1
00:00:00,000 --> 00:00:01,957
You do all your prompting on Ultrafast?

2
00:00:02,040 --> 00:00:04,557
>> I do. >> That's sick. >> Yeah.

3
00:00:04,640 --> 00:00:07,477
>> That's crazy. >> You know,
seven people's thoughts can

4
00:00:07,560 --> 00:00:11,197
working together create something new
that kind of existed in all of our heads

5
00:00:11,280 --> 00:00:15,240
to some degree,
but was like an emergent thing.

6
00:00:21,610 --> 00:00:23,630
>> [music]

7
00:00:26,240 --> 00:00:29,477
>> Hey, this is Dan Shipper. You might
be wondering what happened to AI and I.

8
00:00:29,560 --> 00:00:33,277
This is the same show. We are just
rebranding it to the Every Podcast. I'm

9
00:00:33,360 --> 00:00:35,757
really excited for our first
episode with Sam Altman.

10
00:00:35,840 --> 00:00:37,797
Here it is. >> Sam, welcome to the show.

11
00:00:37,880 --> 00:00:40,277
>> Thanks for having me. >> Psyched
to have you here. So, we're here

12
00:00:40,360 --> 00:00:45,117
at DevDay. I think if you had told me 4
years ago that the coding models were at

13
00:00:45,200 --> 00:00:48,917
this level of capability, I would have
been like, "Wow, I don't know if there

14
00:00:49,000 --> 00:00:50,597
would be a lot of developers at DevDay."

15
00:00:50,680 --> 00:00:53,957
But, there's more developers than ever,
it seems like, and more work to do than

16
00:00:54,040 --> 00:00:58,917
ever. So, what has the like riding the
increasing capabilities of coding models

17
00:00:59,000 --> 00:01:02,557
and seeing the impact on engineering
taught you about the future of

18
00:01:02,640 --> 00:01:06,197
engineering and uh more broadly
like the future of work with AI?

19
00:01:06,280 --> 00:01:09,517
>> I think people just want so much
more stuff. People's expectations are so

20
00:01:09,600 --> 00:01:13,917
high. Uh and I think that's a wonderful
thing. I think this is why we'll all

21
00:01:14,000 --> 00:01:18,117
have this incredible future in front
of us to explore. But, developers they've

22
00:01:18,200 --> 00:01:21,117
never been able to do more, but there's
never been a higher bar either. And I

23
00:01:21,200 --> 00:01:23,437
think that's great. I think that's
like a nice thing about society.

24
00:01:23,520 --> 00:01:24,997
>> Is that something that surprised you?

25
00:01:25,080 --> 00:01:26,757
Did you expect that to happen?

26
00:01:26,840 --> 00:01:30,757
>> I think I was never like
as doomy about the job stuff as other

27
00:01:30,840 --> 00:01:32,957
people, but it still has surprised me

28
00:01:33,040 --> 00:01:36,637
somewhat on the upside of just how much

29
00:01:36,720 --> 00:01:41,357
latent demand existed in the world.
And how And also how much like

30
00:01:41,440 --> 00:01:44,757
people have remained the center
of everything that is happening with AI.

31
00:01:44,840 --> 00:01:47,997
>> Do you think there's a a limit to that,
or do you think that translates pretty

32
00:01:48,080 --> 00:01:50,157
cleanly to other areas of work?

33
00:01:50,240 --> 00:01:53,717
>> I think there will be some areas
of work where there's actually not much

34
00:01:53,800 --> 00:01:55,317
creativity required, and we don't care

35
00:01:55,400 --> 00:01:57,557
about the person doing it. And many of

36
00:01:57,640 --> 00:02:01,157
those tasks or jobs will be done by AI.

37
00:02:01,240 --> 00:02:03,917
But I think there's so much other stuff
people want in the world. There's going

38
00:02:04,000 --> 00:02:06,997
to be no shortage of new things to go
to go do. I think that's been one of the

39
00:02:07,080 --> 00:02:11,357
lessons of recent time.
>> Why do you think it's so human-centered

40
00:02:11,440 --> 00:02:14,357
even though we talk
a lot about automation?

41
00:02:14,440 --> 00:02:17,677
>> I mean, what one observation is we just
we don't care that much about what the

42
00:02:17,760 --> 00:02:21,877
machines do.
And in some sense we never have.

43
00:02:21,960 --> 00:02:24,437
And we're just like we're obsessed with
other people. We want to know about the

44
00:02:24,520 --> 00:02:27,957
person behind the art or the person
behind the company or the idea or the

45
00:02:28,040 --> 00:02:33,197
product. And I think this is like
a wonderful positive surprise.

46
00:02:33,280 --> 00:02:35,037
That we probably should
have all known before.

47
00:02:35,120 --> 00:02:38,837
>> I agree. It's I've been thinking like
if an agent solves an obvious Stokes in a

48
00:02:38,920 --> 00:02:42,357
data center but no one finds out,
did it really solve it, you know?

49
00:02:42,440 --> 00:02:44,437
>> [laughter] >> I think
that was like a great example

50
00:02:44,520 --> 00:02:48,677
where, you know, 10,000 agents worked
super hard for 4 days in perfect

51
00:02:48,760 --> 00:02:52,797
coordination with each other and did
this amazing thing. And almost none of

52
00:02:52,880 --> 00:02:57,277
the public reaction was about that
actually, the problem being solved.

53
00:02:57,360 --> 00:02:59,397
A lot of talk about the interpretation.

54
00:02:59,480 --> 00:03:02,957
A lot of talk about like the people
involved in it in different ways. But,

55
00:03:03,040 --> 00:03:04,597
you know, that was like
a that was a good thing.

56
00:03:04,680 --> 00:03:07,677
>> Yeah. So, it's DevDay.
You launched by my

57
00:03:07,760 --> 00:03:10,077
count 22 new things today,

58
00:03:10,160 --> 00:03:12,597
uh which I've been at all four DevDays.

59
00:03:12,680 --> 00:03:16,397
And uh I think last year was like nine
and the the year before that was like

60
00:03:16,480 --> 00:03:19,757
four or five. So,
it's a lot more than usual.

61
00:03:19,840 --> 00:03:23,997
>> Just wait for next year. >> Okay.
[laughter] I'm tired. I'm tired.

62
00:03:24,080 --> 00:03:25,717
>> Um we cut a bunch
of things that we were

63
00:03:25,800 --> 00:03:27,117
going to launch today just cuz you have

64
00:03:27,200 --> 00:03:30,197
to like fit it in 50 minutes. Uh but

65
00:03:30,280 --> 00:03:35,197
you know, like we try to practice what
we preach. And if we are saying these

66
00:03:35,280 --> 00:03:37,437
tools can make people much more
productive and build more and that

67
00:03:37,520 --> 00:03:39,277
people want more stuff, like

68
00:03:39,360 --> 00:03:41,197
we better build more ourselves.

69
00:03:41,280 --> 00:03:44,837
>> What are the new
kind of bottlenecks and

70
00:03:44,920 --> 00:03:49,917
best practices for doing that? Cuz I
I can see for myself and and I assume

71
00:03:50,000 --> 00:03:53,117
you guys are seeing this too. Like now
that you can build everything, you're

72
00:03:53,200 --> 00:03:56,877
like what what should we build and and
how do we make it all coherent? How do

73
00:03:56,960 --> 00:04:00,757
you think about that? >> Moving to

74
00:04:00,840 --> 00:04:04,437
this world of sort of persistent
intelligence, active intelligence,

75
00:04:04,520 --> 00:04:06,997
whatever people want to call it.
This idea that you're going to have these

76
00:04:07,080 --> 00:04:11,597
agents that are running in the
background 24/7, connected to

77
00:04:11,680 --> 00:04:15,357
your systems, doing stuff proactively
for you. This is clearly a direction

78
00:04:15,440 --> 00:04:19,797
that feels important to explore. I think
we're going to see amazing stuff happen

79
00:04:19,880 --> 00:04:21,437
there. Another thing we really wanted to

80
00:04:21,520 --> 00:04:24,317
do is there's just there are now so many

81
00:04:24,400 --> 00:04:26,917
ideas that are capable of being built.

82
00:04:27,000 --> 00:04:30,597
And what we are missing
is the creative spirit

83
00:04:30,680 --> 00:04:32,637
to figure them out and build them.

84
00:04:32,720 --> 00:04:37,477
So we wanted to really open up
our ecosystem, let people build apps in

85
00:04:37,560 --> 00:04:41,077
ChatGPT and like the same way that we
get to do them, let people bring their

86
00:04:41,160 --> 00:04:45,997
subscription wherever they'd like, have
the marketplace, and and really try to

87
00:04:46,080 --> 00:04:49,997
let people get connected to all the stuff
that people will now be able to build.

88
00:04:50,080 --> 00:04:52,157
>> Yeah, I mean I I
just looking at all the

89
00:04:52,240 --> 00:04:56,237
platform features that you launched
and the way that you talk about platforms,

90
00:04:56,320 --> 00:04:59,717
it reminds me a lot of the way that Bill
Gates used to talk about Windows as a

91
00:04:59,800 --> 00:05:04,077
platform where you create more value
than you capture. What have you learned,

92
00:05:04,160 --> 00:05:07,597
you know, you said this in your keynote
and I've observed it as well. There's at

93
00:05:07,680 --> 00:05:10,557
the end of the day there's always a an app
store. You you've been taking swings

94
00:05:10,640 --> 00:05:14,357
at app stores for a while. What have
you learned about building a platform and

95
00:05:14,440 --> 00:05:17,077
what do you think makes
this one different?

96
00:05:17,160 --> 00:05:20,597
>> I I mean that we we clearly have not
yet figured out the app store equivalent,

97
00:05:20,680 --> 00:05:23,917
but we have figured out how to build
a platform that a lot of people are

98
00:05:24,000 --> 00:05:26,877
building incredible value on top of. I I

99
00:05:26,960 --> 00:05:29,517
think there are really kind of two

100
00:05:29,600 --> 00:05:34,197
schools of thought among people
building AI model companies. There are

101
00:05:34,280 --> 00:05:37,557
people that believe or hope
that there's only going to be

102
00:05:37,640 --> 00:05:41,957
one company in the world. And then there
are people like us that believe we are

103
00:05:42,040 --> 00:05:44,117
going to see a renaissance in the number

104
00:05:44,200 --> 00:05:47,357
of companies and just an absolute

105
00:05:47,440 --> 00:05:49,477
lowering of the bar to

106
00:05:49,560 --> 00:05:54,437
people, one-person companies, small-team
companies creating incredible new

107
00:05:54,520 --> 00:05:58,597
products and services. And we want to go
as hard as we can with the platform to

108
00:05:58,680 --> 00:06:01,477
enable people to do that. And although
I think we haven't figured out the App

109
00:06:01,560 --> 00:06:05,357
Store yet, that's totally fair. I think
we have figured out how to help like

110
00:06:05,440 --> 00:06:07,557
millions of people build and create

111
00:06:07,640 --> 00:06:09,597
significant new products and companies.

112
00:06:09,680 --> 00:06:14,397
>> If you had to narrow in you said
you talked at the keynote about an

113
00:06:14,480 --> 00:06:17,677
industrial revolution versus renaissance
which you just which you just mentioned

114
00:06:17,760 --> 00:06:19,837
now. If you had to narrow in on the key

115
00:06:19,920 --> 00:06:22,757
things that make the difference between

116
00:06:22,840 --> 00:06:26,997
those two worlds and maybe OpenAI's
part in it, but maybe even the broader

117
00:06:27,080 --> 00:06:29,717
ecosystem, what do you
think those things are?

118
00:06:29,800 --> 00:06:33,557
>> One take on this would be that the
renaissance was fundamentally centered

119
00:06:33,640 --> 00:06:37,357
around people and the industrial
revolution was fundamentally

120
00:06:37,440 --> 00:06:39,917
centered around machines.
There were a lot of good things about

121
00:06:40,000 --> 00:06:43,117
the industrial revolution. It had a you
know, hugely positive impact on the

122
00:06:43,200 --> 00:06:48,237
world in in many ways.
But I think a lot of people did not have a

123
00:06:48,320 --> 00:06:49,877
purely positive personal impact from

124
00:06:49,960 --> 00:06:53,997
that. And this feeling that um

125
00:06:54,080 --> 00:06:58,317
people had these like rich lives and did
all these things. And yeah, of course

126
00:06:58,400 --> 00:07:00,797
they like struggled and wanted more
and better stuff. And that's the part the

127
00:07:00,880 --> 00:07:04,277
industrial revolution did deliver
in some ways. But they were kind of turned

128
00:07:04,360 --> 00:07:09,677
into in many cases this like cog in a
machine doing some extremely specialized

129
00:07:09,760 --> 00:07:12,997
thing and kind of like disconnected from
a lot of the parts that made life good.

130
00:07:13,080 --> 00:07:17,197
I get sad to hear people talk about like
oh, AI is going to automate everything.

131
00:07:17,280 --> 00:07:18,597
There's going to be no role for humans.

132
00:07:18,680 --> 00:07:20,597
We're going to be like house cats and

133
00:07:20,680 --> 00:07:25,717
the AI will decide how to reward
us and tell us what to do. And um

134
00:07:25,800 --> 00:07:27,477
I have heard people say that.
>> [laughter]

135
00:07:27,560 --> 00:07:30,117
>> Um that's certainly not what we think.

136
00:07:30,200 --> 00:07:32,677
Um what I am hopeful for is that this is

137
00:07:32,760 --> 00:07:35,597
just a huge new wave of scientific

138
00:07:35,680 --> 00:07:38,437
discovery, art, company formation, like

139
00:07:38,520 --> 00:07:39,997
everybody that wants to start a new

140
00:07:40,080 --> 00:07:43,957
thing. Um and that people do get the

141
00:07:44,040 --> 00:07:47,597
benefits of all of the economic growth
and the new stuff happening here.

142
00:07:47,680 --> 00:07:51,797
But also people get to back
in control their lives and ability

143
00:07:51,880 --> 00:07:54,797
to kind of do the things they want
and spend their time the way they want.

144
00:07:54,880 --> 00:07:57,677
>> One of the things that strikes me about
the Renaissance versus the Industrial

145
00:07:57,760 --> 00:08:01,317
Revolution is in the Renaissance
you have a sort of

146
00:08:01,400 --> 00:08:05,597
different understanding of what it means
to be a person and like an individual

147
00:08:05,680 --> 00:08:08,357
uh, as opposed >> Yeah.
>> as opposed to sort of the cog in the

148
00:08:08,440 --> 00:08:10,797
machine. Do you feel like you're

149
00:08:10,880 --> 00:08:13,477
starting to see any difference in how we

150
00:08:13,560 --> 00:08:18,757
think of ourselves as human that's
comparable to the way that we changed

151
00:08:18,840 --> 00:08:20,677
how we think of ourselves
during the Renaissance?

152
00:08:20,760 --> 00:08:22,997
>> I mean so much happened during
the Renaissance. It's hard to say there's

153
00:08:23,080 --> 00:08:25,317
like this one thing. But but but
definitely a big part of it was about

154
00:08:25,400 --> 00:08:29,397
people like sort of reconnecting
to a lot of things that had

155
00:08:29,480 --> 00:08:34,197
in the in the distant past.
And you do hear and then and then also

156
00:08:34,280 --> 00:08:37,837
like this kind of combination of that
with this crazy futurism and discovery

157
00:08:37,920 --> 00:08:41,957
of science and engineering and just
sort of this this feeling that the human

158
00:08:42,040 --> 00:08:45,437
potential had massively expanded.
And you do hear people talk about both of

159
00:08:45,520 --> 00:08:49,477
those things simultaneously with AI,
which is kind of how I got interested in

160
00:08:49,560 --> 00:08:53,397
this as an analogy in the first place.
I mean this is like a very trivial example

161
00:08:53,480 --> 00:08:58,037
relative to what is going to be
possible or already is possible.

162
00:08:58,120 --> 00:09:00,837
But the Dots, the new thing we launched

163
00:09:00,920 --> 00:09:06,477
today, is the first thing that has given
me back control of my mornings and like

164
00:09:06,560 --> 00:09:09,117
my mental space in the mornings. I used
to love the early mornings. It was my

165
00:09:09,200 --> 00:09:12,517
favorite time of the day to work. It's
like maximum creativity. That's when I

166
00:09:12,600 --> 00:09:15,597
would like do the really deep
complicated project of the day.

167
00:09:15,680 --> 00:09:19,597
And then now I like run this like big
company and something is going wrong all

168
00:09:19,680 --> 00:09:23,677
the time and overnight a lot of stuff
happens. And if I don't check my phone

169
00:09:23,760 --> 00:09:26,637
right away then you know,
sometimes it was fine and

170
00:09:26,720 --> 00:09:29,597
sometimes there was like a real disaster
that if I ignored for an hour just got

171
00:09:29,680 --> 00:09:34,437
much worse. The problem had been that as I
checked my phone, I polluted my brain

172
00:09:34,520 --> 00:09:35,637
with all these other things and I lost

173
00:09:35,720 --> 00:09:39,237
the deep work period. But now I trust my

174
00:09:39,320 --> 00:09:43,757
dot to go have good
judgment about what I need to

175
00:09:43,840 --> 00:09:46,117
think about and I'll think about if
there's something very urgent, otherwise

176
00:09:46,200 --> 00:09:50,277
handle things for me. And I get this
space for kind of like creativity or to

177
00:09:50,360 --> 00:09:54,477
just hang out with my kids uh back and
that has been like, you know, it sounds

178
00:09:54,560 --> 00:09:56,677
small but like a significant
improvement to my life.

179
00:09:56,760 --> 00:10:00,157
>> That's what I was going to ask is so it
look it seems like it's deflecting a lot

180
00:10:00,240 --> 00:10:04,157
of things that you might have had
to deal with immediately that you can just

181
00:10:04,240 --> 00:10:07,357
like leave for later cuz you're
confident you get the thing you need

182
00:10:07,440 --> 00:10:08,437
when when you check it in the morning.

183
00:10:08,520 --> 00:10:13,237
>> Yeah. I mean two two things
that happened to today about that. Um

184
00:10:13,320 --> 00:10:17,597
during the keynote we had like a big
seven production. This is the risk of live

185
00:10:17,680 --> 00:10:20,197
demos and trying to deploy
many new features at once.

186
00:10:20,280 --> 00:10:22,997
>> I was like curling.
I was >> Um >> [laughter]

187
00:10:23,080 --> 00:10:25,557
>> so so that happened. Uh

188
00:10:25,640 --> 00:10:28,357
Thibault got a message from his dot 5

189
00:10:28,440 --> 00:10:32,757
minutes before the demo saying,
"Just to let you know, um

190
00:10:32,840 --> 00:10:35,997
there's I see that you're in the dev day.
You know, I know you were working

191
00:10:36,080 --> 00:10:39,797
on it. I know what the demo is cuz I
was able to like look at the presentation

192
00:10:39,880 --> 00:10:43,197
um in space or whatever and there, you
know, this sub is probably going to make

193
00:10:43,280 --> 00:10:46,197
the demo not work. Would you like me
to try to fix it? There's like a few

194
00:10:46,280 --> 00:10:49,557
minutes to go." And now it couldn't fix

195
00:10:49,640 --> 00:10:54,037
it yet but like maybe the next version
will be able to go fix that in prod. And

196
00:10:54,120 --> 00:10:57,437
like think about that whole like
all the stuff that had to come

197
00:10:57,520 --> 00:11:00,957
together to make that work. Um just
right now before I came here, I've been

198
00:11:01,040 --> 00:11:03,317
like back-to-back stuff all day.

199
00:11:03,400 --> 00:11:06,077
Very very behind on all
of my normal stuff.

200
00:11:06,160 --> 00:11:09,757
Um my dot just before I
had walked into the last

201
00:11:09,840 --> 00:11:14,157
thing said, "You have one little break
between now and this thing you're going

202
00:11:14,240 --> 00:11:17,077
to go to with Dan. You have to pay
attention to this one thing." I never

203
00:11:17,160 --> 00:11:20,637
would have found that otherwise and like
it would have been a big mess. And just

204
00:11:20,720 --> 00:11:24,357
the fact that I like don't have to worry
about that feels feels super different.

205
00:11:24,440 --> 00:11:26,277
>> What's your dot's name? >> Dotty.

206
00:11:26,360 --> 00:11:29,077
>> Dotty. >> I tried I
had it Alfred earlier.

207
00:11:29,160 --> 00:11:31,717
And then like many people are calling
it Dotty for now but I'm trying to like

208
00:11:31,800 --> 00:11:32,997
think of a great name. >> yeah.

209
00:11:33,080 --> 00:11:35,677
>> Mine's name is Boo.
>> That's a good one.

210
00:11:35,760 --> 00:11:37,717
>> a good one. Um >> What do
you think so far? I know it's

211
00:11:37,800 --> 00:11:38,797
like super early and still rough.

212
00:11:38,880 --> 00:11:43,837
>> I I like it. It has become the default
way that I use chat, which I think is a

213
00:11:43,920 --> 00:11:45,837
very promising thing. >> Me, too.

214
00:11:45,920 --> 00:11:49,637
>> And I I shared with you like I think
you know, I'm I'm in the beta, so I was in

215
00:11:49,720 --> 00:11:53,517
the the early period, so I I think it's
it you know, it has limits. It's like

216
00:11:53,600 --> 00:11:56,077
has been a little bit
buggy for me, so >> Yeah.

217
00:11:56,160 --> 00:11:57,517
>> there's a lot
of different surfaces that

218
00:11:57,600 --> 00:12:00,717
it can be used on like mobile and it has
its own cloud computer and like all that

219
00:12:00,800 --> 00:12:03,957
kind of stuff and I think those kinds
of things getting that cleaned up is a big

220
00:12:04,040 --> 00:12:06,837
deal, but yeah. >> That that I
think this is the kind of

221
00:12:06,920 --> 00:12:10,997
thing where it's going to get like better
rapidly each week from here, but it's

222
00:12:11,080 --> 00:12:13,757
going to definitely take some
time to smooth it all out.

223
00:12:13,840 --> 00:12:17,357
>> I had a similar example where I was
trying to reschedule a flight and it was

224
00:12:17,440 --> 00:12:20,317
like, "Hey, you have a meeting during
this flight that you're rescheduling."

225
00:12:20,400 --> 00:12:23,197
but it wasn't on my calendar.
It had just like read it in Slack.

226
00:12:23,280 --> 00:12:24,797
>> Yeah. >> And then it was like, yeah.

227
00:12:24,880 --> 00:12:29,277
>> That that thing about the ability
to bring context together from across your

228
00:12:29,360 --> 00:12:32,757
entire life and like really do things
in ways that would be hard even for an

229
00:12:32,840 --> 00:12:36,797
attentive person to get right has been
super impressive. That's been one of the

230
00:12:36,880 --> 00:12:38,197
things I've found most useful about it.

231
00:12:38,280 --> 00:12:42,437
>> What had to happen from a research
perspective to make this possible?

232
00:12:42,520 --> 00:12:45,677
>> This is like a very complicated
system with a lot of stuff, but maybe most

233
00:12:45,760 --> 00:12:49,837
important was Astra. We needed like
a You needed a smart enough model. And

234
00:12:49,920 --> 00:12:52,797
then there were a lot of other things
you had to build, too, but but getting

235
00:12:52,880 --> 00:12:56,397
the model over the intelligence
bar was a key part.

236
00:12:56,480 --> 00:12:59,517
>> Mhm. You told me that there's like
there've been a couple times, not just

237
00:12:59,600 --> 00:13:04,237
for like scheduling or or or things
that are urgent like uh filtering, but

238
00:13:04,320 --> 00:13:07,637
there've been a couple times where
it's helped you see something or make a

239
00:13:07,720 --> 00:13:11,357
decision with an AI that you could not
have. You said you couldn't have hired

240
00:13:11,440 --> 00:13:14,957
someone to do that. Like it just be too
complicated. Do you have any specific

241
00:13:15,040 --> 00:13:18,317
examples you can share?
>> There was one like new feature that I

242
00:13:18,400 --> 00:13:22,757
was kind of excited about and it
would have been very hard to explain it

243
00:13:22,840 --> 00:13:26,757
to people in enough detail to get it
built. Um and even then there were like

244
00:13:26,840 --> 00:13:29,557
a lot of ways to do it and I wouldn't
have been confident in which one. And I

245
00:13:29,640 --> 00:13:34,237
just like I would just like constantly
jot notes as I thought of something or

246
00:13:34,320 --> 00:13:36,197
as I was like on a walk or whatever to

247
00:13:36,280 --> 00:13:41,597
my dot. And then the dot went off
and built like five or six versions. I was

248
00:13:41,680 --> 00:13:44,157
able to get feedback. I was able to
iterate. And it was just like whenever I

249
00:13:44,240 --> 00:13:49,557
had a spare moment or had an idea and
that kind of like the ability to just like

250
00:13:49,640 --> 00:13:53,237
dump that in raw form, the ability
to like get and give pretty continuous

251
00:13:53,320 --> 00:13:57,277
feedback on it was very cool. And then
there's like many categories like that

252
00:13:57,360 --> 00:14:01,077
where like uh an idea that would be

253
00:14:01,160 --> 00:14:04,317
you'd be so unsure about or it would be
too much effort for you or for someone

254
00:14:04,400 --> 00:14:07,357
else to build.
You can actually just kind of like

255
00:14:07,440 --> 00:14:09,397
continually iterate on and and it's it's

256
00:14:09,480 --> 00:14:14,757
like it's pretty cool to go to sleep
and like just leave a long voice note as

257
00:14:14,840 --> 00:14:17,077
you're kind of drifting
off with a bunch of thoughts

258
00:14:17,160 --> 00:14:19,917
and like wake up in the morning not only
is the thing built, but it like went off

259
00:14:20,000 --> 00:14:23,477
and did like the next seven steps. It's
the best. And it's like all right, this

260
00:14:23,560 --> 00:14:27,997
is like a very cool way to express stuff.
And then and then also like there

261
00:14:28,080 --> 00:14:30,997
there's a set of tasks that I would
normally give up on or that I would have

262
00:14:31,080 --> 00:14:35,677
given up on before. There was a thing
that I was sure I had seen in Slack.

263
00:14:35,760 --> 00:14:39,557
And Slack search is obviously terrible,
so I mostly use Codex to search it. And

264
00:14:39,640 --> 00:14:42,077
I had been searching for this thing cuz
I wanted to find it again. I wanted to

265
00:14:42,160 --> 00:14:47,237
reference it. And I spent like a good
10 or 20 minutes trying to search with

266
00:14:47,320 --> 00:14:50,277
Codex and find this and I kept trying
different things and didn't find it. And

267
00:14:50,360 --> 00:14:53,637
I was like, man, like maybe I dreamed it

268
00:14:53,720 --> 00:14:57,877
and this didn't really happen
because Codex doesn't usually fail.

269
00:14:57,960 --> 00:15:02,837
And I kind of gave up on it.
And then the next day um

270
00:15:02,920 --> 00:15:06,077
my dot was like, hey, I noticed
you were trying to do this thing

271
00:15:06,160 --> 00:15:09,197
and you didn't find it, but I like kept
thinking about it overnight and I found

272
00:15:09,280 --> 00:15:13,157
it for you. >> Wow. >> And I was like,
what? And it was a screenshot.

273
00:15:13,240 --> 00:15:16,957
>> Mhm. >> I had forgotten in my head it
was a screenshot not text in Slack. And it

274
00:15:17,040 --> 00:15:19,357
just like sat there very patiently
looking for something to do to be

275
00:15:19,440 --> 00:15:23,997
helpful to me and like looked at every
image wrecked every screenshot and found

276
00:15:24,080 --> 00:15:27,797
the thing. And I was like that's really
that's really >> have given up on that.

277
00:15:27,880 --> 00:15:31,877
>> I think there's a you're
we're constantly feeling that

278
00:15:31,960 --> 00:15:34,877
there's way too much to read in Slack
on email and all that kind of stuff and

279
00:15:34,960 --> 00:15:38,797
it's it's perfect for that amount
of context to just like get you the right

280
00:15:38,880 --> 00:15:42,797
thing at the right time. >> People have
talked about the need to have an agent

281
00:15:42,880 --> 00:15:47,037
deal with the just onslaught of the
internet and all of the various things

282
00:15:47,120 --> 00:15:50,717
trying to like hack our attention
and distract us and whatever and I have

283
00:15:50,800 --> 00:15:52,637
definitely learned in the few weeks that

284
00:15:52,720 --> 00:15:55,877
I've been using dots that

285
00:15:55,960 --> 00:16:00,157
although technology is very wonderful
in many ways, the kind of like crazy

286
00:16:00,240 --> 00:16:02,317
attention hacking of social media and

287
00:16:02,400 --> 00:16:04,717
phones and notifications had gotten way

288
00:16:04,800 --> 00:16:07,997
worse than I realized for me personally.

289
00:16:08,080 --> 00:16:10,357
>> You hear a lot about AI tools on this

290
00:16:10,440 --> 00:16:15,197
show, but the one that's truly at the
center of your work is your CRM. It's

291
00:16:15,280 --> 00:16:18,837
where all of your customer contacts
lives and where the work either happens

292
00:16:18,920 --> 00:16:23,557
or gets missed. Attio delivers the work
that you never had time to focus on

293
00:16:23,640 --> 00:16:27,117
before and more.
It picks up leads at 2:00 a.m. It

294
00:16:27,200 --> 00:16:29,877
catches renewals before they slip and

295
00:16:29,960 --> 00:16:34,597
drafts the follow-up before you've even
thought to send one. Ask Attio anything

296
00:16:34,680 --> 00:16:39,357
about your business from this week's
forecast to which deals have gone quiet

297
00:16:39,440 --> 00:16:41,037
and get the answer in seconds. And

298
00:16:41,120 --> 00:16:43,677
because it's extensible with an API and

299
00:16:43,760 --> 00:16:46,917
an MCP, your agents work right alongside

300
00:16:47,000 --> 00:16:51,237
your team to build pipeline, advance
deals, and grow accounts around the

301
00:16:51,320 --> 00:16:53,357
clock. Try Attio for free at

302
00:16:53,440 --> 00:16:55,557
attio.com/every.

303
00:16:55,640 --> 00:16:59,597
That's Attio, a t t i o.com/every.

304
00:16:59,680 --> 00:17:01,797
And now, back to the episode.

305
00:17:01,880 --> 00:17:04,477
>> Tell me about Space.
It's it's a big bet

306
00:17:04,560 --> 00:17:08,597
to do three like office suite
type tools in one launch.

307
00:17:08,680 --> 00:17:10,917
>> We didn't set out wanting to do this.

308
00:17:11,000 --> 00:17:13,757
And then we realized, and I'm sure

309
00:17:13,840 --> 00:17:15,517
you've realized your own version of this

310
00:17:15,600 --> 00:17:18,757
too, that software was really not

311
00:17:18,840 --> 00:17:22,757
designed on the whole
for collaborating with AI.

312
00:17:22,840 --> 00:17:25,077
And you can use it yourself or you can

313
00:17:25,160 --> 00:17:28,837
have AI use it on your behalf, but to

314
00:17:28,920 --> 00:17:32,157
really collaborate closely, especially
when it's multiple people and multiple

315
00:17:32,240 --> 00:17:35,917
people's AIs all working together,
like a lot of the current abstractions

316
00:17:36,000 --> 00:17:40,077
we have break down. So, this is
definitely an ambitious bet, but uh

317
00:17:40,160 --> 00:17:44,077
but I think that the space of human
AI collaboration is really important to

318
00:17:44,160 --> 00:17:46,317
explore. >> How are you
working with it? Like are

319
00:17:46,400 --> 00:17:51,037
you reviewing docs in space or are you

320
00:17:51,120 --> 00:17:53,357
making docs and do you have expectations

321
00:17:53,440 --> 00:17:56,117
for how much of it you're typing versus

322
00:17:56,200 --> 00:18:00,837
how much of it your dot is doing for you
and also with the with the docs that you

323
00:18:00,920 --> 00:18:04,157
get? Like are you are you viewing
it from that perspective, too?

324
00:18:04,240 --> 00:18:08,357
>> If there's like a new idea,
um I like writing the old-fashioned way

325
00:18:08,440 --> 00:18:12,397
cuz it helps me like think about it.
So, I still do that and then maybe have my

326
00:18:12,480 --> 00:18:14,637
dot or other people's dot like

327
00:18:14,720 --> 00:18:19,637
do the follow-ups, but one of the things
that I have found most amazing about it

328
00:18:19,720 --> 00:18:22,117
and I don't think we have a word
for this yet, but the for lack of a better

329
00:18:22,200 --> 00:18:26,237
word, these like live documents,
um creating a doc that is going to

330
00:18:26,320 --> 00:18:28,837
automatically update as people interact

331
00:18:28,920 --> 00:18:30,717
or as other things change in the world

332
00:18:30,800 --> 00:18:33,597
and something that like many of our dots

333
00:18:33,680 --> 00:18:38,197
can work with together not only to like
share information, but but actually

334
00:18:38,280 --> 00:18:41,877
like, you know, seven people's dots can
working together create something new

335
00:18:41,960 --> 00:18:45,357
that kind of existed in all of our heads
to some degree, but it was like an

336
00:18:45,440 --> 00:18:49,437
emergent thing. That's been very fun as we
think about like new features to explore.

337
00:18:49,520 --> 00:18:53,197
>> Is there a specific live doc
that has been most useful for you?

338
00:18:53,280 --> 00:18:56,477
>> I have switched like the dashboard,
I think everybody's probably got like a

339
00:18:56,560 --> 00:18:59,797
dashboard that they look at every day
for how the company's doing in the way

340
00:18:59,880 --> 00:19:03,357
they care about and you know, I used
to share one with everybody else or many

341
00:19:03,440 --> 00:19:08,117
other people and I now have
one that is kind of like

342
00:19:08,200 --> 00:19:12,597
a changing summary
of the business just for me

343
00:19:12,680 --> 00:19:15,957
based off of like the evolving things
that I'm caring about or not caring

344
00:19:16,040 --> 00:19:19,397
about. Like looks at all my Slack
messages, makes me a new thing, like

345
00:19:19,480 --> 00:19:22,837
constantly updates it, and it's like, "Oh,
this just feels like a living document."

346
00:19:22,920 --> 00:19:25,157
>> That's sick.
What What do you care about now?

347
00:19:25,240 --> 00:19:29,157
>> A lot on my mind recently is just
how we're thinking about the safety

348
00:19:29,240 --> 00:19:32,797
alignment and security
challenge in front of us. Um

349
00:19:32,880 --> 00:19:36,317
But, you know, 6 months ago, I probably
cared more about revenue and growth and

350
00:19:36,400 --> 00:19:39,957
metrics of this sort of traditional
part of the business. So, watching that

351
00:19:40,040 --> 00:19:43,517
evolve. I I've actually been thinking
about how I can like cuz I'd be very

352
00:19:43,600 --> 00:19:47,997
curious to see what my dot thinks I care
about each day the most over time. And

353
00:19:48,080 --> 00:19:51,237
we don't currently have a good way
to visualize that. I have not been able to

354
00:19:51,320 --> 00:19:55,157
get my dot to figure out a cool way
to do that. And I would love to see that

355
00:19:55,240 --> 00:19:57,677
over the course of years.
>> I think that's really cool. I've been

356
00:19:57,760 --> 00:20:01,397
wanting one that just like it's
a live dashboard of my mood.

357
00:20:01,480 --> 00:20:04,557
Uh because I I have my dot
like when I get a Slack

358
00:20:04,640 --> 00:20:07,157
message and I'm like, I don't know if
I want to look at this Slack message, I

359
00:20:07,240 --> 00:20:10,237
have my dot go look at it, and then
I'm like, "Should I look at this an hour

360
00:20:10,320 --> 00:20:12,757
later?" And it's like,
"You should look later."

361
00:20:12,840 --> 00:20:16,317
Um and I kind of want like a live
thing of my mood, so it knows

362
00:20:16,400 --> 00:20:18,227
Yeah, yeah. I I want that, too.

363
00:20:18,310 --> 00:20:22,517
>> [laughter] >> Um very cool.
Tell me about Decisions API.

364
00:20:22,600 --> 00:20:24,677
I feel like the Decisions API is

365
00:20:24,760 --> 00:20:29,877
one of the sneaky most important things
today because Jev is this new hot thing

366
00:20:29,960 --> 00:20:33,757
that has honestly changed how we think
about building things internally. Is

367
00:20:33,840 --> 00:20:37,517
Decisions API Did you start that from Jev,
or is has it been in the works for

368
00:20:37,600 --> 00:20:39,237
a while? How do you think about it?

369
00:20:39,320 --> 00:20:44,477
>> Diogo uh used to work at OpenAI,
and um Diogo's awesome, and he was talking

370
00:20:44,560 --> 00:20:47,077
about this class of ideas for a long time.

371
00:20:47,160 --> 00:20:52,637
Um and we had been working on

372
00:20:52,720 --> 00:20:56,317
trying to figure out how to make it work.
We We like the number one request

373
00:20:56,400 --> 00:20:59,757
of Luna, and this is like an example
of how people always want like more

374
00:20:59,840 --> 00:21:02,797
cheaper, faster, better. But, the number
one request with Luna, even though

375
00:21:02,880 --> 00:21:07,397
people love that model, was I want
it to be even cheaper and even faster.

376
00:21:07,480 --> 00:21:11,277
And so, we have been thinking about how we
could push it further in that direction.

377
00:21:11,360 --> 00:21:15,677
Um definitely seeing
the response to Jev made us

378
00:21:15,760 --> 00:21:17,557
more excited to prioritize things in

379
00:21:17,640 --> 00:21:20,557
that direction. Um but and like Diogo

380
00:21:20,640 --> 00:21:25,037
has inspired me for a long time
on thinking about this. Um but I think you

381
00:21:25,120 --> 00:21:29,357
will see us try to push you know,
in different ways in the

382
00:21:29,440 --> 00:21:31,757
spirit of make

383
00:21:31,840 --> 00:21:36,557
the technology super fast and super
low cost for common use cases.

384
00:21:36,640 --> 00:21:40,877
>> How do you think about, you know,
if something like Jev comes out, it's a

385
00:21:40,960 --> 00:21:45,317
kind of new type of model, and then
OpenAI can can add like a decisions API?

386
00:21:45,400 --> 00:21:47,717
How do you think about how that changes

387
00:21:47,800 --> 00:21:50,637
the race between new labs and labs and

388
00:21:50,720 --> 00:21:54,797
and what what OpenAI can cover
and what a new lab can focus on?

389
00:21:54,880 --> 00:21:58,157
>> As I understand the approach of Jev,
it's pretty different, and I don't think

390
00:21:58,240 --> 00:22:02,197
we can go I don't well, maybe we could
go do something from scratch like that,

391
00:22:02,280 --> 00:22:03,757
but I think we have other things to go

392
00:22:03,840 --> 00:22:06,517
do. You know, like we want to make

393
00:22:06,600 --> 00:22:11,117
the best, cheapest, fastest,
most capable model at every point on the

394
00:22:11,200 --> 00:22:15,117
Pareto curve of cost and speed and
modality, but I think there will be new

395
00:22:15,200 --> 00:22:18,237
approaches, new architectures,
new technologies that are things we just

396
00:22:18,320 --> 00:22:22,397
can't cover all of. But for the the kind
of main line workhorse, like we will try

397
00:22:22,480 --> 00:22:26,437
to deliver that. >> So I've
been like in your ear a lot

398
00:22:26,520 --> 00:22:29,237
about Codex native apps
for a little while now.

399
00:22:29,320 --> 00:22:32,077
>> did it. >> And you finally
did it. I'm psyched. Um

400
00:22:32,160 --> 00:22:36,997
so plugin extensions, um yeah,
I I and also being able to bring

401
00:22:37,080 --> 00:22:40,557
your tokens from your ChatGPT
subscription to an app developer, like

402
00:22:40,640 --> 00:22:43,677
it changes the equation
for someone like me where

403
00:22:43,760 --> 00:22:47,637
normally you're you're thinking about,
okay, how much AI can I let my users use

404
00:22:47,720 --> 00:22:50,517
so that doesn't bankrupt me? And then
you're like, can I use should I use uh

405
00:22:50,600 --> 00:22:53,557
lower-priced models? And then the
experience is worse, but if someone can

406
00:22:53,640 --> 00:22:55,717
bring their subscription, I'm like,
great, you can use whatever model you

407
00:22:55,800 --> 00:22:58,877
want under your subscription. What are
you most excited about for the kinds of

408
00:22:58,960 --> 00:23:02,757
applications and products that can be
built at the application layer in this

409
00:23:02,840 --> 00:23:04,997
new world. >> I mean,
I'm most excited for the ones

410
00:23:05,080 --> 00:23:08,917
that we haven't thought of. Like there's
many reasons that we wanted to do this,

411
00:23:09,000 --> 00:23:13,037
but one of the biggest is and why we do

412
00:23:13,120 --> 00:23:15,517
this big DevDay thing is just that this

413
00:23:15,600 --> 00:23:18,517
this very deep belief that there are way

414
00:23:18,600 --> 00:23:23,317
more good ideas outside of our building
than inside. And I don't know what I'm

415
00:23:23,400 --> 00:23:25,517
going to love the most as like a ChatGPT

416
00:23:25,600 --> 00:23:28,877
native app a plugin extension, but I am

417
00:23:28,960 --> 00:23:32,717
very confident that there will be things
that I really do love that I cannot

418
00:23:32,800 --> 00:23:35,597
imagine today and that no one at OpenAI
really could imagine with conviction

419
00:23:35,680 --> 00:23:38,877
today. And that's why I mean,
that's one of the reasons why

420
00:23:38,960 --> 00:23:43,957
this is so important to do. Like we we
want to find a way to get that creative

421
00:23:44,040 --> 00:23:47,477
spirit into people's hands and we want
to find a way to help builders connect

422
00:23:47,560 --> 00:23:50,917
with people who are going to find the
technology useful. In terms of bringing

423
00:23:51,000 --> 00:23:54,357
your ChatGPT subscription with you
everywhere, like we just want that

424
00:23:54,440 --> 00:23:57,677
ourselves. There's all this great stuff
that people are building. There's all

425
00:23:57,760 --> 00:23:58,917
these new apps. We all we want to try

426
00:23:59,000 --> 00:24:02,677
them. Um and I like I hate having to

427
00:24:02,760 --> 00:24:06,037
sign up for 100 subscriptions. And also
like when I try something I'm like,

428
00:24:06,120 --> 00:24:09,557
"Hey, I you know, like I'd like to be
able to use this with like full juice

429
00:24:09,640 --> 00:24:14,037
AI." Um so we we want to make it just
super easy for people to take it with

430
00:24:14,120 --> 00:24:18,557
them and it's your subscription like it's
you should get to use wherever you want.

431
00:24:18,640 --> 00:24:21,517
>> Do you think anyone at the app
layer is doing this well today?

432
00:24:21,600 --> 00:24:24,837
>> Like building interesting apps? Well,
not with the new stuff just because it's

433
00:24:24,920 --> 00:24:28,397
brand new, but but yes, I mean,
I think there's an unbelievable

434
00:24:28,480 --> 00:24:32,117
for all of the fear people had a couple
of years ago about, "Man, AI model

435
00:24:32,200 --> 00:24:34,837
companies are just going to eat
everything." There's an unbelievable

436
00:24:34,920 --> 00:24:38,277
amount of significant
new companies, great new

437
00:24:38,360 --> 00:24:42,477
products building this stuff all of the
time. I I've never seen startups get as

438
00:24:42,560 --> 00:24:46,037
successful as quickly. >> Does
that change, you know, you you were

439
00:24:46,120 --> 00:24:50,197
part of Y Combinator before OpenAI,
does that change how you think about when

440
00:24:50,280 --> 00:24:52,797
people ask you about, "How do I start
a startup?" or "How do I think about how

441
00:24:52,880 --> 00:24:54,037
to grow?" like all that kind of stuff?

442
00:24:54,120 --> 00:24:58,197
>> I would love to be a startup investor
again. Um I mean, cuz like oh man, what

443
00:24:58,280 --> 00:25:02,157
a fun time. Like, I that the watching
what people are doing right now, getting

444
00:25:02,240 --> 00:25:04,517
to spend time with some of these
founders, which I don't get nearly as

445
00:25:04,600 --> 00:25:08,717
much as I like,
it is it is just a golden age.

446
00:25:08,800 --> 00:25:10,877
>> What are you looking for?
I mean, obviously besides the founder

447
00:25:10,960 --> 00:25:13,437
themselves, like what are you looking
for in the shape of things that you'd be

448
00:25:13,520 --> 00:25:15,597
like, oh man, I want to invest in that.

449
00:25:15,680 --> 00:25:18,637
>> Mhm. I was always a very founder-driven

450
00:25:18,720 --> 00:25:22,837
investor. Like, I I I have my pet
projects like anybody else, but my my

451
00:25:22,920 --> 00:25:26,957
whole MO as an investor was like,
find the absolute best founders I can that

452
00:25:27,040 --> 00:25:29,517
are working on the ambitious projects
and be willing to take a lot of risk

453
00:25:29,600 --> 00:25:32,197
with them. >> Yeah,
I I feel like there's always that

454
00:25:32,280 --> 00:25:36,917
narrative of the net labs are going to eat
everything and Dev Days like the the

455
00:25:37,000 --> 00:25:40,957
tweet that always works is like,
OpenAI just killed X startup or whatever.

456
00:25:41,040 --> 00:25:44,117
>> The tweet works because people
like doomerism, but it never actually

457
00:25:44,200 --> 00:25:47,157
happens. [laughter] >> Yeah,
why is that? Why does it never happen?

458
00:25:47,240 --> 00:25:51,317
>> Yeah. >> Well, A, I think
we have a lot of conviction that

459
00:25:51,400 --> 00:25:55,197
empowering builders and not just trying
to be like the one giant evil mega corp

460
00:25:55,280 --> 00:26:00,597
is important. So, I I think we took
a lot of steps today that no other AI lab

461
00:26:00,680 --> 00:26:03,717
would do about saying we're going
to give you all of the same tools we use.

462
00:26:03,800 --> 00:26:08,917
We've always pushed the frontier on on
model access. And so, I think we're like

463
00:26:09,000 --> 00:26:11,997
uniquely committed in the industry
to doing this. And we we really do love

464
00:26:12,080 --> 00:26:15,837
developers and I think that's partly I
just also love startups and that's where

465
00:26:15,920 --> 00:26:18,517
I came from. But, I also
think speaking of like a new

466
00:26:18,600 --> 00:26:21,797
renaissance, it's just
there is so much to explore.

467
00:26:21,880 --> 00:26:24,077
People are going to do
such incredible stuff.

468
00:26:24,160 --> 00:26:27,917
We are going to have such a tiny
fraction of that vision ourselves

469
00:26:28,000 --> 00:26:29,997
that, you know, like the world has a

470
00:26:30,080 --> 00:26:34,997
fundamentally new enabler. This is like
I mean, I don't even know what this is

471
00:26:35,080 --> 00:26:38,517
like fire or something. This is like
a real change to what's possible. You

472
00:26:38,600 --> 00:26:40,837
know, maybe we like brought it down from
the mountain, but we want to give it to

473
00:26:40,920 --> 00:26:43,357
everybody and we want people
to like do stuff with it.

474
00:26:43,440 --> 00:26:47,317
>> There's a lot of basic
startup lore, like you should

475
00:26:47,400 --> 00:26:50,597
always have a co-founder or, you know,
even the structure of how Y Combinator

476
00:26:50,680 --> 00:26:56,117
does investments that is based on a
previous, for example, cost structure.

477
00:26:56,200 --> 00:26:59,237
If you were going to start those things
from scratch or think about them from

478
00:26:59,320 --> 00:27:00,997
scratch, like what do you think changes?

479
00:27:01,080 --> 00:27:04,277
>> There are a few things
that are probably just like

480
00:27:04,360 --> 00:27:07,317
invariants. You know, you got to like
make something people want. That advice

481
00:27:07,400 --> 00:27:09,477
was good or agents want 100 years ago.

482
00:27:09,560 --> 00:27:14,397
>> Yeah, okay. Well, >> yes, but at least

483
00:27:14,480 --> 00:27:17,157
by our understanding of the world today,
like at some level people got to want

484
00:27:17,240 --> 00:27:21,157
it. You got it. So,
there's some stuff that just

485
00:27:21,240 --> 00:27:24,597
hasn't changed. And then there's
a lot of other things like

486
00:27:24,680 --> 00:27:27,397
how much work you can get done with one
person or two people or whatever that

487
00:27:27,480 --> 00:27:31,837
feels totally different.
And I don't think there has been enough

488
00:27:31,920 --> 00:27:35,997
re-examination of that
in this sort of startup

489
00:27:36,080 --> 00:27:37,637
conveyor belt of Silicon Valley yet.

490
00:27:37,720 --> 00:27:40,757
>> As a person who's
running a gigantic company

491
00:27:40,840 --> 00:27:43,637
and generally probably wouldn't have
time for building stuff and it now seems

492
00:27:43,720 --> 00:27:47,197
like you are, how does that change

493
00:27:47,280 --> 00:27:51,237
how you are able to lead or what you're
able to do that you can like make stuff

494
00:27:51,320 --> 00:27:54,717
on the side? >> I mean, we are
not a gigantic company, first of

495
00:27:54,800 --> 00:27:57,997
all. Like there's probably, I don't know,
like maybe the big tech companies

496
00:27:58,080 --> 00:27:59,437
have like 100 times more people than we

497
00:27:59,520 --> 00:28:02,037
do, something like that. But,

498
00:28:02,120 --> 00:28:05,597
I am way I I I I really believe in if

499
00:28:05,680 --> 00:28:09,877
you lower friction a little
bit, people build way more.

500
00:28:09,960 --> 00:28:12,317
And I have really felt that now with

501
00:28:12,400 --> 00:28:15,837
dots and these like, you know, little

502
00:28:15,920 --> 00:28:19,797
chunks of time I have or even a like
a long chunk of time where I can like have

503
00:28:19,880 --> 00:28:21,877
these interactive things get a lot done.

504
00:28:21,960 --> 00:28:26,197
Uh I I I feel like I will do
something that is not just like a

505
00:28:26,280 --> 00:28:29,357
little side project for myself like I
used to do with Codex, but like I hope I

506
00:28:29,440 --> 00:28:31,837
launch something to the world. Like I
hope I actually make a cool new thing

507
00:28:31,920 --> 00:28:33,997
and like get it over the threshold.

508
00:28:34,080 --> 00:28:37,117
>> And like does that go out on like
samaltman.com or you're talking about

509
00:28:37,200 --> 00:28:38,997
like a real like OpenAI product?

510
00:28:39,080 --> 00:28:42,117
>> Uh I meant like I
would try to like build a

511
00:28:42,200 --> 00:28:44,477
plugin extension
or something and put it out.

512
00:28:44,560 --> 00:28:47,167
>> That's that's >> Maybe I Maybe
I'll ship a real OpenAI product.

513
00:28:47,250 --> 00:28:50,317
>> [laughter] >> Feels >>
like all bets are off now.

514
00:28:50,400 --> 00:28:52,677
And so, you know, you have
to make that decision. Do I

515
00:28:52,760 --> 00:28:54,397
>> You know what?
I will do that. I'm inspired now.

516
00:28:54,480 --> 00:28:56,317
>> What's Do you have >> I don't know.
I want to go think about

517
00:28:56,400 --> 00:28:57,277
it and I want to do something.

518
00:28:57,360 --> 00:29:00,037
>> I love it. I love it. But you you do
have to make that decision of like you

519
00:29:00,120 --> 00:29:04,557
could just put a team on that. So,
what do you get to What does that add that

520
00:29:04,640 --> 00:29:07,237
instead of having to put a team on it,
you can now be like, "Oh, I'm going to

521
00:29:07,320 --> 00:29:10,317
make a plug-in." >> Uh well,
I think today was an example of

522
00:29:10,400 --> 00:29:12,997
this. Like you talked about how
the amount of stuff every dev day has gone

523
00:29:13,080 --> 00:29:17,637
up. Like we could not have done all of
this stuff without having AI to help us.

524
00:29:17,720 --> 00:29:21,597
And if this works, like next year we will

525
00:29:21,680 --> 00:29:24,837
put the teams on it and I will do
something and more people will do

526
00:29:24,920 --> 00:29:28,717
something and we'll have like
a crazy amount of amazing stuff.

527
00:29:28,800 --> 00:29:32,797
Or just one really great thing.
That'd be fine, too. But like hopefully

528
00:29:32,880 --> 00:29:36,637
the total amount of like utility
we put out into the world goes up.

529
00:29:36,720 --> 00:29:39,237
>> Do you do all your
prompting on ultrafast?

530
00:29:39,320 --> 00:29:42,717
>> I do. >> That's sick. >> Yeah.
[laughter] >> That's crazy.

531
00:29:42,800 --> 00:29:44,117
>> No, I've been I've been thinking about

532
00:29:44,200 --> 00:29:48,677
this. Um I I worry that in some ways

533
00:29:48,760 --> 00:29:52,557
we kind of get out of touch with people
building on our tools and we try we've

534
00:29:52,640 --> 00:29:56,317
tried to like put all these constraints
on ourselves. You know, I'm like very

535
00:29:56,400 --> 00:29:59,957
aware that not everybody is going to be
able to use ultrafast all the time for

536
00:30:00,040 --> 00:30:03,437
everything. But I have such
a hard time dragging it back over.

537
00:30:03,520 --> 00:30:07,677
>> Given that you're like living in that
future that I think, you know, all of us

538
00:30:07,760 --> 00:30:10,797
can experience a little taste of it,
what is what is different about the

539
00:30:10,880 --> 00:30:13,557
ultrafast world in terms of how you can

540
00:30:13,640 --> 00:30:15,077
use these tools?

541
00:30:15,160 --> 00:30:18,037
>> There was an amazing talk like, I don't

542
00:30:18,120 --> 00:30:20,837
know, 15 20 years ago from Bret Victor

543
00:30:20,920 --> 00:30:25,677
about kind of the principles
that he thought people like

544
00:30:25,760 --> 00:30:28,997
how he did his work.
And one of the things I'm going to get

545
00:30:29,080 --> 00:30:31,677
this a little bit wrong, but it
was something like creators should have an

546
00:30:31,760 --> 00:30:32,997
immediate connection to what they're

547
00:30:33,080 --> 00:30:35,637
creating. And how speed and the feedback

548
00:30:35,720 --> 00:30:39,237
loop was so important. And I did not

549
00:30:39,320 --> 00:30:41,917
realize what a difference that makes to

550
00:30:42,000 --> 00:30:47,477
my kind of iterative thinking loop
to have responses come in so quickly. But

551
00:30:47,560 --> 00:30:49,717
But in terms of trying to build
something, trying to think through a

552
00:30:49,800 --> 00:30:52,797
problem, it's like a wild difference.

553
00:30:52,880 --> 00:30:54,717
>> I know you've been thinking for a long

554
00:30:54,800 --> 00:30:58,237
time about AI as a democratizing force

555
00:30:58,320 --> 00:31:00,237
to allow to allow anyone to build stuff.

556
00:31:00,320 --> 00:31:04,477
And in the sort of ultra fast world
or in the Astra is a more expensive model

557
00:31:04,560 --> 00:31:08,157
world, I know you're also thinking about
that means something that people can pay

558
00:31:08,240 --> 00:31:11,797
more for better intelligence. So, like
like what are you What are you thinking

559
00:31:11,880 --> 00:31:14,917
there? >> So, historically,
what we have mostly

560
00:31:15,000 --> 00:31:18,957
focused on and I think done a great job
of is bringing down the like price of

561
00:31:19,040 --> 00:31:23,237
IQ. And we have made
hugely more capable models

562
00:31:23,320 --> 00:31:27,557
that get dramatically dramatically
cheaper every few months or every year.

563
00:31:27,640 --> 00:31:31,037
Now, there's this new vector that is
clearly going to be very important of

564
00:31:31,120 --> 00:31:34,877
speed. And we are putting the investments
in place. Like our own chip

565
00:31:34,960 --> 00:31:39,477
is going to be an incredible contributor
to letting us get very fast inference in

566
00:31:39,560 --> 00:31:43,557
the hands of everybody to say nothing
of what the models will do to improve

567
00:31:43,640 --> 00:31:47,077
research and our our inference
stack and everything else.

568
00:31:47,160 --> 00:31:50,877
And I think we now realize that we've
got to go invest in the democratization

569
00:31:50,960 --> 00:31:54,797
of this like we have
for model intelligence.

570
00:31:54,880 --> 00:31:59,997
>> So, when can I expect to use
ultra fast for all my problems?

571
00:32:00,080 --> 00:32:01,437
>> You can pay us for it and do it now.

572
00:32:01,520 --> 00:32:03,017
Like uh >> Yeah,
I don't think I can afford it.

573
00:32:03,100 --> 00:32:07,317
[laughter] >> I think we can deliver 8x

574
00:32:07,400 --> 00:32:11,757
at a great price in the not distant
future. But then we'll have 100x faster.

575
00:32:11,840 --> 00:32:16,157
And like, you know, and there
they'll also be like, you know, GPT-7

576
00:32:16,240 --> 00:32:18,677
galaxy or whatever. Um

577
00:32:18,760 --> 00:32:21,557
And so, I I think I I think the right

578
00:32:21,640 --> 00:32:23,757
way to think about this is

579
00:32:23,840 --> 00:32:29,437
as far as I can see there will
always be a real frontier in

580
00:32:29,520 --> 00:32:33,477
terms of intelligence or speed
or whatever else at a higher price. But I

581
00:32:33,560 --> 00:32:35,717
think you can also bet that we will keep

582
00:32:35,800 --> 00:32:37,677
doing what we've done of just a

583
00:32:37,760 --> 00:32:40,997
ridiculous like put Moore's law to shame

584
00:32:41,080 --> 00:32:44,637
relentless focus on driving value per

585
00:32:44,720 --> 00:32:48,597
dollar up and up and up kind
of in this big company way.

586
00:32:48,680 --> 00:32:50,317
>> Amazing. Sam, thank you so much.

587
00:32:50,400 --> 00:32:52,800
>> Thank you.

588
00:32:53,540 --> 00:32:55,560
>> [applause]

589
00:32:56,240 --> 00:32:57,877
>> THANK YOU ALL FOR COMING. I DIDN'T
REALIZE THERE WAS GOING TO BE LIKE A

590
00:32:57,960 --> 00:33:00,157
LIVE AUDIENCE, but that was fun.

591
00:33:00,240 --> 00:33:02,840
See you. >> See you.