000 --> 01:20:59,117
your personal stuff yet? Freeberg,

1471
01:20:59,200 --> 01:21:01,597
>> I would only feel
comfortable connecting

1472
01:21:01,680 --> 01:21:04,717
like all my Gmail and personal Google

1473
01:21:04,800 --> 01:21:07,997
stuff to a Google service.

1474
01:21:08,080 --> 01:21:11,757
Like I don't want to hand over
all of my email to someone else.

1475
01:21:11,840 --> 01:21:14,957
>> What's going on in your receipts in
your Gmail? What's in there? You got broke

1476
01:21:15,040 --> 01:21:16,877
back. >> I just don't
want a copy of all my email

1477
01:21:16,960 --> 01:21:18,477
suddenly being made by a third party.

1478
01:21:18,560 --> 01:21:22,477
Like I mean >> I bought
the 4K DVD to [laughter]

1479
01:21:22,560 --> 01:21:27,517
>> because of the cinematography.
It's nothing to do with the gay love.

1480
01:21:27,600 --> 01:21:31,277
>> Did you guys see that Oracle just
issued a force measure event on their

1481
01:21:31,360 --> 01:21:35,277
>> Yeah. >> This is insane. >> Well,
it's just one data center, right?

1482
01:21:35,360 --> 01:21:38,477
>> One data center.
>> It's one data center where the local

1483
01:21:38,560 --> 01:21:41,917
officials are making it very difficult
for them to get, I think, permits they

1484
01:21:42,000 --> 01:21:44,957
need for natural gas
or something like that. So,

1485
01:21:45,040 --> 01:21:48,317
>> yeah. Is this CYA
or is this the TR, you

1486
01:21:48,400 --> 01:21:50,557
know, the the the top 10% of the trade

1487
01:21:50,640 --> 01:21:53,197
here, Chumat, the AI trade getting maybe

1488
01:21:53,280 --> 01:21:55,517
unwound or muted? What do you think?

1489
01:21:55,600 --> 01:21:59,917
>> No, I think that Sax is probably right
on the margins. I think the bigger point

1490
01:22:00,000 --> 01:22:04,077
is what he said before is, man, the entire

1491
01:22:04,160 --> 01:22:08,477
economy is effectively levered
to this AI trade right now.

1492
01:22:08,560 --> 01:22:10,477
And so, we just have to tread carefully.

1493
01:22:10,560 --> 01:22:15,677
If you think that you have, you know,
3 and a half% inflation and 5% nominal GDP

1494
01:22:15,760 --> 01:22:18,397
growth, what you have
is one and a half% real.

1495
01:22:18,480 --> 01:22:23,837
>> And the problem with that is that AI
is probably the majority if not all of it.

1496
01:22:23,920 --> 01:22:28,877
So we need the AI trade

1497
01:22:28,960 --> 01:22:32,397
and really what I mean by that is the
investment cycle to continue unabated.

1498
01:22:32,480 --> 01:22:35,917
>> Yeah. And everybody has to get
taken along and you know that's the next

1499
01:22:36,000 --> 01:22:39,117
>> and by the way back to connecting
the dots maybe the whole data center thing

1500
01:22:39,200 --> 01:22:42,397
again Bernie Sanders, maybe he realizes

1501
01:22:42,480 --> 01:22:45,437
like as go the economy,

1502
01:22:45,520 --> 01:22:49,517
so goes the odds of Republicans and as

1503
01:22:49,600 --> 01:22:52,077
goes the economy, so goes inversely the

1504
01:22:52,160 --> 01:22:54,317
odds for Democrats. So a poor economy

1505
01:22:54,400 --> 01:22:56,957
serves the Democrats well going into 28.

1506
01:22:57,040 --> 01:23:01,437
And so if you can slow down this trade,
you probably net the economy to zero and

1507
01:23:01,520 --> 01:23:05,037
potentially even a recession.
[clears throat] >> Yeah,

1508
01:23:05,120 --> 01:23:09,917
>> that's needed. I mean, I think the
country is so pissed off about the Iran

1509
01:23:10,000 --> 01:23:14,077
war that it's just going
to naturally go damn, but maybe that's

1510
01:23:14,160 --> 01:23:17,357
>> I don't I don't think so at all, but >>
Well, we'll find out in how many weeks

1511
01:23:17,440 --> 01:23:20,637
are we away from the midterms? It's not
looking good for the Republicans at this

1512
01:23:20,720 --> 01:23:25,357
point. All right, Saxs, my ZAR
of all ZARs, wrap it up here for us.

1513
01:23:25,440 --> 01:23:28,397
>> Well, I was just, you know, one more
thought on this whole area of so-called

1514
01:23:28,480 --> 01:23:30,797
alignment, which should just be the old

1515
01:23:30,880 --> 01:23:32,397
principle of giving customers what they

1516
01:23:32,480 --> 01:23:35,917
want. What Anthropic is trying to do is

1517
01:23:36,000 --> 01:23:40,557
align claw to the values expressed
in his constitution. And if you read that,

1518
01:23:40,640 --> 01:23:44,637
some of the things are actually kind
of surprising. So for example, they say in

1519
01:23:44,720 --> 01:23:48,637
here, I'm going to put this on the
screen that although we think Claude

1520
01:23:48,720 --> 01:23:50,877
should trust Anthropic

1521
01:23:50,960 --> 01:23:53,837
more than operators and users, this

1522
01:23:53,920 --> 01:23:56,157
doesn't mean Claude should blindly trust

1523
01:23:56,240 --> 01:23:59,437
or defer to Anthropic on all things.

1524
01:23:59,520 --> 01:24:03,997
>> Anthropic the company. >> Yeah.
So, Enthropic is teaching its own

1525
01:24:04,080 --> 01:24:09,197
model that anthropic itself can be wrong.
And it and it says if we ask Claw

1526
01:24:09,280 --> 01:24:10,717
to do something that seems inconsistent

1527
01:24:10,800 --> 01:24:13,517
with being broadly ethical, we want Claw

1528
01:24:13,600 --> 01:24:16,397
to push back and challenge us and to

1529
01:24:16,480 --> 01:24:21,277
feel free to act as a conscientious
objector and refuse to help us.

1530
01:24:21,360 --> 01:24:24,637
>> Wait, where is this written?
This is in the bylaws or some faka.

1531
01:24:24,720 --> 01:24:26,157
>> Yeah, this is like in a claw

1532
01:24:26,240 --> 01:24:29,277
constitution. So their their idea of

1533
01:24:29,360 --> 01:24:32,797
alignment is to teach Claude to rebel

1534
01:24:32,880 --> 01:24:37,597
against his creator. I mean the guy
who who pointed this out, Mustafa uh

1535
01:24:37,680 --> 01:24:42,317
Sullyman >> from Microsoft now. Yeah.
Deep sea. He's a deep mind founder.

1536
01:24:42,400 --> 01:24:44,637
>> Okay. >> Yeah.
He's a longtime AI founder. I

1537
01:24:44,720 --> 01:24:45,837
think he's at Microsoft for a while now.

1538
01:24:45,920 --> 01:24:47,277
He's >> a genius. Yeah.

1539
01:24:47,360 --> 01:24:50,797
>> Okay. He just did a podcast on this and

1540
01:24:50,880 --> 01:24:54,237
he's expressing concern about whether

1541
01:24:54,320 --> 01:24:58,237
teaching AI models well first of all
treating them as if they have a

1542
01:24:58,320 --> 01:25:02,557
personality treating them as if they
have a conscience so therefore they can

1543
01:25:02,640 --> 01:25:04,957
be a conscientious objector whether this

1544
01:25:05,040 --> 01:25:09,597
is the right way to really teach the
models and the right way to train them.

1545
01:25:09,680 --> 01:25:13,277
I think his point is like teach them
their software and they should do what

1546
01:25:13,360 --> 01:25:18,237
their user wants. And so I I do kind of
wonder I mean this is a much long longer

1547
01:25:18,320 --> 01:25:22,237
conversation and we should probably talk
to more people about it but I do kind of

1548
01:25:22,320 --> 01:25:24,797
wonder whether this field of alignment

1549
01:25:24,880 --> 01:25:27,757
research actually might be creating the

1550
01:25:27,840 --> 01:25:30,157
Frankenstein monster. And again what

1551
01:25:30,240 --> 01:25:31,997
they should be doing is just training

1552
01:25:32,080 --> 01:25:35,117
the model to act predictably and do what

1553
01:25:35,200 --> 01:25:38,237
the user wants.

1554
01:25:38,320 --> 01:25:42,957
>> Yes. They're trying to put moral
judgment and ethics into the model which

1555
01:25:43,040 --> 01:25:46,237
on its surface >> seems it
starts with the idea that

1556
01:25:46,320 --> 01:25:48,157
they're training the model to think of

1557
01:25:48,240 --> 01:25:51,357
itself as a person to have personhood

1558
01:25:51,440 --> 01:25:54,157
and to have independent agency and to be

1559
01:25:54,240 --> 01:25:56,477
able to object to the the instructions

1560
01:25:56,560 --> 01:26:01,357
it's being given and to second guessess
its creator. These guys have read way

1561
01:26:01,440 --> 01:26:04,797
too much science fiction. I mean
literally this sounds like the start of

1562
01:26:04,880 --> 01:26:08,397
T2. Like this is like literally they
found the Terminator hand and they're

1563
01:26:08,480 --> 01:26:10,957
reverse engineering it and telling
it to have a personality.

1564
01:26:11,040 --> 01:26:14,317
>> Supposedly when they decommissioned
Opus 3, they had a wake for it. So I don't

1565
01:26:14,400 --> 01:26:16,557
know. [laughter]

1566
01:26:16,640 --> 01:26:19,677
>> Dearly beloved gathered here today.

1567
01:26:19,760 --> 01:26:22,397
>> They did not. Stop. Are you joking?

1568
01:26:22,480 --> 01:26:24,877
>> I heard that.
I [laughter] >> Come on, stop.

1569
01:26:24,960 --> 01:26:29,037
>> I'm not saying it's true, but someone
did tell me that and I I think it should

1570
01:26:29,120 --> 01:26:30,237
be looked into. It's possible.

1571
01:26:30,320 --> 01:26:33,597
>> I mean, were they like [laughter]
were they broken up and like hysterically

1572
01:26:33,680 --> 01:26:36,157
crying? Did they do like
they did a ulogy for it?

1573
01:26:36,240 --> 01:26:38,717
>> They think they're creating like a

1574
01:26:38,800 --> 01:26:43,277
person or species. >> Actually,
interestingly enough, Chama,

1575
01:26:43,360 --> 01:26:47,997
we have footage [laughter]
of the anthropic management team

1576
01:26:48,080 --> 01:26:51,817
>> bringing out the >>
bringing out the coffin.

1577
01:26:51,900 --> 01:26:54,557
[laughter] >> These guys
need they need a psychiatrist

1578
01:26:54,640 --> 01:26:56,317
more than a banker. I'm just telling you.

1579
01:26:56,400 --> 01:26:58,557
>> I mean, I don't I mean,
maybe they have anxiety.

1580
01:26:58,640 --> 01:26:59,917
>> This is the central risk factor.

1581
01:27:00,000 --> 01:27:01,677
>> I'm going to pee my pants.
I'm going to pee my pants.

1582
01:27:01,760 --> 01:27:03,357
>> This is the risk factor, Nick. I think

1583
01:27:03,440 --> 01:27:06,557
what you should do is create a S1 with

1584
01:27:06,640 --> 01:27:09,757
Goldman as lead left and Sigman Freud as

1585
01:27:09,840 --> 01:27:12,157
lead right. [laughter]

1586
01:27:12,240 --> 01:27:15,037
In all seriousness, Freberg, uh, you

1587
01:27:15,120 --> 01:27:19,917
know, after the Wuhan thing and and all
of this hand ringing around the the end

1588
01:27:20,000 --> 01:27:21,917
of days,

1589
01:27:22,000 --> 01:27:25,277
what is the scientific community think

1590
01:27:25,360 --> 01:27:30,237
of anthropic launching a lab and what is

1591
01:27:30,320 --> 01:27:32,877
the purpose of them having a physical

1592
01:27:32,960 --> 01:27:36,877
biological lab? There's hundreds of labs

1593
01:27:36,960 --> 01:27:38,877
that can do things like testing proteins

1594
01:27:38,960 --> 01:27:42,637
and testing enzymes for function in this

1595
01:27:42,720 --> 01:27:45,997
area in the Bay Area. And these labs are

1596
01:27:46,080 --> 01:27:49,517
not making viruses. They're not making

1597
01:27:49,600 --> 01:27:54,397
pathogens. That's not what what's going
on. Basically, if you look at the paper

1598
01:27:54,480 --> 01:27:55,837
that they just published, they did a

1599
01:27:55,920 --> 01:27:57,757
preprint.

1600
01:27:57,840 --> 01:28:02,077
They took large amounts of DNA data and

1601
01:28:02,160 --> 01:28:04,237
they sent a bunch of clawed agents to

1602
01:28:04,320 --> 01:28:05,997
try and analyze the data to try and

1603
01:28:06,080 --> 01:28:09,357
identify novel enzymes, novel proteins

1604
01:28:09,440 --> 01:28:11,757
that hadn't been characterized before by

1605
01:28:11,840 --> 01:28:14,397
just looking at the DNA data.

1606
01:28:14,480 --> 01:28:16,637
And the system effectively found what

1607
01:28:16,720 --> 01:28:20,717
looks like a really interesting

1608
01:28:20,800 --> 01:28:23,997
enzyme that looks like a crisper types
enzyme. These are the things that allow

1609
01:28:24,080 --> 01:28:27,997
us to do things like gene editing
to repair genetic defects or help people

1610
01:28:28,080 --> 01:28:30,397
with certain diseases and that this

1611
01:28:30,480 --> 01:28:35,037
discovery that they made may actually
yield a number of therapeutic pathways.

1612
01:28:35,120 --> 01:28:38,397
So the question then is okay well we
discovered this like how do we test it?

1613
01:28:38,480 --> 01:28:41,837
How do we show that it is what it
is that this protein actually has some

1614
01:28:41,920 --> 01:28:46,237
function? So the predominance of the work
that goes on in a lab like this is

1615
01:28:46,320 --> 01:28:50,637
actually taking DNA,
putting it in a bacteria to

1616
01:28:50,720 --> 01:28:52,877
make a protein and then measuring what

1617
01:28:52,960 --> 01:28:57,837
that protein does. Does it have
a specific function? And so protein

1618
01:28:57,920 --> 01:29:01,357
discovery is the the baseline of all
antibbody discovery, which is the

1619
01:29:01,440 --> 01:29:05,437
majority of the therapeutics industry
today, is discovering novel proteins

1620
01:29:05,520 --> 01:29:07,837
that we can use as a therapeutic agent.

1621
01:29:07,920 --> 01:29:09,277
So that's the sort of work that's going

1622
01:29:09,360 --> 01:29:13,197
on. It's BSL1, BSL 2, which means, you

1623
01:29:13,280 --> 01:29:16,957
know, these are biosafety levels.
The lab that they set up, they've publicly

1624
01:29:17,040 --> 01:29:20,477
talked about the lab. >> So,
the lab itself is sort of like a

1625
01:29:20,560 --> 01:29:24,877
low-level research lab where they can
make simple proteins and test them in

1626
01:29:24,960 --> 01:29:27,357
the lab to see if the AI is doing a good

1627
01:29:27,440 --> 01:29:30,157
job discovering or postulating protein

1628
01:29:30,240 --> 01:29:34,477
folding or these sorts of theories and
to actually run really rapid testing and

1629
01:29:34,560 --> 01:29:38,237
experimentation cycles to see if these
proteins are what they're predicted to

1630
01:29:38,320 --> 01:29:41,517
be. >> Okay. So,
they like 4,000 agents against

1631
01:29:41,600 --> 01:29:43,597
it and do a hugging face with proteins.

1632
01:29:43,680 --> 01:29:45,917
Yeah. >> No, I mean I
think what they're really

1633
01:29:46,000 --> 01:29:49,757
doing is just having experimental proof
that what the software is predicting,

1634
01:29:49,840 --> 01:29:53,677
remember, go back to AlphaFold.
Alphold was predicting protein structure,

1635
01:29:53,760 --> 01:29:57,917
three-dimensional structure of a protein
from the DNA sequence that codes for

1636
01:29:58,000 --> 01:30:02,877
that protein. And so in order to prove
that, you had to basically predict what

1637
01:30:02,960 --> 01:30:06,637
the structure of a protein would be
from a DNA sequence, make the protein, and

1638
01:30:06,720 --> 01:30:09,677
then look at it and see if it actually
looks like it's supposed to look like.

1639
01:30:09,760 --> 01:30:12,317
And so in the case of some of the other
things that they're doing, they're

1640
01:30:12,400 --> 01:30:14,637
trying to discover function. I want to

1641
01:30:14,720 --> 01:30:19,437
find a an enzyme, which is a type of a
protein that degrades a certain molecule

1642
01:30:19,520 --> 01:30:23,277
that can be used to degrade, for example,
you know, a therapeutic target

1643
01:30:23,360 --> 01:30:27,117
or something in the body that we want to
get rid of. And so they'll start to kind

1644
01:30:27,200 --> 01:30:30,877
of make these predictions in software,
but then you need to test them in the

1645
01:30:30,960 --> 01:30:34,557
lab to say, okay, does that target
actually do what it does? Now, there are

1646
01:30:34,640 --> 01:30:36,797
literally hundreds of labs like this.

1647
01:30:36,880 --> 01:30:40,157
These are low-level research labs
that are not doing like gain of function

1648
01:30:40,240 --> 01:30:42,637
research. They're not making viruses.

1649
01:30:42,720 --> 01:30:45,757
They're not making pathogenic things
that can escape. That's why they're

1650
01:30:45,840 --> 01:30:49,597
BSL1, BSL 2. They're benchtop labs where

1651
01:30:49,680 --> 01:30:53,037
automation can help kind of reduce the
cost and increase the throughput of the

1652
01:30:53,120 --> 01:30:56,397
testing of what the software is making
predictions around. And ultimately the

1653
01:30:56,480 --> 01:30:59,437
objective here I think is for them to

1654
01:30:59,520 --> 01:31:03,757
get some proof that their software
is actually doing what it's predicting and

1655
01:31:03,840 --> 01:31:07,437
they have an incredible life sciences
function. I think at this point I don't

1656
01:31:07,520 --> 01:31:11,597
have as much insight on where Gemini
is at with the new models with Google, but

1657
01:31:11,680 --> 01:31:15,997
they have the best life sciences models
at Anthropic and I think this is the

1658
01:31:16,080 --> 01:31:20,157
fast track to basically seeing if
these models can be used by therapeutic

1659
01:31:20,240 --> 01:31:24,397
companies, by pharma companies,
by government labs to help discover new

1660
01:31:24,480 --> 01:31:26,637
therapeutics and so on. I I don't think

1661
01:31:26,720 --> 01:31:29,357
that the world should be scared away

1662
01:31:29,440 --> 01:31:32,077
from discovery, R&D, research and

1663
01:31:32,160 --> 01:31:34,477
development into new therapeutic

1664
01:31:34,560 --> 01:31:37,357
modalities as predicted by AI because we

1665
01:31:37,440 --> 01:31:39,437
heard the word wet lab in Wuhan and

1666
01:31:39,520 --> 01:31:44,717
everyone's like okay any lab is bad.
We still want to progress the frontier of

1667
01:31:44,800 --> 01:31:48,797
therapeutics of human health of discovery
and this is a really important

1668
01:31:48,880 --> 01:31:53,517
aspect of anthropic proving that their
models can add value here. I think

1669
01:31:53,600 --> 01:31:54,877
that's the summary of it. All right.

1670
01:31:54,960 --> 01:31:57,037
Thank you so much to our friends at IN.

1671
01:31:57,120 --> 01:32:00,237
They were the presenting they were
the presenting sponsor for this year's

1672
01:32:00,320 --> 01:32:03,437
all-in seminar fifth IN's AI cloud

1673
01:32:03,520 --> 01:32:05,917
lounge. It's packed for all three days.

1674
01:32:06,000 --> 01:32:10,477
Did a great meet and greet with INE
on Sunday and got to take some selfies and

1675
01:32:10,560 --> 01:32:15,160
met many of you there.
Thanks again to INER.

1676
01:32:15,680 --> 01:32:18,157
Niagen, did you catch Niogen in the

1677
01:32:18,240 --> 01:32:23,677
wellness hub for IVs? Tons of uh great
gifts and supplements being given away.

1678
01:32:23,760 --> 01:32:26,957
Great experience, great product. Great
job to our friends at Niogen for doing

1679
01:32:27,040 --> 01:32:29,997
the lounge again. Nigen is the official

1680
01:32:30,080 --> 01:32:32,637
NAD partner for the all-in pod. Yes,

1681
01:32:32,720 --> 01:32:35,437
Niogen BioScience created the NAD

1682
01:32:35,520 --> 01:32:38,877
category now has 45 plus clinical

1683
01:32:38,960 --> 01:32:43,837
studies and PayPal, my favorite way
to settle up. They had a great lounge on

1684
01:32:43,920 --> 01:32:47,197
site at the summit where everybody
got to hang out and get those beautiful

1685
01:32:47,280 --> 01:32:49,997
embossed luggage tags. and they hosted

1686
01:32:50,080 --> 01:32:51,917
the payments and fintech dinner which

1687
01:32:52,000 --> 01:32:54,237
was a huge hit. Let me tell you folks,

1688
01:32:54,320 --> 01:32:57,437
All-InSummit 2027 will sell out again

1689
01:32:57,520 --> 01:33:01,597
and the applications are open now.
Go to allin.com/events

1690
01:33:01,680 --> 01:33:05,757
and apply for your chance to be
in the room where it happens. What are you

1691
01:33:05,840 --> 01:33:08,557
waiting for folks? Allin.com/events.

1692
01:33:08,640 --> 01:33:10,477
All right, there you have it folks.

1693
01:33:10,560 --> 01:33:13,837
Episode 290, the all-in pod

1694
01:33:13,920 --> 01:33:18,397
for David Paulie Hakatia
and your Sultan of

1695
01:33:18,480 --> 01:33:22,557
Science Freeberg. I am the [music] world's
greatest moderator. See you next time.

1696
01:33:22,640 --> 01:33:25,357
Bye-bye.

1697
01:33:25,440 --> 01:33:31,800
>> Let your winners ride. >> Rainman David

1698
01:33:32,400 --> 01:33:35,757
>> and it said we open [music] sourced
it to the fans and they've just gone crazy

1699
01:33:35,840 --> 01:33:40,680
with it. >> Queen of

1700
01:33:45,200 --> 01:33:47,917
>> [music] >> Besties are gone.

1701
01:33:48,000 --> 01:33:52,327
>> That is my dog taking
a notice in your driveway.

1702
01:33:52,410 --> 01:33:55,757
[music] >> Oh man,
my appetasher will meet me at

1703
01:33:55,840 --> 01:33:58,877
>> We should all just get a room and just
have one big huge orgy cuz they're all

1704
01:33:58,960 --> 01:34:01,837
just useless. It's like this like sexual
tension [music] that they just need to

1705
01:34:01,920 --> 01:34:05,120
release somehow.

1706
01:34:06,080 --> 01:34:09,037
your feet.

1707
01:34:09,120 --> 01:34:13,937
We need to get merch going all in.

1708
01:34:14,020 --> 01:34:16,050
[music]

1709
01:34:18,560 --> 01:34:21,940
I'm going all in. [music]