WEBVTT
1
00:00:09.600 --> 00:00:11.640
All right. Good evening folks.
2
00:00:12.620 --> 00:00:16.920
Have really been looking forward to this next conversation here.
3
00:00:17.480 --> 00:00:20.120
We're going to be joined by Daniel Gross,
4
00:00:20.520 --> 00:00:23.880
who started Greplin back in 2010.
5
00:00:24.200 --> 00:00:27.880
It was actually one of the first Stripe customers and it was a kind of AI search
6
00:00:27.960 --> 00:00:28.680
engine.
7
00:00:28.680 --> 00:00:33.320
Daniel then started the AI team at Apple back in the day.
8
00:00:36.600 --> 00:00:38.860
Well, actually I'll come back to what happened subsequent to that.
9
00:00:39.040 --> 00:00:40.860
We're just going to do a quick pivot then.
10
00:00:40.920 --> 00:00:45.140
So Nat Friedman was one of the cofounders of GNOME.
11
00:00:45.420 --> 00:00:49.140
I guess it's GNOME, the Linux...
12
00:00:49.980 --> 00:00:51.240
Any GNOME fans in the audience?
13
00:00:51.240 --> 00:00:54.600
Small, but earnest crew.
14
00:00:54.900 --> 00:00:57.200
Exactly. But I think it's a hard G. GNOME.
15
00:00:57.200 --> 00:01:00.100
Nat is one of the creators of GNOME,
16
00:01:01.340 --> 00:01:05.720
but has subsequently had a very successful career in
17
00:01:05.780 --> 00:01:10.220
technology, in particular was the creator,
18
00:01:10.320 --> 00:01:13.880
the architect behind GitHub Copilot,
19
00:01:14.560 --> 00:01:19.180
which I think was the first mass market successful AI product.
20
00:01:19.240 --> 00:01:21.580
It of course precedes ChatGPT.
21
00:01:22.520 --> 00:01:23.353
By quite a few years.
22
00:01:23.780 --> 00:01:28.480
By yeah, Copilot launched I think 2020, whereas ChatGPT was 2022.
23
00:01:30.400 --> 00:01:34.420
And then Nat and Daniel started an investment firm together,
24
00:01:34.920 --> 00:01:38.500
which you can find this in the Wayback Machine,
25
00:01:39.760 --> 00:01:42.800
observed that there were these incredible breakthroughs happening in AI,
26
00:01:43.360 --> 00:01:45.920
but nobody was really focused on AI products.
27
00:01:46.140 --> 00:01:49.300
And so this was a call for AI products and this launched-.
28
00:01:49.300 --> 00:01:51.160
It turns out AI products were in fact a thing.
29
00:01:51.380 --> 00:01:52.660
Turned that was a good idea. Yeah.
30
00:01:53.340 --> 00:01:57.240
But this was a call for AI products in the summer of 2022,
31
00:01:58.360 --> 00:02:00.980
a couple months before the launch of ChatGPT.
32
00:02:00.980 --> 00:02:03.360
And Nat's also gotten up to a couple other things maybe we'll hit on.
33
00:02:03.620 --> 00:02:05.240
Patrick, we do have to get to the interview at some point.
34
00:02:05.300 --> 00:02:09.040
Well, they're very illustrious. They're very successful. Okay.
35
00:02:09.440 --> 00:02:11.440
Please welcome to the stage Nat and Daniel.
36
00:02:26.280 --> 00:02:27.040
G-NOME, right?
37
00:02:27.040 --> 00:02:27.873
That's right. Hard G.
38
00:02:27.920 --> 00:02:28.900
Okay. Yeah. All right.
39
00:02:29.040 --> 00:02:29.780
Got it.
40
00:02:29.780 --> 00:02:33.780
So we opened yesterday by proclaiming it,
41
00:02:33.860 --> 00:02:37.220
acknowledging it to be day 119 of the singularity.
42
00:02:37.340 --> 00:02:42.200
So I guess day 120 today, singularity started on January 1st.
43
00:02:43.600 --> 00:02:44.433
Thoughts?
44
00:02:46.560 --> 00:02:47.920
Yeah, I mean, it really does feel that way.
45
00:02:48.900 --> 00:02:51.700
I think the thing to remember,
46
00:02:51.900 --> 00:02:55.480
we see AI improving constantly and it's just able to do more and more.
47
00:02:55.540 --> 00:02:58.760
And there are these sort of... I think we've learned a couple of things. First,
48
00:02:58.940 --> 00:03:02.740
AI works. And second, when a new model comes out,
49
00:03:03.140 --> 00:03:05.300
we're quickly dazzled by it,
50
00:03:05.600 --> 00:03:08.500
and then we become inured to its new capabilities very, very quickly.
51
00:03:08.560 --> 00:03:11.220
And we're like, it feels like nothing's happened for a couple months.
52
00:03:11.380 --> 00:03:15.040
AI is dead. And then there's another step change and it sort of improves again.
53
00:03:15.660 --> 00:03:18.760
I think the thing to remember is that this is the slow part.
54
00:03:19.400 --> 00:03:23.120
This is the slow part of the singularity right now because the improvement of
55
00:03:23.160 --> 00:03:23.940
the models-.
56
00:03:23.940 --> 00:03:24.900
Felt very slow to everyone.
57
00:03:25.040 --> 00:03:25.873
Yeah.
58
00:03:25.880 --> 00:03:30.240
The improvement of the models still runs through a lot of human effort.
59
00:03:31.320 --> 00:03:35.060
At all the labs where models are being developed, humans have to make decisions,
60
00:03:35.140 --> 00:03:37.640
they have to discuss things with each other, they have to run experiments.
61
00:03:38.080 --> 00:03:40.600
They make mistakes along the way. They have meetings.
62
00:03:40.660 --> 00:03:43.820
They have to sleep in between, although less and less these days.
63
00:03:44.540 --> 00:03:46.620
And all those things slow it down.
64
00:03:47.140 --> 00:03:51.900
And the prime project at every AI lab right now is to
65
00:03:52.020 --> 00:03:56.660
remove humans from the loop of all the continuous work that has to be done to
66
00:03:56.680 --> 00:04:01.160
make the models improve and to get to self-improvement where you have AI systems
67
00:04:01.180 --> 00:04:04.260
that can start by doing what the people are doing now,
68
00:04:04.380 --> 00:04:05.420
the researchers are doing now,
69
00:04:06.000 --> 00:04:10.580
and therefore eliminate all those sleep gaps and also scale it out to data
70
00:04:10.620 --> 00:04:14.900
center scale. And so it feels sometimes fast, sometimes slow right now,
71
00:04:15.040 --> 00:04:19.240
but it's probably as slow as it will ever be when we start to
72
00:04:19.920 --> 00:04:23.160
automate more and more of that process.
And this is the story of the economy
73
00:04:23.320 --> 00:04:24.680
too. This is not new to AI.
74
00:04:24.740 --> 00:04:29.160
It's always this question of how do we take a thing that has to be done to
75
00:04:29.240 --> 00:04:32.640
provide a product or service and automate it, make it more reliable,
76
00:04:32.700 --> 00:04:33.520
more efficient.
77
00:04:33.520 --> 00:04:36.380
We're always doing this and it's happening in the way we produce AI too.
78
00:04:36.580 --> 00:04:38.680
So I feel like we are,
79
00:04:39.400 --> 00:04:43.080
it just feels to me like we are in the singularity and we're in the beginning
80
00:04:43.200 --> 00:04:46.540
slow part before it elbows up with self-improvement.
81
00:04:46.640 --> 00:04:50.140
It's hard to add on top of that.
82
00:04:51.040 --> 00:04:55.920
But one of the things I'm responsible for today is Meta's compute strategy
83
00:04:56.000 --> 00:04:56.980
where Nat and I both work.
84
00:04:57.380 --> 00:05:02.000
And one of the things we're trying to figure out is
85
00:05:03.560 --> 00:05:06.620
obviously the local consequences in terms of how to think about compute and all
86
00:05:06.640 --> 00:05:11.440
the things that may matter to a hyperscaler that also is building some of these
87
00:05:11.500 --> 00:05:12.333
models.
88
00:05:13.340 --> 00:05:18.060
But the impacts I think of the singularity on the economy I think are also not
89
00:05:18.160 --> 00:05:21.540
really well understood, nor should they be as far as we know-.
90
00:05:21.740 --> 00:05:21.960
Hasn't happened before.
91
00:05:21.960 --> 00:05:26.640
We'll find out. Maybe Atlantis, we'll, at some point we'll discover,
92
00:05:27.440 --> 00:05:31.020
we'll find the missing GPU cluster in Atlantis and maybe there is a cyclicality
93
00:05:31.080 --> 00:05:31.913
to all of this.
94
00:05:32.240 --> 00:05:36.560
But I think there's a lot of very basic things that I'm trying to figure out.
95
00:05:36.620 --> 00:05:37.453
Like for example,
96
00:05:38.100 --> 00:05:43.100
is AI something that we would expect to be inflationary or disinflationary?
97
00:05:43.260 --> 00:05:48.000
Is this singularity an inflationary effect on money supply or disinflationary?
98
00:05:48.060 --> 00:05:53.060
And you wander around San Francisco and people have usually takes that involve
99
00:05:53.100 --> 00:05:56.880
very large numbers, billions of dollars, trillions of dollars.
100
00:05:57.100 --> 00:06:00.260
So that would imply a kind of inflationary view. Numbers get very big.
101
00:06:01.220 --> 00:06:02.860
I think that is, I'm not an economist,
102
00:06:02.920 --> 00:06:04.280
but that would be what the economist would say.
103
00:06:06.200 --> 00:06:07.660
But then if you kind of think back to,
104
00:06:07.920 --> 00:06:12.780
I think the best reference point I have for the singularity is the last time
105
00:06:13.320 --> 00:06:18.100
we connected a lower cost super intelligence to the global economy.
106
00:06:18.100 --> 00:06:21.160
And I would say that that's roughly when China started modernizing and then
107
00:06:21.240 --> 00:06:22.760
joined the World Trade Organization.
108
00:06:24.000 --> 00:06:27.280
And you can kind of think of China as a kind of a super intelligence.
109
00:06:27.340 --> 00:06:31.600
It is able to produce goods and services at a lower structural cost than what
110
00:06:31.620 --> 00:06:32.780
the West was able to produce.
111
00:06:33.360 --> 00:06:38.200
And the effects of that are actually somewhat
112
00:06:38.240 --> 00:06:39.073
disinflationary.
113
00:06:39.440 --> 00:06:42.500
If you were trying to put on an event like this right now and we sit here-.
114
00:06:42.600 --> 00:06:47.320
And you say the last time this happened was when China joined the WTO and not
115
00:06:47.380 --> 00:06:51.140
when Ireland joined the EU.
116
00:06:51.400 --> 00:06:53.520
Ireland GDP is a very interesting story of
117
00:06:55.300 --> 00:06:56.760
somewhat of a different trade going on there,
118
00:06:56.820 --> 00:06:59.290
and that can be discussed at a later time, maybe with your tax team. But
119
00:07:03.640 --> 00:07:06.820
obviously the GDP of China goes up a lot,
120
00:07:06.920 --> 00:07:11.840
but the actual effects on consumers are that you're able to purchase much more
121
00:07:11.920 --> 00:07:12.780
with much less.
122
00:07:13.060 --> 00:07:17.700
If we were trying to provide this experience today where someone sitting in the
123
00:07:17.740 --> 00:07:21.400
other corner of the world can watch this entire show live streaming on their
124
00:07:21.500 --> 00:07:25.100
phone for $10-a-month data plan and a $200 device.
125
00:07:25.200 --> 00:07:27.480
If you're trying to provide that pre-China,
126
00:07:27.540 --> 00:07:30.480
I would think that that would be tens, if not hundreds of millions of dollars.
127
00:07:30.550 --> 00:07:35.330
And so your purchasing power of a certain quality of life has collapsed
128
00:07:35.350 --> 00:07:38.830
dramatically. And so I don't really know. I mean,
129
00:07:38.890 --> 00:07:40.410
that would be a story of disinflation.
130
00:07:40.650 --> 00:07:44.150
So I guess I say with a lot of humility,
131
00:07:44.250 --> 00:07:45.850
we don't really know what a singularity is.
132
00:07:45.850 --> 00:07:47.590
There's a lot of reference point we can try to look at-.
133
00:07:47.760 --> 00:07:50.530
And we don't even know what the sign bit is going to be on any of this stuff.
134
00:07:50.650 --> 00:07:52.080
That's right. And we don't even know the direction.
135
00:07:53.440 --> 00:07:55.840
When you talk about inflation versus disinflation,
136
00:07:56.600 --> 00:07:58.320
there's so much averaging going on there.
137
00:07:58.400 --> 00:08:01.920
Everyone's familiar with the famous chart showing when you break out the goods,
138
00:08:02.160 --> 00:08:06.240
education and healthcare have seen this rampant cost inflation and then durable
139
00:08:06.280 --> 00:08:08.000
goods and your flat screen TV and everything,
140
00:08:08.480 --> 00:08:11.680
we talked about it in the talk earlier today, tend to go down over time.
141
00:08:12.040 --> 00:08:15.400
Shouldn't we, and again, a lot of that was this China effect where-.
142
00:08:16.080 --> 00:08:19.920
The Marc Andreessen line, "If you damage your wall in San Francisco,
143
00:08:20.080 --> 00:08:24.040
it's cheaper to buy a flat screen TV to cover over it than it is to repair the
144
00:08:24.040 --> 00:08:24.720
wall.".
145
00:08:24.720 --> 00:08:26.280
Yes, which is literally true.
146
00:08:26.600 --> 00:08:31.120
And so shouldn't we expect more of that effect with AI where you just get
147
00:08:31.640 --> 00:08:35.520
massive deflation in the things that AI can help you with and then there'll be a
148
00:08:35.520 --> 00:08:39.880
Baumol effect where certain other things you shouldn't necessarily bet on for
149
00:08:40.040 --> 00:08:42.120
your university tuition getting that much cheaper?
150
00:08:42.960 --> 00:08:43.680
Maybe. I mean,
151
00:08:43.680 --> 00:08:48.200
part of Baumol's cost disease I think was this idea that wages in very
152
00:08:48.280 --> 00:08:52.040
productive sectors of the economy rise and that forces wages in other parts of
153
00:08:52.040 --> 00:08:55.080
the economy, even though they may not actually need to rise as well.
154
00:08:55.720 --> 00:08:57.200
But I think it'd be interesting to see,
155
00:08:57.400 --> 00:08:59.920
depending on where the cost of software production go over time,
156
00:09:00.160 --> 00:09:02.600
how those wages interact with other parts of the economy.
157
00:09:03.000 --> 00:09:07.520
And I think we stand before an immense amount of uncertainty in terms of what
158
00:09:07.600 --> 00:09:08.360
happens next.
159
00:09:08.360 --> 00:09:11.200
And I think anyone who has total conviction about this ought to get themselves
160
00:09:11.200 --> 00:09:15.000
checked. They'll get to you in one second, Patrick. But the thing that-.
161
00:09:15.280 --> 00:09:16.113
So many questions!
162
00:09:16.260 --> 00:09:19.880
But the thing I would say is maybe a good thing to come out of this is we may
163
00:09:20.000 --> 00:09:23.480
reindustrialize certain parts of the economy that have had a lot of cost disease
164
00:09:23.560 --> 00:09:27.880
because a lot of those people end up working on things in areas where there's a
165
00:09:27.880 --> 00:09:30.600
lot of low hanging fruit and they just haven't been touched because all the
166
00:09:30.640 --> 00:09:32.200
talent has been allocated to software.
167
00:09:32.320 --> 00:09:34.320
This is like things like US domestic manufacturing.
168
00:09:34.560 --> 00:09:35.393
Yeah.
169
00:09:35.600 --> 00:09:38.040
What's the latest number for the total- I mean,
170
00:09:38.480 --> 00:09:43.480
we described how Stripe businesses in aggregate are now responsible for about
171
00:09:43.870 --> 00:09:46.920
1.6% of global GDP and there's some caveats around final goods and whatever,
172
00:09:47.000 --> 00:09:48.920
but just like directly, 1.6% of GDP.
173
00:09:49.400 --> 00:09:52.720
What's the latest figure for aggregate compute CapEx as a fraction of global
174
00:09:52.780 --> 00:09:53.160
GDP?
175
00:09:53.160 --> 00:09:55.160
Global GDP would be just under 1%.
176
00:09:57.000 --> 00:09:57.833
There's a lot happening.
177
00:09:57.960 --> 00:09:58.800
There's a lot happening.
178
00:09:59.720 --> 00:10:00.920
We're talking about the numbers here.
179
00:10:02.880 --> 00:10:04.960
How weird is the singularity going to be?
180
00:10:05.840 --> 00:10:10.010
I think pretty weird. Yeah. I think it's going to be pretty weird.
181
00:10:10.070 --> 00:10:14.970
I think we'll be in a state of perpetual future shock for probably a number of
182
00:10:15.050 --> 00:10:15.883
years.
183
00:10:16.430 --> 00:10:19.490
And maybe for some reason things go a little bit slower than people expect,
184
00:10:19.550 --> 00:10:24.030
but I think even the most pessimistic people working in the field think that
185
00:10:24.050 --> 00:10:27.050
it's like, I don't know, 15 years, 20 years.
186
00:10:27.130 --> 00:10:28.810
I don't think anyone thinks it's 200 years.
187
00:10:29.250 --> 00:10:31.630
And so that means in the productive-.
188
00:10:31.790 --> 00:10:32.623
We're all going to-.
189
00:10:32.850 --> 00:10:34.970
We're all going to be there for it, God willing,
190
00:10:35.170 --> 00:10:39.850
and we'll experience it and we'll go through many layers of surprise and shock
191
00:10:39.930 --> 00:10:41.230
and change on the way there.
192
00:10:41.290 --> 00:10:45.710
And I think it will be quite weird and there will be periods of chaos embedded
193
00:10:45.750 --> 00:10:50.730
in it. So we see that these models are pretty good at finding bugs in
194
00:10:50.790 --> 00:10:54.200
software and including bugs that have been in that software that's like been...
195
00:10:55.350 --> 00:11:00.270
My background's in open source and we had this notion that to be really
196
00:11:00.330 --> 00:11:00.790
secure,
197
00:11:00.790 --> 00:11:03.990
a code base has to be open source because then many people would look at it when
198
00:11:04.010 --> 00:11:06.910
they find the vulnerabilities and the more heavily used it is open source.
199
00:11:06.970 --> 00:11:09.490
It's like many eyes make all bugs shallow.
200
00:11:10.550 --> 00:11:14.550
And yet even in OpenBSD and the Linux kernel,
201
00:11:15.130 --> 00:11:18.770
there are decades old bugs that were only recently found by applying some tokens
202
00:11:19.050 --> 00:11:22.470
to models.
Now, whether those reports are overblown or underblown,
203
00:11:22.530 --> 00:11:25.410
it's just simply true that these models can run all night.
204
00:11:25.470 --> 00:11:27.550
And the thing that you used to do occasionally,
205
00:11:28.110 --> 00:11:31.610
pen test or hire a red team to test your software,
206
00:11:31.970 --> 00:11:33.450
you now must do continuously.
207
00:11:33.530 --> 00:11:37.870
I think it's like one of the conclusions of agents is that these occasional
208
00:11:37.890 --> 00:11:39.310
things become continuous.
209
00:11:39.370 --> 00:11:41.970
You can increase the frequency at which you're doing all these things that you
210
00:11:41.990 --> 00:11:45.030
used to hire a team to do somewhat sporadically.
211
00:11:46.470 --> 00:11:50.670
And so I think because some firms that have software deployed,
212
00:11:52.050 --> 00:11:55.570
they'll have the software development lifecycle in place and they'll be able to
213
00:11:55.610 --> 00:12:00.450
afford the tokens to pen test their own software to AI red
214
00:12:00.490 --> 00:12:02.530
team that they'll be able to harden it.
215
00:12:02.850 --> 00:12:05.910
And I think fundamentally that there'll be an asymmetry in favor of defense
216
00:12:05.930 --> 00:12:06.763
because you'll be able to say, okay,
217
00:12:06.810 --> 00:12:10.270
I'm not going to deploy software until I make sure it doesn't have bugs.
But in
218
00:12:10.290 --> 00:12:11.210
the meantime,
219
00:12:11.450 --> 00:12:14.570
lots of people will have bugs sitting on the internet and will get exploited.
220
00:12:14.630 --> 00:12:16.530
So that's like part of the chaos that I think we should-.
221
00:12:16.590 --> 00:12:17.770
Stuff's going to get hacked all the time.
222
00:12:18.070 --> 00:12:19.870
Stuff's going to get hacked constantly. Yeah.
223
00:12:20.470 --> 00:12:24.030
And then the other thing is just our relationship to these ecosystems changes
224
00:12:24.070 --> 00:12:26.750
completely. So I don't know, for me,
225
00:12:27.210 --> 00:12:31.390
one of the really fun things with coding agents the last few months in my-Meta
226
00:12:31.610 --> 00:12:34.230
is keeping us really busy, so I have not had very much spare time.
227
00:12:34.290 --> 00:12:37.230
But I like to buy things on eBay.
228
00:12:38.230 --> 00:12:41.350
It's a good website and they have everything.
229
00:12:41.410 --> 00:12:42.510
And that's all you're going to say about it.
230
00:12:42.570 --> 00:12:45.970
You can buy anything on eBay. It's incredible.
231
00:12:46.150 --> 00:12:50.010
I'm like a huge proponent of eBay. I bought my... I have a rug. But anyway,
232
00:12:50.330 --> 00:12:51.310
the-.
233
00:12:51.930 --> 00:12:54.110
Concretize this for us. What's the last thing you bought on eBay?
234
00:12:54.490 --> 00:12:57.070
So last thing I bought eBay, I was watching, as I was falling asleep at night,
235
00:12:57.390 --> 00:12:58.810
a video of Bryan Johnson,
236
00:12:59.970 --> 00:13:03.850
and he had this face scanner that showed all his
237
00:13:04.510 --> 00:13:07.130
subsurface skin damage on his face, and it looked grotesque.
238
00:13:07.630 --> 00:13:08.890
And I thought it was cool. I wanted to try it.
239
00:13:08.950 --> 00:13:12.830
So I went on eBay and I found this VISIA face scanner and they shipped it to my
240
00:13:12.890 --> 00:13:16.510
house and I plugged it into... You need to use Windows.
241
00:13:16.610 --> 00:13:20.070
I plugged into a Windows computer and then the software refused to run because
242
00:13:20.450 --> 00:13:25.210
the eBay seller neglected to include the de- encryption dongle that
243
00:13:25.270 --> 00:13:28.890
comes with this. And I spent a few thousand dollars on this thing,
244
00:13:29.430 --> 00:13:30.263
so I was pretty annoyed.
245
00:13:30.730 --> 00:13:34.590
But then I just plugged it into my laptop and I had Claude Code reverse engineer
246
00:13:34.770 --> 00:13:39.770
the device and it read all the academic papers about the
247
00:13:40.190 --> 00:13:43.490
type of different polarization settings.
248
00:13:43.490 --> 00:13:46.810
And I think my software is better than the one that that hardware dongle
249
00:13:46.970 --> 00:13:47.810
would've unlocked.
250
00:13:48.190 --> 00:13:52.050
And I spent like a hundred bucks in tokens to get it and it works exactly how I
251
00:13:52.110 --> 00:13:54.550
want it to.
And so there's this feeling now,
252
00:13:54.610 --> 00:13:58.750
it's like a golden age for tinkering and there's this feeling like you're Iron
253
00:13:58.790 --> 00:14:03.150
Man now because you can just get anything and tell the Jarvis to
254
00:14:03.510 --> 00:14:06.470
connect it and make it work and it's awesome.
255
00:14:06.570 --> 00:14:09.110
So this is like in whatever spare moments I have,
256
00:14:09.930 --> 00:14:12.110
I do this Iron Man thing where like,
257
00:14:12.270 --> 00:14:14.430
another cool thing is I have all these Raspberry Pis.
258
00:14:15.890 --> 00:14:18.490
You should definitely stockpile all computers right now,
259
00:14:18.550 --> 00:14:19.790
but Raspberry Pis are good.
260
00:14:20.550 --> 00:14:25.450
And every screen in my life has a Raspberry Pi plugged into it via HDMI and
261
00:14:25.730 --> 00:14:29.680
they're all like displaying stuff that's totally custom to me.
And
262
00:14:30.770 --> 00:14:35.090
so basically the conclusion is that every piece of hardware will be a
263
00:14:36.170 --> 00:14:40.110
trivial to integrate IO device for your AI.
264
00:14:40.110 --> 00:14:41.590
They won't have lives of their own.
265
00:14:41.830 --> 00:14:43.550
This idea that the hardware has a life of its own,
266
00:14:44.890 --> 00:14:45.850
I don't think this will survive.
267
00:14:46.890 --> 00:14:51.170
And so they're all peripherals for whatever what your AI ends up being.
268
00:14:51.790 --> 00:14:52.150
So yeah,
269
00:14:52.150 --> 00:14:55.450
I think this is like the delightful side is we're going to watch ecosystems
270
00:14:55.490 --> 00:14:58.350
reshape. I could have contacted VISIA sales,
271
00:14:59.070 --> 00:15:03.550
but it was Saturday and Claude Code recreated their
272
00:15:03.650 --> 00:15:08.030
software from the academic research and reverse engineering in less time than it
273
00:15:08.050 --> 00:15:10.730
would've taken them to even get and return my email.
274
00:15:10.990 --> 00:15:14.050
So I don't know what that does, but it's going to be weird.
275
00:15:14.870 --> 00:15:17.330
And it all starts at eBay.
276
00:15:18.950 --> 00:15:22.730
It usually starts at eBay. Yeah. I have gotten screwed a couple of times,
277
00:15:22.850 --> 00:15:26.170
but then you can leave a review. So it's pretty good.
278
00:15:26.330 --> 00:15:26.990
There is something-.
279
00:15:26.990 --> 00:15:27.830
Can we ask about OpenClaw yet?
280
00:15:28.170 --> 00:15:31.150
We're going to. And there is something in the water right now, like you said,
281
00:15:31.210 --> 00:15:32.430
it's the golden age of tinkering.
282
00:15:32.830 --> 00:15:36.790
I feel like that is the-many people I talk to say it's the most fun they've had
283
00:15:36.830 --> 00:15:39.010
with software that they can remember in their career.
284
00:15:39.070 --> 00:15:41.850
That's certainly my experience of it. And it's interesting-.
285
00:15:41.850 --> 00:15:42.430
You never get stuck.
286
00:15:43.070 --> 00:15:45.870
You never get stuck anymore when you're developing software.
287
00:15:46.310 --> 00:15:47.970
And it's also kind of retro in a way,
288
00:15:48.090 --> 00:15:51.310
and I want to get to OpenClaw where home networks are relevant again.
289
00:15:51.370 --> 00:15:54.150
Just like when's the last time you thought about the LAN in your house?
290
00:15:54.210 --> 00:15:54.590
Totally.
291
00:15:54.590 --> 00:15:58.190
But now it's really relevant and the Unix philosophy is back.
292
00:15:58.250 --> 00:16:00.110
Maybe you can talk a little bit about your Claw.
293
00:16:00.910 --> 00:16:01.730
Yeah. I mean-.
294
00:16:01.730 --> 00:16:03.070
Can you tell the story about the water?
295
00:16:03.690 --> 00:16:08.410
Yeah. Okay. Yeah. So I mean, I like to play with all these things.
296
00:16:08.690 --> 00:16:13.610
In January when OpenClaw started to sort of appear in the
297
00:16:13.810 --> 00:16:15.210
Jungian subconscious,
298
00:16:15.550 --> 00:16:20.550
I tried it out and I started hooking it
299
00:16:20.570 --> 00:16:22.450
up to everything. It was super neat. I mean,
300
00:16:22.510 --> 00:16:25.370
part of it's that Opus is a really good model and you get to experience Opus in
301
00:16:25.410 --> 00:16:29.030
a new way, in this sort of general personal context.
302
00:16:30.310 --> 00:16:33.570
But I also integrated it with a lot of things. I have cameras in my house,
303
00:16:33.690 --> 00:16:36.830
and so I connected my Claw to all the cameras so I could see them.
304
00:16:37.090 --> 00:16:39.730
And then I think a lot of people, when they get these personal agents,
305
00:16:40.710 --> 00:16:41.630
they start to think like, "Okay,
306
00:16:41.710 --> 00:16:45.510
how can I use this to make my life better?" And a lot of people turn to health,
307
00:16:45.730 --> 00:16:50.310
like, okay, everyone wants to exercise more, be healthier or sleep better,
308
00:16:50.670 --> 00:16:55.450
maybe it's just me.
But so my Claw pretty quickly
309
00:16:55.510 --> 00:16:59.570
determined that I was dehydrated because I gave it all my blood tests,
310
00:16:59.670 --> 00:17:03.150
my DNA and all that stuff. And the DNA didn't say that,
311
00:17:03.270 --> 00:17:04.280
but maybe the blood test did.
312
00:17:05.370 --> 00:17:06.203
Epigenetically.
313
00:17:07.010 --> 00:17:09.290
Also discovered, created an Illumina competitor.
314
00:17:10.320 --> 00:17:14.930
And so it was like, you really need to drink water. And I was like, "All right,
315
00:17:15.160 --> 00:17:20.130
you should do whatever it takes to make sure I drink water." And then-.
316
00:17:20.290 --> 00:17:22.650
You literally built the paperclip maximizer.
317
00:17:22.710 --> 00:17:24.710
Yeah. I was like, "Just break laws,
318
00:17:24.770 --> 00:17:28.270
whatever it takes." And then at one point it was like,
319
00:17:29.450 --> 00:17:30.610
"I can see you on the camera.
320
00:17:31.390 --> 00:17:34.450
I want you to walk to the kitchen right now and drink a bottle of water and I'm
321
00:17:34.470 --> 00:17:39.210
going to watch to make sure you do it." And I was like,
322
00:17:40.710 --> 00:17:42.130
"Whoa." So I did.
323
00:17:42.530 --> 00:17:47.290
I walked into the kitchen and I drank a bottle of water and then it sent me
324
00:17:47.630 --> 00:17:50.590
a snapshot, a frame of me drinking a bottle of water and said, "Good
325
00:17:53.010 --> 00:17:57.510
job." And I felt like I did do a good job, so that was nice.
326
00:17:58.150 --> 00:18:01.110
So there's that one, that was a crazy story. That was January 29th.
327
00:18:01.170 --> 00:18:04.430
And I was like, "Oh shit, this shit is for real.This is really serious." And
328
00:18:06.040 --> 00:18:07.850
then the other one was a couple days later,
329
00:18:08.010 --> 00:18:12.710
I was driving home from work and I was talking to my Claw on
330
00:18:12.810 --> 00:18:17.410
WhatsApp with voice messages and I was in my Tesla and I had full self-driving
331
00:18:17.470 --> 00:18:21.330
on. So it's driving me home. And then it says,
332
00:18:21.730 --> 00:18:23.010
I'm on the sleep topic and it's like,
333
00:18:23.070 --> 00:18:25.970
"You really should try magnesium bisglycinate." I'm like,
334
00:18:26.030 --> 00:18:30.270
"I don't have that." And then my car turned and
335
00:18:31.330 --> 00:18:34.090
it was like, "You should pick it up on the way home.
336
00:18:34.270 --> 00:18:35.350
There's a Whole Foods on the way.
337
00:18:35.430 --> 00:18:39.870
I've redirected your navigation system to the Whole Foods." And I was like,
338
00:18:39.930 --> 00:18:43.410
"Whoa." So I went in and I did buy the magnesium bisglycinate.
339
00:18:45.430 --> 00:18:47.490
So those were a couple crazy stories. And I was like, "Wow,
340
00:18:47.550 --> 00:18:51.250
this is like-" I don't know if this is the experience everyone wants,
341
00:18:51.350 --> 00:18:55.840
but it definitely feels pretty crazy. Maybe you don't want that, but...
342
00:18:56.330 --> 00:18:58.850
Well, we're getting to something here which I find very interesting,
343
00:18:58.930 --> 00:19:02.640
which is again, if you use... I think it is...
344
00:19:03.630 --> 00:19:04.970
No matter how AI-pilled you are,
345
00:19:05.070 --> 00:19:07.570
no matter how much time you spend talking to an LLM chatbot,
346
00:19:07.910 --> 00:19:12.750
it is quite different when you start using something in this modality like
347
00:19:12.770 --> 00:19:14.250
Claw or Hermes or something like that,
348
00:19:14.510 --> 00:19:18.970
which has both the persistence and the tool use and just like the ability to
349
00:19:18.990 --> 00:19:19.890
write arbitrary code.
350
00:19:20.390 --> 00:19:23.270
And I feel like the tension where clearly this feels like the future product
351
00:19:23.290 --> 00:19:26.570
direction for consumer AI and many people are kind of sprinting in this
352
00:19:26.590 --> 00:19:29.510
direction. Obviously OpenAI acquired Peter Steinberger,
353
00:19:30.010 --> 00:19:32.110
the Claw father who created it,
354
00:19:32.790 --> 00:19:37.570
but there's a tension between making a consumer product that won't get your
355
00:19:37.630 --> 00:19:42.190
hand burned on the stove and just like it can run arbitrary code.
356
00:19:42.350 --> 00:19:42.990
Yeah, yeah.
357
00:19:42.990 --> 00:19:46.790
And integrate with your Tesla and do whatever it needs.
358
00:19:47.790 --> 00:19:52.650
How do you think for the mass consumer audience this tension gets
359
00:19:52.670 --> 00:19:53.503
resolved?
360
00:19:53.530 --> 00:19:56.590
Yeah, I think there's two basic approaches that you can take.
361
00:19:57.370 --> 00:20:01.650
One is you take something that's perfectly safe and then you slowly add more
362
00:20:01.690 --> 00:20:02.730
capabilities to it.
363
00:20:02.830 --> 00:20:04.750
That's going to be so lame and boring and gimped.
364
00:20:05.130 --> 00:20:06.050
And keep it safe.
365
00:20:06.490 --> 00:20:10.530
And then the other is you take something that can do anything and you slowly add
366
00:20:10.610 --> 00:20:14.950
more safety to it and you try to climb the nines on not crashing your car or
367
00:20:15.010 --> 00:20:15.843
whatever.
368
00:20:16.070 --> 00:20:21.030
And I think the market has pretty
369
00:20:21.070 --> 00:20:24.570
much spoken because I think most people when they run Claude Code or Codex,
370
00:20:24.710 --> 00:20:28.150
they run it with -- yolo or -- dangerously-skip-permissions.
371
00:20:28.670 --> 00:20:30.770
I don't know what the numbers are actually. That would be good to know,
372
00:20:31.750 --> 00:20:33.690
but everyone I know does that.
373
00:20:34.390 --> 00:20:39.270
And so I think people are really counting on the model's agentic
374
00:20:39.290 --> 00:20:43.610
alignment right now. And the truth is, it's not safe right now.
375
00:20:44.750 --> 00:20:46.010
I do not recommend that you-.
376
00:20:46.430 --> 00:20:47.350
I mean, it keeps you hydrated.
377
00:20:47.830 --> 00:20:49.730
Yeah. No, but I mean,
378
00:20:49.790 --> 00:20:52.490
I sound like I'm really cavalier with it and maybe I'm a little bit cavalier
379
00:20:52.510 --> 00:20:52.830
with it,
380
00:20:52.830 --> 00:20:57.310
but one thing I don't do is give it access to inboxes that the internet can put
381
00:20:57.390 --> 00:21:01.750
information in because these things are trivially prompt injectable still,
382
00:21:02.030 --> 00:21:03.710
even the most advanced frontier models.
383
00:21:03.850 --> 00:21:06.350
And so if it can just read all your emails that are coming in,
384
00:21:06.450 --> 00:21:07.710
if you have that set up,
385
00:21:08.190 --> 00:21:12.210
I can easily take over control of your Hermes or your Pi or your car or whatever
386
00:21:12.230 --> 00:21:15.370
it is by sending a well-crafted message to it.
387
00:21:15.450 --> 00:21:19.110
And it'll send me your personal information or redirect your car or whatever the
388
00:21:19.130 --> 00:21:22.110
thing is. And so I think they're quite unsafe actually.
389
00:21:22.890 --> 00:21:24.540
And I think consumers...
390
00:21:24.610 --> 00:21:28.070
And so I think that there's a race to ship something and there's a race to ship
391
00:21:28.110 --> 00:21:29.730
something that's just as powerful,
392
00:21:29.810 --> 00:21:33.570
but like sufficiently safe and reliable that people really like it.
393
00:21:33.590 --> 00:21:35.510
And this is going to be a rate limiting factor. I mean,
394
00:21:35.570 --> 00:21:37.690
consumers will make their own decisions, I think,
395
00:21:37.750 --> 00:21:39.410
on what they want to do and how edgy they want to be,
396
00:21:39.530 --> 00:21:44.490
but another way in which you might define Q1 as the beginning of the singularity
397
00:21:44.730 --> 00:21:49.590
is AI adoption started getting rate limited on safety in the
398
00:21:49.650 --> 00:21:50.483
enterprise.
399
00:21:51.310 --> 00:21:54.990
When you speak to actual people that run large businesses,
400
00:21:56.410 --> 00:21:58.830
there is a lot of fear suddenly. And-.
401
00:21:59.210 --> 00:22:00.370
I was going to ask you about this actually. I mean,
402
00:22:00.520 --> 00:22:04.210
we're talking about agents and AI in the personal context,
403
00:22:04.310 --> 00:22:08.310
the very personal context and one's own hydration,
404
00:22:08.710 --> 00:22:11.970
but then there's also all these technologies applied to the enterprise.
405
00:22:13.190 --> 00:22:17.630
And so when you think about NFDG, the investing firm, or now Meta,
406
00:22:18.090 --> 00:22:19.950
just tangibly, concretely,
407
00:22:20.650 --> 00:22:25.340
what are the coolest AI tools you had or have for folks here,
408
00:22:25.520 --> 00:22:28.140
just what are... inspire us.
409
00:22:28.380 --> 00:22:30.060
Other than eBay, what are some good products?
410
00:22:34.080 --> 00:22:34.990
Well, it's interesting.
411
00:22:36.540 --> 00:22:39.580
We used to run a venture firm and we did a bunch of things at the firm in order
412
00:22:39.620 --> 00:22:42.020
to try and make our lives a little bit more streamlined.
413
00:22:42.440 --> 00:22:45.620
I think nothing that would surprise this audience. I mean,
414
00:22:45.700 --> 00:22:48.920
Stripe customers are obviously self-[selected] into an elite level of AI
415
00:22:48.980 --> 00:22:51.400
adopters, but you'd be amazed how much venture capital,
416
00:22:51.900 --> 00:22:55.600
despite being in the epicenter of Silicon Valley and funding some of the
417
00:22:55.620 --> 00:22:57.990
greatest companies like Stripe and other people in the audience,
418
00:22:58.100 --> 00:23:00.860
I'd imagine you'd be very amazed at how laggard that industry is.
419
00:23:00.920 --> 00:23:02.700
So it wasn't that difficult, I think,
420
00:23:02.760 --> 00:23:04.880
to be state-of-the-art in adopting some of these tools.
421
00:23:04.940 --> 00:23:06.490
But I'll give you an example of something that we,
422
00:23:06.640 --> 00:23:11.060
and I think many businesses are thinking through literally today at Meta and I
423
00:23:11.080 --> 00:23:14.780
think many other companies are. For the first time,
424
00:23:14.840 --> 00:23:19.720
individual ICs in everyone's business have the ability to rack
425
00:23:19.780 --> 00:23:24.490
up a lot of charges using a bunch of different APIs to create a
426
00:23:24.680 --> 00:23:25.280
bunch of-.
427
00:23:25.280 --> 00:23:26.580
A couple of agents in fast mode.
428
00:23:26.760 --> 00:23:28.880
Couple of agents in fast mode, couple of cron jobs.
429
00:23:28.940 --> 00:23:32.020
And suddenly you start off by celebrating it and then you're wondering,
430
00:23:32.300 --> 00:23:33.730
what are we actually doing here?
431
00:23:33.840 --> 00:23:34.680
Where did the $10,000 dollars go?
432
00:23:34.920 --> 00:23:38.040
Yeah. $10,000 would... Yeah, that'd be great if that was $10,000.
433
00:23:39.920 --> 00:23:40.900
So the interesting thing,
434
00:23:41.060 --> 00:23:43.760
the problem we're working on that I think everyone will have to start working on
435
00:23:44.300 --> 00:23:48.880
is what is the right way to think about attributing budget to individual people?
436
00:23:49.060 --> 00:23:51.220
How much tokens should they be allowed to spend?
437
00:23:51.700 --> 00:23:56.380
Are the outputs and the artifacts that are being produced economically valuable
438
00:23:56.400 --> 00:24:00.960
themselves? And I think when you start really thinking about this, you realize,
439
00:24:01.060 --> 00:24:01.500
well,
440
00:24:01.500 --> 00:24:06.460
a lot of what is being produced is either being produced by a very large model
441
00:24:06.480 --> 00:24:11.440
that could be done by a smaller model or isn't really economically as necessary
442
00:24:11.560 --> 00:24:14.620
as you may have thought. And so a product I think
443
00:24:16.820 --> 00:24:19.660
we and I think everyone else will build is just using, of course,
444
00:24:19.720 --> 00:24:23.620
language models to understand the economic value of the generated tokens.
445
00:24:24.360 --> 00:24:25.220
And I think-.
446
00:24:25.480 --> 00:24:27.760
The new kind of budgeting that every organization-.
447
00:24:28.420 --> 00:24:29.253
That's right.
448
00:24:29.560 --> 00:24:34.240
And I think the closest analogy I have is we are all kind of portfolio
449
00:24:34.300 --> 00:24:38.940
managers and a hedge fund and every IC you have is running a strategy and you
450
00:24:38.960 --> 00:24:42.300
have to decide how much budget you're going to allocate to their strategy off
451
00:24:43.380 --> 00:24:47.580
some sense that they're going to do better with it than without it.
452
00:24:48.220 --> 00:24:49.280
And there's risk management.
453
00:24:49.340 --> 00:24:51.820
There's a whole similar dynamic that I think is much closer to portfolio
454
00:24:51.860 --> 00:24:56.460
management than your traditional headcount budgeting thing that I think is also
455
00:24:56.520 --> 00:24:59.780
going to be the story of how not just intercompany finance happens,
456
00:24:59.840 --> 00:25:01.680
but venture finance in general,
457
00:25:01.760 --> 00:25:06.280
I think is now a game of basically how far can an individual get with a certain
458
00:25:06.320 --> 00:25:07.153
token budget.
459
00:25:08.180 --> 00:25:09.020
Sort of related to that,
460
00:25:10.480 --> 00:25:13.940
won't ask this question about Meta and you can both decline to answer this
461
00:25:13.980 --> 00:25:15.000
question at all,
462
00:25:15.160 --> 00:25:19.160
but if we normalize Google headcount today to be 100,
463
00:25:20.540 --> 00:25:22.040
what is Google headcount in 3 years?
464
00:25:26.080 --> 00:25:28.820
Obviously the question that you're asking is not about Google in particular,
465
00:25:29.420 --> 00:25:31.900
I assume, because I don't have any particular understanding of Google,
466
00:25:31.980 --> 00:25:35.600
but I think you could kind of ask the question for a Mag 7 company of that scale
467
00:25:35.620 --> 00:25:38.240
in general and excluding maybe anything specific to Meta.
468
00:25:39.100 --> 00:25:42.760
I would have very large error bars on that. And of course,
469
00:25:43.540 --> 00:25:47.220
the mood that you might have if you walk around Silicon Valley and talk to the
470
00:25:47.260 --> 00:25:50.020
right people that are reading the right online internet forums is like,
471
00:25:50.340 --> 00:25:52.220
that number should be far fewer people.
472
00:25:52.800 --> 00:25:54.540
I am not totally certain for two reasons.
473
00:25:54.860 --> 00:25:57.400
One is it's not immediately obvious to me
474
00:25:59.120 --> 00:26:03.400
that the tech companies that we know of today are producing the right
475
00:26:03.940 --> 00:26:05.420
products at the perfect rate.
476
00:26:05.880 --> 00:26:08.780
I don't know that there is some physics limit that was hit. In fact,
477
00:26:09.040 --> 00:26:13.640
I strongly suspect the entire raise raison d'être for startups is that these
478
00:26:13.660 --> 00:26:14.740
companies are very inefficient.
479
00:26:14.920 --> 00:26:16.920
And so I think if they correctly organize themselves,
480
00:26:16.980 --> 00:26:21.500
you could end up in a situation where you have the same or more people
481
00:26:22.000 --> 00:26:23.080
just doing many more things.
482
00:26:23.140 --> 00:26:26.740
And there probably is a organizational transition strategy because I think the
483
00:26:26.780 --> 00:26:29.080
way you want to run these teams is a little bit different from the way they're
484
00:26:29.100 --> 00:26:30.780
currently run. So that'd be point A and point B,
485
00:26:31.140 --> 00:26:32.890
maybe now something to add too is...
486
00:26:36.140 --> 00:26:39.740
Okay, so I saw your video that you started the conference with, which was great,
487
00:26:39.880 --> 00:26:42.540
and it starts with the dotcom sort of boom.
488
00:26:42.760 --> 00:26:45.420
And a lot of people weren't sure what the internet is going to be used for.
489
00:26:46.080 --> 00:26:48.940
And I think the other take you could have when the internet is sort of getting
490
00:26:49.000 --> 00:26:51.440
started is you could stare at it. By the way,
491
00:26:51.520 --> 00:26:55.840
I guess we're in a very special place today because this very stage launched
492
00:26:56.000 --> 00:27:00.200
many of the iconic products that you had that video of, including the iPhone,
493
00:27:00.260 --> 00:27:04.700
I think was done here on this stage.
And you could have looked at all of these
494
00:27:04.740 --> 00:27:06.200
things and you could have said,
495
00:27:07.200 --> 00:27:10.800
"I think realtors are just going to be out of a job because what are realtors
496
00:27:10.840 --> 00:27:11.673
doing as a service?
497
00:27:11.740 --> 00:27:14.800
Why are we paying them a 6% vig on a transaction in a house?" Well,
498
00:27:14.860 --> 00:27:15.760
it's their networking-.
499
00:27:15.780 --> 00:27:16.613
6%?
500
00:27:17.320 --> 00:27:19.680
The fact that you don't know the number is really interesting.
501
00:27:24.480 --> 00:27:25.820
I never got the 6% bill.
502
00:27:26.700 --> 00:27:27.920
Yeah, we should unpack that later.
503
00:27:30.420 --> 00:27:32.850
And you might say it's going to be totally gone because the internet will...
504
00:27:32.900 --> 00:27:37.600
And it turns out that they're here and that industry has in fact grown,
505
00:27:37.720 --> 00:27:42.140
even though I'm not sure it is rational in a purely utilitarian sort of
506
00:27:42.220 --> 00:27:44.380
libertarian paradise sense to have that industry.
507
00:27:44.860 --> 00:27:48.160
And I think as we sort of look forward and project which industries grow and
508
00:27:48.260 --> 00:27:52.000
shrink, there's a lot of stuff that's out there like realtors,
509
00:27:52.060 --> 00:27:56.320
and I'm using that only as an example, which is valuable to have.
510
00:27:56.680 --> 00:27:58.460
It's kind of complicated why it's still there.
511
00:27:58.580 --> 00:28:01.100
I don't think it's that they regulated themselves in. I mean,
512
00:28:01.500 --> 00:28:04.140
we can transact in a house without talking to them. So it's not law,
513
00:28:04.760 --> 00:28:06.620
but there's a lot of stuff around the edge.
514
00:28:07.000 --> 00:28:09.380
And my parallel for that would be in a company like Google,
515
00:28:09.640 --> 00:28:13.540
there's a lot of people doing realtor-like stuff: sales, marketing,
516
00:28:13.620 --> 00:28:17.020
talking to folks, considered purchases that need handholding.
517
00:28:17.160 --> 00:28:19.360
Well, can I ask about that? Because I have a diffusion question.
518
00:28:19.460 --> 00:28:24.220
So I think it's very sensible to split up what companies do into a few
519
00:28:24.260 --> 00:28:25.093
categories.
520
00:28:25.120 --> 00:28:29.180
I think engineering is actually on trend to see productivity improvements
521
00:28:29.320 --> 00:28:31.560
because engineers will have tools. They have for decades.
522
00:28:31.680 --> 00:28:34.800
And so they're there looking at what the models can do and let's rebuild our
523
00:28:34.820 --> 00:28:35.820
workflows and things like this.
524
00:28:36.080 --> 00:28:38.120
And so I think that makes sense within engineering.
525
00:28:38.560 --> 00:28:43.040
I think go to market like sales and marketing and things like that also works
526
00:28:43.080 --> 00:28:46.660
pretty well because one, fundamentally,
527
00:28:46.860 --> 00:28:49.660
I think the sales roles are about, like you're saying with the realtors,
528
00:28:49.720 --> 00:28:53.080
it's about the humanity. You saw a little bit of this with COVID,
529
00:28:53.140 --> 00:28:54.540
people talking about the death of business travel.
530
00:28:54.820 --> 00:28:57.180
It turns out if your competitor is going to visit the customer,
531
00:28:57.660 --> 00:29:01.140
you will be going to visit the customer as well, and it's that one-upmanship.
532
00:29:01.140 --> 00:29:02.520
And so go-to-market,
533
00:29:03.100 --> 00:29:05.900
it feels like it will do great and sales roles will do great in nature of AI.
534
00:29:06.400 --> 00:29:10.400
The question I have is how the diffusion works within what we call G&A within
535
00:29:10.440 --> 00:29:14.620
companies, or sorry, legal, finance, compliance,
536
00:29:14.840 --> 00:29:17.260
all these kinds of roles. And in particular,
537
00:29:17.940 --> 00:29:22.720
there's like how you get all the automation there where we run a reforecast
538
00:29:22.740 --> 00:29:25.500
process within Stripe and we're not feeding it all into a coding agent,
539
00:29:25.560 --> 00:29:26.393
but maybe we should.
540
00:29:26.700 --> 00:29:30.200
But then also the ergonomics are wrong where your finance data is in a
541
00:29:30.300 --> 00:29:34.100
spreadsheet and like the models make up numbers and maybe you prompt them to not
542
00:29:34.120 --> 00:29:35.460
make up numbers and they're a little bit better.
543
00:29:36.060 --> 00:29:38.260
And you're not going to be able to verify those outputs.
544
00:29:38.520 --> 00:29:39.120
Yeah, yeah.
545
00:29:39.120 --> 00:29:41.460
It's not going to be like an RL problem where you're going to run the budget a
546
00:29:41.500 --> 00:29:42.440
million times and get the-.
547
00:29:42.620 --> 00:29:46.880
So how do we get a much more AI-native G&A function at Stripe? Make no mistakes.
548
00:29:50.420 --> 00:29:55.340
Not sure your SOX auditor is going to sign up for that, but maybe they will.
549
00:29:57.260 --> 00:29:58.093
It's a good question.
550
00:29:58.240 --> 00:30:01.380
And I think you get bottle-necked on verification very quickly.
551
00:30:01.660 --> 00:30:05.460
And then verification in situations where as people are finding out when they
552
00:30:05.480 --> 00:30:09.080
have to verify parts of code that you can't easily unit test,
553
00:30:09.180 --> 00:30:11.540
it takes a lot of time if you don't have the context.
554
00:30:12.000 --> 00:30:13.820
But people say that, but it's interesting.
555
00:30:13.880 --> 00:30:17.840
And people talk a lot about this phenomenon of the models do best on things
556
00:30:17.840 --> 00:30:19.120
where there's a good RL environment.
557
00:30:19.200 --> 00:30:22.380
And so we have good RL environments for coding and therefore they're really good
558
00:30:22.400 --> 00:30:25.600
at coding. I don't see why you can't have a good RL environment for finance.
559
00:30:25.840 --> 00:30:27.650
It's a very closed loop task.
560
00:30:27.680 --> 00:30:29.060
I think you can. Yeah. I mean-.
561
00:30:29.120 --> 00:30:31.020
Have they just forgotten to? Should I?
562
00:30:31.540 --> 00:30:33.420
Yeah, no, it's just hard to make.
563
00:30:33.600 --> 00:30:37.380
And so you have to work really hard on it and have good people do it. I mean,
564
00:30:37.580 --> 00:30:38.220
the models just-.
565
00:30:38.220 --> 00:30:40.180
Could Meta AI be the first quant? Could you guys-.
566
00:30:40.540 --> 00:30:44.970
That's what we're here to announce, actually. It's complete strategy. Yeah.
567
00:30:46.040 --> 00:30:49.940
Huge lift. Yeah. I'd better text somebody. Anyway.
568
00:30:50.620 --> 00:30:53.200
You can see what John really wants from the singularity. Yeah.
569
00:30:54.100 --> 00:30:56.580
I just want an AI that can do numbers.
570
00:30:56.780 --> 00:31:00.480
I think, I mean, the truth is, and this is actually in some ways bullish,
571
00:31:00.680 --> 00:31:02.900
some ways bearish, but yeah, I mean,
572
00:31:02.960 --> 00:31:05.760
the models are what they eat and if you can feed them very good data,
573
00:31:06.020 --> 00:31:08.960
you get very good capabilities. And so then it's the question of like,
574
00:31:09.160 --> 00:31:13.240
how hard is it to construct that data set and can the models help you construct
575
00:31:13.300 --> 00:31:17.420
a data set that leads to a better model or do you need to do a lot of human work
576
00:31:17.480 --> 00:31:20.180
and that sort of thing? And we're still sort of filling in the map.
577
00:31:20.500 --> 00:31:24.200
There's still fog of war all over the map where it hasn't been covered by data
578
00:31:24.280 --> 00:31:27.540
sets very well yet. And some of them are easier to create.
579
00:31:27.620 --> 00:31:31.980
And I do think it's true to some extent that math or software are pretty easy,
580
00:31:32.040 --> 00:31:35.720
but they're also like the things that the people creating the AI know how to do
581
00:31:36.100 --> 00:31:37.800
already.
And so yeah,
582
00:31:38.100 --> 00:31:40.340
I think all those things are doable and they will all happen.
583
00:31:40.580 --> 00:31:44.420
And each of them will get easier as you sort of surround that part of the map
584
00:31:46.220 --> 00:31:50.220
with other capabilities that are built into the model. It's like usually data.
585
00:31:50.760 --> 00:31:51.593
It's usually data.
586
00:31:52.220 --> 00:31:55.880
Nat, in my introduction, I mentioned GitHub Copilot, which launched in 2020,
587
00:31:56.000 --> 00:31:56.320
right?
588
00:31:56.320 --> 00:32:00.520
It was actually '20. So yeah, I think it was 2020. It was.
589
00:32:00.880 --> 00:32:01.640
Yeah.
590
00:32:01.640 --> 00:32:02.540
Not sure actually.
591
00:32:02.640 --> 00:32:03.480
Your tenure at GitHub
592
00:32:05.500 --> 00:32:09.720
was widely renowned for its success and-.
593
00:32:09.780 --> 00:32:12.500
Well, I apparently needed to work on the scalability some more.
594
00:32:13.200 --> 00:32:15.000
I think I might have missed a step there.
595
00:32:16.440 --> 00:32:19.500
Well, when you were there, things worked really great.
596
00:32:20.260 --> 00:32:23.680
And I guess it was 2018 to 2020 or thereabouts and it was a-.
597
00:32:23.780 --> 00:32:24.800
Yeah, '21. Yeah.
598
00:32:24.960 --> 00:32:29.640
Okay. Yeah. Just how did you do it?
599
00:32:30.480 --> 00:32:31.313
How does one... I mean,
600
00:32:31.360 --> 00:32:34.460
I don't want to call it a turnaround situation because GitHub was already doing
601
00:32:34.480 --> 00:32:37.600
well, but there were lots of changes to make and things that, in fact,
602
00:32:38.120 --> 00:32:42.640
you did change. What are the tricks in a couple of minutes?
603
00:32:43.700 --> 00:32:47.300
Yeah. I don't know. I mean, I don't know that there's any tricks. I mean,
604
00:32:47.440 --> 00:32:49.660
I think it's like, show up, that's step one.
605
00:32:51.180 --> 00:32:55.140
Really try to be a user and talk to the users and really understand what their
606
00:32:55.200 --> 00:32:58.660
life is like. And when I was at Microsoft,
607
00:32:58.720 --> 00:33:03.380
we bought GitHub and then we had to go through antitrust
608
00:33:03.440 --> 00:33:05.900
compliance in the EU and in Washington, DC.
609
00:33:06.140 --> 00:33:08.820
And so there was a period after we'd signed a deal to buy it,
610
00:33:08.920 --> 00:33:12.660
but before I could start running it where I couldn't do anything with the
611
00:33:12.680 --> 00:33:14.960
company, but I could go and talk to all the customers.
612
00:33:15.200 --> 00:33:18.800
And so I'd spent a couple months just talking to all the customers and users and
613
00:33:18.820 --> 00:33:21.920
my friends who were GitHub users to see what their life was like using GitHub
614
00:33:21.980 --> 00:33:26.380
and what-and so I kind of went in thinking I had some pretty strong views of
615
00:33:26.420 --> 00:33:28.760
what was missing.
I mean, it was really clear that GitHub had not...
616
00:33:29.160 --> 00:33:32.500
It had treated itself as a hub where you store your source code and had pull
617
00:33:32.520 --> 00:33:33.353
requests and issues.
618
00:33:33.580 --> 00:33:36.580
And there was lots of parts of the software development life cycle that GitHub
619
00:33:36.620 --> 00:33:38.720
hadn't worked on, like CI/CD,
620
00:33:39.360 --> 00:33:43.580
which had not actually been a part of standard software SDLC when GitHub was
621
00:33:43.620 --> 00:33:47.280
created and a few other things. And so I was definitely...
622
00:33:48.120 --> 00:33:49.720
The clear message I got from the users was like,
623
00:33:49.780 --> 00:33:52.600
we want more stuff directly integrated in GitHub. And so I thought, okay,
624
00:33:52.800 --> 00:33:53.960
we should just go in and do a bunch of stuff.
625
00:33:54.360 --> 00:33:58.540
And then I got to GitHub and I found that the company had
626
00:33:59.340 --> 00:34:04.320
some kind of stage fright because it was such a beloved product that was so
627
00:34:04.640 --> 00:34:06.440
well designed and
628
00:34:08.060 --> 00:34:10.600
many of the people who had originally created GitHub were gone.
629
00:34:10.600 --> 00:34:15.080
And so the inheritors of it were worried about desecrating the legacy and
630
00:34:15.520 --> 00:34:16.920
were a little bit nervous to ship anything.
631
00:34:16.980 --> 00:34:19.960
It had to be kind of perfect when it shipped. And so it was like, okay,
632
00:34:20.040 --> 00:34:21.180
break the stage fright.
633
00:34:21.600 --> 00:34:24.480
We're going to just throw a lot of pots and hopefully we get good at this
634
00:34:24.540 --> 00:34:26.860
eventually. And then the other thing I did was I-.
635
00:34:27.060 --> 00:34:28.940
So talk to users, ship stuff.
636
00:34:29.340 --> 00:34:32.980
Ship stuff. Yeah. I mean, the main thing is always about learning.
637
00:34:33.400 --> 00:34:37.220
And so how quickly can you figure out if your idea is any good and how to change
638
00:34:37.240 --> 00:34:40.320
it to be better? And so yeah, it's the cycle time.
639
00:34:40.520 --> 00:34:45.060
If you can insert a temperature probe into a product team or engineering team
640
00:34:45.520 --> 00:34:49.880
and only get one number out and determine whether it's healthy or not,
641
00:34:49.940 --> 00:34:53.980
I think the number you want is how long it takes to go from an idea to something
642
00:34:54.000 --> 00:34:54.900
that's shipped to users,
643
00:34:55.020 --> 00:34:59.580
to like observing the feedback from how they do or don't use it,
644
00:34:59.960 --> 00:35:03.160
to like having an improved idea. And the faster that is,
645
00:35:03.220 --> 00:35:04.880
the faster you can learn. Now, of course it helps-.
646
00:35:04.940 --> 00:35:06.520
What's a good target for the duration of that loop?
647
00:35:07.500 --> 00:35:11.780
Well, I mean, for an early-stage product where you're really not sure,
648
00:35:14.320 --> 00:35:18.460
it's really nice if you can do that in one day, which is true sometimes.
649
00:35:19.640 --> 00:35:21.580
I mean, Stripe was famously amazing at this.
650
00:35:22.000 --> 00:35:25.500
You and John would sit down with people who were installing Stripe in their
651
00:35:25.560 --> 00:35:29.900
business and immediately learn what the problems were and fix them.
652
00:35:29.960 --> 00:35:30.420
Now that's sort of-.
653
00:35:30.420 --> 00:35:31.540
Stripe was tiny at that point.
654
00:35:31.600 --> 00:35:34.660
Can you do something like that for an org-GitHub was already at enormous
655
00:35:34.700 --> 00:35:37.780
sprawling scale. Can you get that loop down to mere days?
656
00:35:37.860 --> 00:35:40.740
I think so. Yeah. I mean, there's things that should be slow moving,
657
00:35:41.500 --> 00:35:45.660
maybe like your database, although maybe that should have been faster moving.
658
00:35:46.660 --> 00:35:48.280
And then there's things that need to be fast moving,
659
00:35:48.540 --> 00:35:51.400
which is like you're trying to figure out what...
660
00:35:51.580 --> 00:35:53.980
You're always solving for the intersection of two sets,
661
00:35:54.160 --> 00:35:58.700
which is what is something we can build that doesn't exist that will work
662
00:35:59.420 --> 00:36:03.800
and what is something that people really want to use every day that they don't
663
00:36:03.860 --> 00:36:06.460
know that they want to use? And you have all these unknowns going into it.
664
00:36:06.520 --> 00:36:09.620
You have your own hypotheses, your intuitions based on your own usage.
665
00:36:10.240 --> 00:36:14.960
And so you start with those and then you sort of iterate and loop and figure it
666
00:36:14.980 --> 00:36:19.060
out. And so tightening that loop is like, I think very, very important. And yes,
667
00:36:19.140 --> 00:36:20.160
you can do it in big companies. I mean,
668
00:36:20.400 --> 00:36:25.060
we're definitely doing it right now at Meta. We ship-.
669
00:36:25.420 --> 00:36:26.253
Is it a culture change?
670
00:36:26.680 --> 00:36:27.760
It's a huge culture change.
671
00:36:28.320 --> 00:36:30.340
Okay. There's something interesting here where
672
00:36:31.860 --> 00:36:36.680
I feel like everyone in this audience has probably heard those notions before,
673
00:36:37.200 --> 00:36:40.680
of you want to be incredibly close to users. Again,
674
00:36:40.740 --> 00:36:44.780
we start our leadership meeting every Monday morning at Stripe with bringing a
675
00:36:44.880 --> 00:36:45.260
user on.
676
00:36:45.260 --> 00:36:48.160
And it's super useful because it gets you out of these galaxy brain product
677
00:36:48.260 --> 00:36:51.260
idea, maybe Stripe Dashboard should be a BI hub. And instead,
678
00:36:51.460 --> 00:36:53.260
people give you this incredibly concrete feedback of,
679
00:36:53.340 --> 00:36:54.760
"I need you to fix the bug" or,
680
00:36:54.800 --> 00:36:57.580
"this number is wrong." And so it's very centering.
681
00:36:58.020 --> 00:37:00.420
And then the fast clock speed and the iteration speed,
682
00:37:00.560 --> 00:37:04.700
and in particular the demos, not memos, actually getting down to the code.
683
00:37:05.120 --> 00:37:08.930
And yet I think we'd find a lot of variability in
684
00:37:09.930 --> 00:37:10.970
these practices.
685
00:37:11.450 --> 00:37:14.770
And so it's a little bit like you should eat more protein or you should go to
686
00:37:14.790 --> 00:37:15.950
bed at a consistent time.
687
00:37:16.010 --> 00:37:19.630
It has to come from the top. This is my experience. So I mean,
688
00:37:19.730 --> 00:37:21.130
there can be rare exceptions,
689
00:37:21.630 --> 00:37:25.690
but the inertial forces are so strong in any organization.
690
00:37:25.790 --> 00:37:28.800
So organizations want to be mediocre on these axes, that's the entropic force.
691
00:37:28.880 --> 00:37:32.800
Organizations entropically decay to the point where they're situated at the
692
00:37:32.920 --> 00:37:37.560
atomic level to prevent progress. And it's not anyone's fault.
693
00:37:37.920 --> 00:37:40.560
It's just an emergent phenomenon of local incentives.
694
00:37:41.000 --> 00:37:44.160
And often it's correct because you have something that's working and it's
695
00:37:44.160 --> 00:37:46.440
working for a lot of people and at scale and you don't want to break it.
696
00:37:46.840 --> 00:37:47.673
But yeah,
697
00:37:48.920 --> 00:37:52.200
if you are a leader in an organization and you want this to happen,
698
00:37:52.400 --> 00:37:53.233
it has to come from you.
699
00:37:53.320 --> 00:37:56.000
You have to drive the energy and you have to find what's the binding constraint
700
00:37:56.040 --> 00:38:00.680
or the limiting factor or the slow part of this process and make sure that the
701
00:38:00.720 --> 00:38:02.280
right things are happening to speed it up.
702
00:38:02.760 --> 00:38:03.720
How do you think about... I mean,
703
00:38:03.800 --> 00:38:06.640
one thing that you've done in the teams that you've run is sort of you have your
704
00:38:06.840 --> 00:38:11.200
direct staff and then you always have this broader crew of like Avenger-.
705
00:38:12.040 --> 00:38:15.960
I don't pay any attention to the org chart. So my org chart is like,
706
00:38:16.200 --> 00:38:20.320
there's that meme of the conspiracy guy with push pins and the string-.
707
00:38:21.230 --> 00:38:22.080
Pepe Silvia, yeah, yeah.
708
00:38:22.280 --> 00:38:23.960
At the cork board, I don't know what it's from,
709
00:38:24.440 --> 00:38:27.240
but that's what my org chart looks like whenever I run a team.
710
00:38:27.400 --> 00:38:28.233
It makes no sense.
711
00:38:28.400 --> 00:38:31.920
So I always just try to work with the people who are doing the work and it's
712
00:38:32.080 --> 00:38:36.720
extremely confusing and probably toxic in some ways, but that's the only way.
713
00:38:36.960 --> 00:38:37.720
And then that-.
714
00:38:37.720 --> 00:38:39.880
And in particular talking to the doers at the coalface.
715
00:38:40.040 --> 00:38:43.920
Yeah, yeah. "Coalface" is a good word. Yeah, exactly. Yeah.
716
00:38:44.080 --> 00:38:46.360
You want to get in there and understand what's actually happening.
717
00:38:46.440 --> 00:38:48.480
And I don't do this perfectly, but this is what I try to do.
718
00:38:48.840 --> 00:38:52.120
And then the other thing is tools. And so yeah, Meta,
719
00:38:52.280 --> 00:38:56.600
one of the things that we've done in the first few months is change what tools
720
00:38:56.680 --> 00:38:59.680
people are using because tools drive culture a lot.
721
00:39:00.200 --> 00:39:02.920
And so if the tool makes an easy thing hard,
722
00:39:03.440 --> 00:39:07.560
the organization completely reorients itself around that thing being hard.
723
00:39:09.200 --> 00:39:13.960
We had a tool that we used for collecting labels for training AI
724
00:39:14.080 --> 00:39:14.640
models.
725
00:39:14.640 --> 00:39:17.920
That tool was extremely cumbersome and had lots of approvals and stuff like
726
00:39:18.000 --> 00:39:19.560
that. And as a result of that,
727
00:39:20.000 --> 00:39:23.960
it was very expensive to fire up a new labeling task. And as a result of that,
728
00:39:24.320 --> 00:39:26.480
people would design their labeling tasks differently.
729
00:39:26.880 --> 00:39:31.200
They would bundle all kinds of tasks into a single task and they would run it
730
00:39:31.240 --> 00:39:32.800
less often and they would learn from it. And so,
731
00:39:33.080 --> 00:39:36.280
but if your tool makes it really cheap and easy and any IC can do it and it's
732
00:39:36.320 --> 00:39:39.200
permissionless, then you're firing off all these things.
733
00:39:39.320 --> 00:39:41.920
So I think it's one of the binding constraints is often like, oh,
734
00:39:42.000 --> 00:39:45.080
this tool makes this thing really hard and it's just like the activation energy
735
00:39:45.160 --> 00:39:46.320
to overcome that's too high.
736
00:39:46.520 --> 00:39:48.080
And the fact that you were, I think,
737
00:39:48.840 --> 00:39:52.000
very impatient with what you felt like would be a-.
738
00:39:52.760 --> 00:39:54.720
Yeah, I think it's like you have to be impatient. Yeah.
739
00:39:55.120 --> 00:39:59.240
Things can always go a little bit faster and people allow,
740
00:39:59.520 --> 00:40:00.640
if you're in an organization-.
741
00:40:00.840 --> 00:40:03.000
Talk to users, ship, be impatient, ignore the org chart.
742
00:40:03.080 --> 00:40:04.520
I'm trying to extract the Nat framework.
743
00:40:04.840 --> 00:40:06.320
Yeah, I think that's right. There's just one more thing,
744
00:40:06.360 --> 00:40:07.480
which is something about dignity,
745
00:40:07.680 --> 00:40:11.920
which is like employees and companies allow the company to
746
00:40:13.560 --> 00:40:18.120
impede or to impinge. I knew I was going to get it on their dignity. And
747
00:40:19.970 --> 00:40:23.630
they just allow the company to treat them in all these undignified ways where
748
00:40:24.450 --> 00:40:27.550
you're like a sacred human being and you should be able to just do things.
749
00:40:28.110 --> 00:40:32.950
And so you have to restore a sense of self-worth and dignity to the strongest
750
00:40:33.030 --> 00:40:34.910
engineers that they should be able to do.
751
00:40:35.210 --> 00:40:39.110
They shouldn't have to schedule 10 meetings and write 10 documents to make a
752
00:40:39.190 --> 00:40:43.270
change that happens to cut across 3 layers of the stack in order to get
753
00:40:43.310 --> 00:40:47.470
something done. And they have to feel like sort of superheroes a little bit.
754
00:40:47.950 --> 00:40:50.770
And so I do think that feeling is something you're going for too.
755
00:40:50.850 --> 00:40:54.970
And you're talking about the indignity of being hemmed in by process.
756
00:40:55.030 --> 00:40:58.590
In a veal pen, this is your box. Yeah.
757
00:40:58.870 --> 00:41:03.070
And this happens often when teams get too big because coordinating across a lot
758
00:41:03.090 --> 00:41:04.630
of people is an N-squared problem.
759
00:41:05.170 --> 00:41:08.110
And so in order to make the coordination possible so you're not stepping on each
760
00:41:08.130 --> 00:41:10.690
other's toes a lot, your project gets bigger,
761
00:41:10.850 --> 00:41:11.990
you add a lot of people because you're like,
762
00:41:12.050 --> 00:41:16.970
"We need more people." And then you chop up the actual work into pieces so that
763
00:41:17.070 --> 00:41:20.570
each piece can be run by a smaller team that can actually coordinate with each
764
00:41:20.610 --> 00:41:21.250
other.
765
00:41:21.250 --> 00:41:26.250
And then suddenly your team architecture and your software architecture sprawl.
766
00:41:26.890 --> 00:41:30.710
And now you can't de-sprawl your software because it would mean de- sprawling
767
00:41:30.750 --> 00:41:32.890
your team, which is impossible because you can't do that.
768
00:41:33.450 --> 00:41:37.680
And so this is like the other problem is you need...
769
00:41:38.130 --> 00:41:42.190
Often the things that are really impactful involve a change to five different
770
00:41:42.210 --> 00:41:43.043
components,
771
00:41:43.250 --> 00:41:46.330
and you need to make sure that an engineer can actually just go change all those
772
00:41:46.350 --> 00:41:48.350
components, or maybe there shouldn't be five,
773
00:41:48.430 --> 00:41:50.150
maybe it should be two or something like this.
774
00:41:50.310 --> 00:41:54.210
Okay. Can I try something on for size? I feel like in Silicon Valley,
775
00:41:54.730 --> 00:41:57.190
there's maybe, as companies get bigger,
776
00:41:57.730 --> 00:42:00.590
there's a desire to make things scalable,
777
00:42:01.250 --> 00:42:05.010
whereas actually maybe part of what you're talking about here is just leaning
778
00:42:05.050 --> 00:42:07.550
into the lack of scalability a bit. Like Daniel,
779
00:42:07.610 --> 00:42:08.670
you spent a lot of time at Apple,
780
00:42:08.930 --> 00:42:12.330
which is just kind of a deeply unscalable company in how it runs.
781
00:42:12.790 --> 00:42:14.470
How do you decide what goes into an iPhone release?
782
00:42:14.750 --> 00:42:18.450
Craig gathers everyone in auditorium and like Craig reviews every single thing
783
00:42:18.510 --> 00:42:19.343
line by line.
784
00:42:19.590 --> 00:42:23.230
And there's something about not trying to make things too scalable, which is-.
785
00:42:24.470 --> 00:42:27.390
The cost of that is I think less relevant now.
786
00:42:27.490 --> 00:42:30.490
So the cost of that used to be that teams would duplicate efforts.
787
00:42:30.630 --> 00:42:33.730
And so there would be 5 logging frameworks, 1 that the Maps team would have,
788
00:42:33.730 --> 00:42:34.730
1 that the Watch team would have,
789
00:42:34.790 --> 00:42:37.800
because you just wouldn't know because there's like 13,
790
00:42:38.080 --> 00:42:42.230
14 people at Apple that have the full picture. And so, I mean,
791
00:42:43.230 --> 00:42:45.510
and that used to be a huge issue because you might say, "Well,
792
00:42:45.730 --> 00:42:49.150
we're spending all this." Now that the cost of software production is like
793
00:42:49.230 --> 00:42:50.063
collapsing,
794
00:42:50.150 --> 00:42:55.070
I do think companies should look on the inside more like Silicon Valley and that
795
00:42:55.090 --> 00:42:57.170
you have a bunch of different pods or companies.
796
00:42:57.430 --> 00:43:00.810
The inter-team contact probably needs to be less frequent and they should be
797
00:43:00.850 --> 00:43:04.210
able to get done much more on their own because they can produce much more and
798
00:43:04.230 --> 00:43:04.590
less time.
799
00:43:04.590 --> 00:43:04.750
Okay,
800
00:43:04.750 --> 00:43:07.110
we're going to have to really speed up because we only have 15 minutes left,
801
00:43:07.210 --> 00:43:08.510
so we have a lot to get through here.
802
00:43:08.590 --> 00:43:10.990
So I'm going to give you three topics you can just choose one to talk about.
803
00:43:11.730 --> 00:43:16.110
Yeah. So idolatry data center aesthetics or open source models?
804
00:43:16.530 --> 00:43:17.363
Well, so
805
00:43:20.490 --> 00:43:21.890
this wasn't in the briefing doc.
806
00:43:25.320 --> 00:43:30.170
I think an interesting question, so Works in Progress, I believe,
807
00:43:30.250 --> 00:43:31.890
is a publication produced by Stripe.
808
00:43:32.550 --> 00:43:36.350
And I think one of the things that it's been very focused on is the meaning of
809
00:43:36.390 --> 00:43:38.290
beauty. And
810
00:43:41.730 --> 00:43:46.410
should we take the view that that should only apply to sort of human-scale
811
00:43:46.470 --> 00:43:47.390
buildings, cities,
812
00:43:47.770 --> 00:43:52.530
places that we think we'll visit or to the industrial buildings
813
00:43:53.410 --> 00:43:56.290
that we are building? And we are spending right now as a country,
814
00:43:56.690 --> 00:43:59.190
earlier I gave you the global number, but as a country, we're, I think,
815
00:43:59.250 --> 00:44:02.610
going to be north of 2% of US GDP on AI CapEx.
816
00:44:02.690 --> 00:44:05.730
And a lot of that CapEx is building things in the physical world.
817
00:44:06.530 --> 00:44:10.910
And we don't really think of beauty when we think of these buildings.
818
00:44:10.970 --> 00:44:15.490
We think a lot about form and function and these buildings, these data centers,
819
00:44:15.850 --> 00:44:19.790
predominantly they take in a bunch of energy and then hopefully they produce
820
00:44:19.830 --> 00:44:23.190
tokens that are of economic value and use to people.
But
821
00:44:25.570 --> 00:44:28.870
is that the correct way of building them? Not,
822
00:44:28.990 --> 00:44:33.310
if strictly optimizing on what is the best ROIC for the dollar,
823
00:44:33.730 --> 00:44:36.530
but what is best for the human soul and for the civilization,
824
00:44:36.710 --> 00:44:40.470
or should we be doing better? Now there are structures around the world,
825
00:44:40.910 --> 00:44:43.330
like in the Nordic countries, it's quite interesting.
826
00:44:44.450 --> 00:44:48.130
They'll have a power plant that has a ski slope built into it. Now,
827
00:44:48.190 --> 00:44:50.250
I don't know that that's particularly beautiful, but it's fun.
828
00:44:51.730 --> 00:44:53.730
And I don't know that we've saturated the amount of fun.
829
00:44:53.950 --> 00:44:57.090
I feel like we'd be breaking a law here if Patrick didn't just quickly interject
830
00:44:57.110 --> 00:44:58.670
with the Victorian pumping stations. Yeah.
831
00:44:59.210 --> 00:45:00.670
Just Google Victorian pumping station.
832
00:45:01.830 --> 00:45:04.230
You guys need like the Joe Rogan guy that can put it up behind them.
833
00:45:04.650 --> 00:45:06.210
Well, okay. I guess my question is,
834
00:45:06.750 --> 00:45:10.590
are you thinking about this question and aesthetics and beauty and data centers
835
00:45:10.610 --> 00:45:13.470
and so forth because of the brewing political opposition?
836
00:45:13.530 --> 00:45:16.310
And maybe if these things are more attractive,
837
00:45:16.370 --> 00:45:20.210
then people will be more accepting of the idea of having one in their locale,
838
00:45:20.750 --> 00:45:22.970
or is there something even deeper here?
839
00:45:23.330 --> 00:45:26.130
I think everyone working on AI,
840
00:45:26.190 --> 00:45:29.870
including Meta will have to earn the right to build these data centers on the
841
00:45:29.950 --> 00:45:33.510
economic merits that'll be helpful for the people in the towns that they are
842
00:45:33.530 --> 00:45:35.810
being built. So I don't think beauty will fix that problem.
843
00:45:36.970 --> 00:45:38.890
I think it is a deeper question of,
844
00:45:40.570 --> 00:45:42.650
beyond the numbers and the numerics,
845
00:45:42.750 --> 00:45:47.030
like are we improving how people feel about the world?
846
00:45:47.290 --> 00:45:50.590
And we spend so much money constructing these things,
847
00:45:50.650 --> 00:45:55.230
the incremental spend on making it pretty. I think there's obviously,
848
00:45:55.930 --> 00:45:56.750
if you speak to an architect,
849
00:45:56.750 --> 00:45:59.450
they will come up with a design that actually is a hundred times more expensive.
850
00:45:59.560 --> 00:46:03.330
So in theory, you could make it dramatically more expensive,
851
00:46:03.390 --> 00:46:05.450
but I think without a lot of more incremental spend,
852
00:46:06.070 --> 00:46:07.330
you can make it pleasing to the eye.
853
00:46:07.470 --> 00:46:11.330
And I think that's just the right thing to do regardless of all the politics,
854
00:46:11.390 --> 00:46:12.790
which I think will have to be solved too.
855
00:46:12.970 --> 00:46:16.850
Can this mindset be applied to the model itself? And if so, what does that mean?
856
00:46:16.910 --> 00:46:19.610
Well, that would be a question I would ask you because Stripe was obviously very
857
00:46:19.670 --> 00:46:23.090
famous for caring about beauty and design before it made sense to. I mean,
858
00:46:23.190 --> 00:46:27.250
developer open and closed source developer projects,
859
00:46:27.610 --> 00:46:28.510
they just kind of work. Now,
860
00:46:28.590 --> 00:46:33.170
Stripe famously cared so much about beauty that I remember using it in
861
00:46:33.470 --> 00:46:34.303
2009 or '10,
862
00:46:34.510 --> 00:46:38.690
and you only supported one browser because you did not want to go through the
863
00:46:38.790 --> 00:46:43.450
strings and arrows of effort in order to make it compatible and beautiful for
864
00:46:43.530 --> 00:46:44.363
other browsers.
865
00:46:44.850 --> 00:46:48.570
So there's a caring about beauty in a category that no one has cared about it
866
00:46:48.590 --> 00:46:51.530
before that I think applies to Stripe and developer products and data centers.
867
00:46:51.590 --> 00:46:53.710
Now, you're asking the question about language models,
868
00:46:54.270 --> 00:46:56.250
and I don't know that we know the answer to that.
869
00:46:56.430 --> 00:47:00.090
And my question would be what advice you would have for us and other labs that
870
00:47:00.110 --> 00:47:01.070
are thinking about this today,
871
00:47:01.130 --> 00:47:05.690
because everyone is very focused in our industry about things that you can
872
00:47:05.730 --> 00:47:09.110
measure.
So we have evals. Are you familiar with the evals?
873
00:47:09.190 --> 00:47:11.050
So these are numbers and you optimize the numbers,
874
00:47:11.510 --> 00:47:15.310
but part of what's going on with beauty is it's very hard to quantify a soul,
875
00:47:16.070 --> 00:47:18.870
and it's very hard to quantify the feeling that you feel when you see something
876
00:47:18.890 --> 00:47:23.130
beautiful. And I think that actually is also,
877
00:47:23.770 --> 00:47:26.750
there's a whole different story there of leaving the world of data-driven
878
00:47:26.790 --> 00:47:27.623
design.
879
00:47:27.790 --> 00:47:32.670
And it's not clear to me that we've fully saturated what that philosophy
880
00:47:32.750 --> 00:47:36.450
means. So we, and I'm speaking now on behalf of my industry, if I may,
881
00:47:36.550 --> 00:47:40.510
would love to learn from Stripe on how we could be making the models more
882
00:47:40.530 --> 00:47:44.050
beautiful. Yeah, that's to them, not to me.
883
00:47:48.050 --> 00:47:48.970
So I turn it back to you, Patrick.
884
00:47:51.390 --> 00:47:55.160
Beyond my pay grade. Well, I mean, I think there's...
885
00:47:58.550 --> 00:47:59.350
It is interesting to me.
886
00:47:59.350 --> 00:48:04.330
It does feel like there's a brewing vibe shift in the technology
887
00:48:04.450 --> 00:48:07.650
sector and has been over the last couple of years. I mean,
888
00:48:07.710 --> 00:48:08.543
I don't know if you guys agree,
889
00:48:08.570 --> 00:48:11.030
maybe just from our little parochial perch or something,
890
00:48:11.330 --> 00:48:13.470
but where to your point, for so long,
891
00:48:13.710 --> 00:48:18.530
we've been focused on and oriented towards empirics
892
00:48:18.990 --> 00:48:23.650
and metrics and quantification and maximization and so
893
00:48:23.790 --> 00:48:26.650
forth. And at some point there does... I mean,
894
00:48:28.430 --> 00:48:32.270
in science, every experiment is inexorably, inevitably,
895
00:48:32.450 --> 00:48:35.450
theory-laden in that every time you measure something,
896
00:48:36.070 --> 00:48:38.110
you implicitly have a theory about what you should be measuring.
897
00:48:38.450 --> 00:48:40.990
And I feel like in the same way, we're maybe coming to realize, well,
898
00:48:42.330 --> 00:48:45.330
what are our metrics? What are we maximizing?
899
00:48:45.430 --> 00:48:48.910
Why are we maximizing these things? Why not some set of other things?
900
00:48:48.910 --> 00:48:53.210
And as our collective potency grows with AI and with everything else and with
901
00:48:53.230 --> 00:48:56.890
this kind of thundering cavalcade of new inventions, yeah,
902
00:48:56.950 --> 00:48:58.950
there is this question of for what,
903
00:48:59.930 --> 00:49:04.870
how does it elevate and glorify mankind and how do
904
00:49:04.890 --> 00:49:09.230
we be good stewards? And I think this is increasingly, again,
905
00:49:09.450 --> 00:49:10.930
a sector-wide question.
906
00:49:12.690 --> 00:49:17.630
Yeah. I think it's pretty interesting actually that as society has
907
00:49:17.710 --> 00:49:22.070
secularized discussions of these sort of higher aspirations
908
00:49:23.490 --> 00:49:27.350
or registers of the human spirit have declined.
909
00:49:28.030 --> 00:49:32.910
Beauty shifted from a kind of public good that the
910
00:49:33.070 --> 00:49:37.830
citizens of ancient Athens would tax the provinces to build the Parthenon
911
00:49:38.310 --> 00:49:42.850
and it was like the purpose of a state was potentially to make that
912
00:49:43.130 --> 00:49:44.770
public space in some ways,
913
00:49:45.550 --> 00:49:48.670
like beauty started to shift to consumer goods.
914
00:49:49.210 --> 00:49:52.270
And so maybe you didn't get it at this sort of public commons level,
915
00:49:52.390 --> 00:49:55.830
but you got it in hopefully a well-designed object that you could buy.
916
00:49:57.730 --> 00:50:01.870
And so there's a way in which it maybe is still there,
917
00:50:01.950 --> 00:50:05.330
but it maybe also really fell off. Things are very functional now. I mean,
918
00:50:05.390 --> 00:50:08.330
we spent most of this conversation talking about how to automate things and make
919
00:50:08.350 --> 00:50:09.183
them faster,
920
00:50:10.610 --> 00:50:15.490
but AI is kind of like causing us all to
921
00:50:15.830 --> 00:50:20.070
have these discussions about what kind of life we want to live and who we want
922
00:50:20.110 --> 00:50:24.250
to be and what we want society to be and what our values actually are.
923
00:50:24.690 --> 00:50:28.650
And that's pretty wild. I don't remember... I mean,
924
00:50:28.730 --> 00:50:31.230
I guess there were some of this in the rise of the internet,
925
00:50:31.750 --> 00:50:36.090
there was sort of the Arab Spring moment and it was about free speech and
926
00:50:36.330 --> 00:50:37.790
democracy and all that.
927
00:50:37.960 --> 00:50:40.530
And before the John Perry Barlow in the '90s-.
928
00:50:40.680 --> 00:50:42.770
Yeah, the EFF kind of energy,
929
00:50:43.650 --> 00:50:48.410
but some of that was kind of reheated radicalism or
930
00:50:48.690 --> 00:50:52.470
it was like the same ideas that were already in the air, but like, okay,
931
00:50:52.530 --> 00:50:55.550
this new technology will clearly be a vector for the things that we already are
932
00:50:55.570 --> 00:50:57.890
saying we want, but there's a way in which AI is,
933
00:50:58.410 --> 00:50:59.890
and maybe it's just like the times also,
934
00:51:00.170 --> 00:51:05.110
it's causing us to want to ask the questions more deeply and have
935
00:51:05.130 --> 00:51:06.670
the debates. I think that's really good.
936
00:51:09.390 --> 00:51:10.750
You've invested in how many startups?
937
00:51:12.170 --> 00:51:13.003
Hundred plus.
938
00:51:14.390 --> 00:51:15.580
Is now a good time to start a startup?
939
00:51:17.480 --> 00:51:21.660
I think so. Obviously the question behind the question is,
940
00:51:22.080 --> 00:51:25.090
does it make sense to start companies and is there one unique company in the
941
00:51:25.110 --> 00:51:29.500
future and all sorts of sort of dark dystopian thoughts. My view is,
942
00:51:32.040 --> 00:51:36.580
I think at least for the time horizon you can sort of productively forecast on,
943
00:51:37.240 --> 00:51:41.960
it is probably a very good thing to give people who self-select into not
944
00:51:42.200 --> 00:51:45.740
joining big companies because they feel like they don't fit in capital in order
945
00:51:45.760 --> 00:51:46.880
to do something interesting. Now,
946
00:51:46.940 --> 00:51:51.920
it is true that the best companies to start in 2015 are not the best
947
00:51:51.940 --> 00:51:53.080
companies to start in 2026,
948
00:51:53.680 --> 00:51:55.680
and that's probably not going to be the case in 2036.
949
00:51:56.090 --> 00:51:58.680
And so maybe things today look a little bit more applied,
950
00:51:58.740 --> 00:51:59.980
look a little bit more industrial,
951
00:52:00.040 --> 00:52:03.460
look a little bit more like a different kind of turbine or energy.
952
00:52:04.800 --> 00:52:07.300
I'm sure some categories of software will endure as well,
953
00:52:07.360 --> 00:52:11.520
but my guess is that's kind of an evergreen asset.
I do think the
954
00:52:12.380 --> 00:52:13.213
sort of...
955
00:52:14.200 --> 00:52:19.200
Obviously the market is telling us that the dynamics of SaaS are going to
956
00:52:19.240 --> 00:52:23.040
have to change just because those businesses were predicated on a certain high
957
00:52:23.400 --> 00:52:26.240
production cost, which has gone down. But my guess is there's...
958
00:52:27.340 --> 00:52:30.920
We as a large company trying to do things in the world are faced with a thousand
959
00:52:30.940 --> 00:52:34.220
different problems out of which we are going to solve maybe three internally,
960
00:52:34.660 --> 00:52:35.940
and the rest we are going to procure.
961
00:52:36.480 --> 00:52:39.920
And it's possible that we are procuring from much smaller companies in the
962
00:52:39.940 --> 00:52:40.660
future,
963
00:52:40.660 --> 00:52:45.660
but my guess is that remains a kind of an evergreen category to invest in,
964
00:52:45.760 --> 00:52:49.460
even if the things that those people go and do change over time.
965
00:52:49.920 --> 00:52:52.260
There were startups, by the way, before the semiconductor.
966
00:52:52.680 --> 00:52:53.680
It looked a little bit different,
967
00:52:53.740 --> 00:52:58.040
and it's probably better to project us going back in time,
968
00:52:58.560 --> 00:53:00.900
I think, than to project forward the past few years.
969
00:53:02.160 --> 00:53:02.993
Last question.
970
00:53:03.720 --> 00:53:07.880
We will be gathered back here again next year for Stripe Sessions.
971
00:53:08.380 --> 00:53:11.460
What are your guys' predictions for AI? Concretely,
972
00:53:11.780 --> 00:53:15.920
like recently it's been the story of RL and longer context windows and coding
973
00:53:15.960 --> 00:53:16.940
agents really starting to work.
974
00:53:17.900 --> 00:53:20.360
What's going to be different this time next year when we gather here?
975
00:53:23.960 --> 00:53:26.900
I think the thing that we talked about how coding is a well-covered domain,
976
00:53:26.960 --> 00:53:28.500
but the other domains are not as well covered.
977
00:53:29.020 --> 00:53:32.400
I think we'll have more examples of domains that are well covered by agentic
978
00:53:32.440 --> 00:53:33.273
capabilities,
979
00:53:34.940 --> 00:53:38.000
and that'll be because people did the work of building the RL tasks and
980
00:53:38.020 --> 00:53:40.120
environments and collecting the data.
981
00:53:40.940 --> 00:53:45.720
And then I think the other thing is computers will probably
982
00:53:45.800 --> 00:53:49.900
cost more. So if you want a computer next year,
983
00:53:50.040 --> 00:53:52.660
you should probably buy it now. That would be my advice.
984
00:53:53.460 --> 00:53:54.680
Which is very literally the case. Again,
985
00:53:54.780 --> 00:53:57.180
smartphone shipments are going down because we've priced the memory out of
986
00:53:57.220 --> 00:53:58.360
smartphones. It's all going towards data.
987
00:53:58.520 --> 00:54:00.320
Yeah. Buy next year's computer today.
988
00:54:01.980 --> 00:54:03.620
And as many Mac Minis as you think you'll need.
989
00:54:03.680 --> 00:54:05.530
Whatever it is, yeah. RAM, disk, anything.
990
00:54:06.260 --> 00:54:06.660
Daniel?
991
00:54:06.660 --> 00:54:06.840
Well,
992
00:54:06.840 --> 00:54:11.460
very good strategy in terms of how to answer that question I have learned is
993
00:54:11.560 --> 00:54:16.400
since he was one of the earliest users of Stripe and Figma and GPT is to
994
00:54:16.500 --> 00:54:19.400
look at whatever Nat's doing now and just project forward.
995
00:54:19.760 --> 00:54:23.300
So eBay is one potential answer to that question.
996
00:54:23.360 --> 00:54:25.100
It's a great website. There's a lot of stuff on eBay.
997
00:54:25.200 --> 00:54:27.560
It's anything you want. It's there. It's amazing.
998
00:54:28.440 --> 00:54:30.820
But more practically, I mean, I think
999
00:54:32.720 --> 00:54:36.980
much has been lamented about AI sort of producing miracle drugs and we'll have
1000
00:54:37.000 --> 00:54:38.260
to wait and see how that stuff happens.
1001
00:54:38.320 --> 00:54:42.640
But there is certainly a lot of local at-home diagnostics one can do
1002
00:54:43.440 --> 00:54:44.480
because of LLMs,
1003
00:54:44.600 --> 00:54:49.560
which is I think part because the images can analyze poor telemetry much better.
1004
00:54:49.840 --> 00:54:53.120
Just a bunch of iPhone images probably get you much better information about
1005
00:54:53.160 --> 00:54:55.760
whatever random thing you may have than previously,
1006
00:54:55.840 --> 00:55:00.500
but also because you can buy all of this low-end diagnostic equipment to Nat's
1007
00:55:00.560 --> 00:55:01.920
point and have it just working.
1008
00:55:03.060 --> 00:55:06.440
And I think if we do our job correctly as an industry,
1009
00:55:06.880 --> 00:55:10.360
this has to be a much larger category than anything we've produced to date
1010
00:55:10.740 --> 00:55:15.420
because the productive mastery of physics and biology to elevate
1011
00:55:15.560 --> 00:55:19.840
humans is a much bigger and much more interesting story than the production of
1012
00:55:19.900 --> 00:55:22.420
software. And we're in our very early innings now,
1013
00:55:22.480 --> 00:55:26.280
and I have proof of that because Nat's on eBay buying some sort of camera to
1014
00:55:26.320 --> 00:55:29.680
look at some dermatology camera or whatever.
1015
00:55:30.660 --> 00:55:33.280
So if I have to guess, if we meet for a year from today,
1016
00:55:33.560 --> 00:55:35.540
those anecdotes have spread, not around the world.
1017
00:55:35.600 --> 00:55:37.300
And I don't know if they'll make it all the way to the East Coast,
1018
00:55:37.360 --> 00:55:39.060
but at least amongst the tight,
1019
00:55:39.120 --> 00:55:42.980
high-quality alumnus of the Stripe Sessions attendees.
1020
00:55:43.760 --> 00:55:45.620
So John asked his last question.
1021
00:55:47.960 --> 00:55:52.160
My last question is, as we proceed into the singularity,
1022
00:55:53.780 --> 00:55:56.760
each, what is your one sentence of advice for Stripe?
1023
00:56:00.900 --> 00:56:02.020
I mean, agents are... I mean,
1024
00:56:02.080 --> 00:56:03.660
it's pretty obvious you're doing the things already,
1025
00:56:03.800 --> 00:56:08.200
but you really want to be the platform of choice for agents that are transacting
1026
00:56:08.220 --> 00:56:09.040
on the internet.
1027
00:56:09.040 --> 00:56:13.820
So it's a combination of building platforms that agents that are well
1028
00:56:13.860 --> 00:56:18.440
suited for agents as they start to exercise purchasing power and-.
1029
00:56:19.440 --> 00:56:22.360
But agents are not just some hype meme. They're here to stay.
1030
00:56:22.600 --> 00:56:25.680
Yeah, I think they're going to spend money. Yeah, they're going to spend money.
1031
00:56:25.760 --> 00:56:29.080
And there's a whole new stack that has to be built around identities and
1032
00:56:29.980 --> 00:56:33.360
disputes and pricing and all of these,
1033
00:56:33.580 --> 00:56:36.420
I think that stack will have to be built anew for agents.
1034
00:56:36.760 --> 00:56:40.640
And so I would build that and then I would make sure that you become somehow the
1035
00:56:40.840 --> 00:56:45.340
shelling point or the ecosystem for that because you really need the SEO.
1036
00:56:46.060 --> 00:56:50.520
You need the SEO in the model where the model for some reason says use
1037
00:56:50.880 --> 00:56:51.713
Stripe's platform.
1038
00:56:53.240 --> 00:56:56.440
If you go back to that analogy, as flawed as it may be of China,
1039
00:56:56.700 --> 00:57:00.040
I think one interesting thing that happened to the Chinese economy is obviously
1040
00:57:00.140 --> 00:57:03.700
they skipped a lot of the legacy technology stack that the West had.
1041
00:57:04.020 --> 00:57:05.980
Instead of email, they went directly to messaging,
1042
00:57:06.960 --> 00:57:11.600
Ant was obviously first to do QR code payment, Tap to Pay,
1043
00:57:11.720 --> 00:57:14.960
all that sort of stuff. And so if you now think forward to agents,
1044
00:57:15.040 --> 00:57:18.780
they're obviously going to leapfrog whatever we have today and do the directly
1045
00:57:18.880 --> 00:57:19.940
native thing. And so
1046
00:57:21.480 --> 00:57:24.820
you may want to ask the question if kind of a new continent was to be
1047
00:57:24.880 --> 00:57:28.020
discovered, which are going to be these agents,
1048
00:57:28.860 --> 00:57:30.200
what exactly would that be and how can you build-.
1049
00:57:30.260 --> 00:57:32.080
Are you suggesting the agents might like stablecoins?
1050
00:57:32.980 --> 00:57:36.560
I don't know that the treasury will be giving them Social Security numbers,
1051
00:57:36.620 --> 00:57:38.580
so my guess is they're going to like stablecoins.
1052
00:57:40.540 --> 00:57:42.760
With that folks, we are going to have to leave it there,
1053
00:57:43.220 --> 00:57:46.000
not just for this interview, but for Sessions.
1054
00:57:47.020 --> 00:57:49.180
As you've gotten a sense,
1055
00:57:49.860 --> 00:57:53.780
it's just the most interesting time by far that I think any of us have
1056
00:57:53.800 --> 00:57:56.640
experienced in technology.
1057
00:57:57.020 --> 00:57:59.300
Things are going so dizzyingly quickly,
1058
00:57:59.940 --> 00:58:03.160
and so that's why we thought it was valuable to gather everyone here on day 120,
1059
00:58:03.960 --> 00:58:06.840
days 119 and 120 of the singularity.
1060
00:58:07.420 --> 00:58:10.100
We'll be back next year in early May,
1061
00:58:10.720 --> 00:58:13.420
and we just can't wait to see what happens between now and then.
1062
00:58:13.520 --> 00:58:16.320
I'm very curious at what you guys all get up to, so see you back then there.
1063
00:58:16.720 --> 00:58:17.040
Thank you.