1
00:00:00,227 --> 00:00:01,207
Hey everyone.

2
00:00:10,427 --> 00:00:11,567
I'm Dillon Rolnick.

3
00:00:11,807 --> 00:00:14,287
I'm the CEO of Nous Research.

4
00:00:14,667 --> 00:00:20,307
And today we're going to talk about building on foundations and platforms that you can actually own.

5
00:00:27,367 --> 00:00:28,447
There we go.

6
00:00:28,947 --> 00:00:29,347
All right.

7
00:00:30,347 --> 00:00:34,607
How do you make back a centibillion dollar infrastructure and research bet?

8
00:00:36,227 --> 00:00:40,547
Labs individually have spent hundreds of billions of dollars

9
00:00:40,547 --> 00:00:44,627
training and building ever more sophisticated intelligence system

10
00:00:44,627 --> 00:00:47,927
and the industry as a whole has spent trillions, perhaps.

11
00:00:54,687 --> 00:01:02,087
The problem is, though, that the models have begun to commoditize. You guys in this room,

12
00:01:02,227 --> 00:01:08,147
who are hopefully practitioners of AI, have known this intuitively. You switch among models readily,

13
00:01:08,527 --> 00:01:15,307
you switch between Astra, Fable, Opus, Kimi, whatever it may be, without too much regard for

14
00:01:15,307 --> 00:01:18,807
who's on the back end or what it may be, you want to try the new model.

15
00:01:20,007 --> 00:01:21,607
We've seen this in data too.

16
00:01:21,787 --> 00:01:26,967
There were some charts going around Twitter this week showing the average cost of a token

17
00:01:26,967 --> 00:01:31,427
going down while the total spent on tokens surges.

18
00:01:31,987 --> 00:01:37,047
And the infrastructure costs associated with serving those tokens has stayed flat.

19
00:01:37,447 --> 00:01:43,607
Meaning where there used to be margin in a very, very performant model, it's now starting

20
00:01:43,607 --> 00:01:49,127
to narrow as we see competition and a baseline level of intelligence be available to people.

21
00:01:51,467 --> 00:01:58,187
Okay, so if the models are starting to commoditize and you've spent hundreds of billions, trillions

22
00:01:58,187 --> 00:02:01,587
of dollars on them, you need to make that back.

23
00:02:02,787 --> 00:02:08,907
you need to find a different way to do it. And in my opinion, there are two pretty obvious ways to do it.

24
00:02:10,607 --> 00:02:23,527
One, you find a different layer of the stack to compete at. You go down to the infrastructure or you go up to the user. Or two, you build God and you build God faster than anyone else and you monetize God.

25
00:02:25,647 --> 00:02:28,787
If you're in front of the user, you have an opportunity for stickiness.

26
00:02:29,307 --> 00:02:30,387
Stickiness means margin.

27
00:02:31,247 --> 00:02:34,947
This is why the fight has largely shifted to the agent layer lately.

28
00:02:35,087 --> 00:02:40,887
We started talking, I heard somebody previously talk about 5% of a workforce engaging in agents.

29
00:02:41,087 --> 00:02:45,547
This is becoming more and more the zeitgeist of where the AI battle is taking place.

30
00:02:47,447 --> 00:02:51,287
Consumers, though, are fickle and they're terrible customers.

31
00:02:52,907 --> 00:02:58,207
Businesses, on the other hand, spend tons of money on tokens, and they have economically

32
00:02:58,207 --> 00:03:00,207
productive use cases for those tokens.

33
00:03:01,287 --> 00:03:05,347
So if you're going to be competing at the user layer, and you're going to be bringing

34
00:03:05,347 --> 00:03:08,487
this fight somewhere else, you're going to be bringing it to businesses.

35
00:03:10,187 --> 00:03:14,107
The best thing to do is to get ingrained in a business, because when you're in a business,

36
00:03:14,267 --> 00:03:15,667
it's hard to take you out, right?

37
00:03:16,387 --> 00:03:26,747
And if you think about your business two, three years into the future, how much of your workforce, as measured by its output, is agentic?

38
00:03:28,107 --> 00:03:31,427
70% conservatively, probably more.

39
00:03:32,267 --> 00:03:41,467
And if 70% of your workforce is organized by somebody else who may or may not have adverse incentives to you, they can be organized into a strike.

40
00:03:41,927 --> 00:03:45,767
And so what you would do if you had to make back a trillion-dollar investment

41
00:03:46,587 --> 00:03:50,027
was you would embed, you would use your asymmetric leverage,

42
00:03:50,027 --> 00:03:51,607
and you would change the rules of the game.

43
00:03:53,947 --> 00:03:55,847
Or, like I said, you would build God.

44
00:03:59,027 --> 00:04:01,407
If you succeed at building God,

45
00:04:02,507 --> 00:04:04,467
what you would do is you would point God

46
00:04:04,467 --> 00:04:06,507
at the most lucrative problems in the world.

47
00:04:08,627 --> 00:04:14,007
The best shot at earning it back is not to sell it for an ever-decreasing margin to others

48
00:04:14,007 --> 00:04:16,227
to solve these hard and lucrative problems.

49
00:04:16,227 --> 00:04:17,847
The best shot is to solve them yourselves.

50
00:04:27,887 --> 00:04:34,827
Labs think that markets are solvable problems, and that's exactly how they view the financial

51
00:04:34,827 --> 00:04:36,307
industry, as a problem.

52
00:04:36,507 --> 00:04:43,347
to be solved. And a tractable problem at that, perhaps not even a very hard one.

53
00:04:45,027 --> 00:04:50,447
Right now, though, they know, at least the smart ones know, that they don't have enough context,

54
00:04:50,547 --> 00:04:55,667
know-how, or experience in the financial industry to be able to do that themselves. So they will

55
00:04:55,667 --> 00:05:01,607
partner. They hire investment banking analysts. They partner with some of the biggest firms on

56
00:05:01,607 --> 00:05:06,287
the globe, and they start to build their foundation here.

57
00:05:08,047 --> 00:05:11,207
My belief is that they will eventually go to the source because, like I said, that's

58
00:05:11,207 --> 00:05:14,027
where the money is and we're trying to make back a trillion dollars.

59
00:05:16,647 --> 00:05:19,747
And so you might say, this is nothing new.

60
00:05:19,867 --> 00:05:24,087
We have always been trying to trap people into these games and change the margin, and

61
00:05:24,087 --> 00:05:27,247
once we have you and you're stuck, we are able to raise the prices.

62
00:05:27,247 --> 00:05:29,207
But the difference is that the stakes are much higher.

63
00:05:29,207 --> 00:05:32,007
We're talking about meaningful

64
00:05:32,747 --> 00:05:35,127
Deca percentages of the Earth's economy

65
00:05:36,747 --> 00:05:38,447
And you in this room

66
00:05:40,527 --> 00:05:41,627
Need AI

67
00:05:43,207 --> 00:05:44,967
Everybody has seen what they can do

68
00:05:44,967 --> 00:05:45,947
What these models can do

69
00:05:45,947 --> 00:05:47,267
What agentic systems can do

70
00:05:47,267 --> 00:05:48,227
And there's no going back

71
00:05:50,007 --> 00:05:51,987
People in this room who are

72
00:05:54,427 --> 00:05:55,547
Making decisions

73
00:05:55,547 --> 00:05:58,087
Vice presidents and executives

74
00:05:58,087 --> 00:05:59,987
and engineers about what they're building on

75
00:05:59,987 --> 00:06:02,307
are going to have to make decisions now

76
00:06:02,307 --> 00:06:05,487
about where they grow roots around these tools and processes

77
00:06:05,487 --> 00:06:07,547
that will affect the future of these businesses.

78
00:06:09,327 --> 00:06:12,207
And whatever you pick, you'll start to harden.

79
00:06:13,087 --> 00:06:14,347
You'll gain know-how.

80
00:06:14,527 --> 00:06:17,827
It'll sop up the intelligence of the organization itself.

81
00:06:17,827 --> 00:06:20,467
You will start to build processes and systems around it.

82
00:06:20,907 --> 00:06:24,947
I implore you, do not build on quicksand foundations.

83
00:06:25,527 --> 00:06:26,987
Build on open foundations.

84
00:06:28,087 --> 00:06:29,427
Build on Hermes agent.

85
00:06:32,167 --> 00:06:36,967
So I just spent a fair amount of time convincing you, hopefully,

86
00:06:37,447 --> 00:06:40,787
that the highest expected value move in AI right now

87
00:06:40,787 --> 00:06:43,647
is to either trap you and jack up the prices

88
00:06:44,627 --> 00:06:46,367
or replace you directly.

89
00:06:46,847 --> 00:06:48,527
So why am I standing here?

90
00:06:50,587 --> 00:06:53,187
We view the world very differently.

91
00:06:54,567 --> 00:06:58,007
And the first thing I hope you guys will learn about me

92
00:06:58,007 --> 00:07:00,367
and about Nous Research is that we built this company

93
00:07:00,367 --> 00:07:02,767
primarily for moral and philosophical reasons.

94
00:07:03,667 --> 00:07:08,127
And a lot of times in business, morality and incentives don't always align,

95
00:07:08,127 --> 00:07:10,987
and when that happens, it's probably easiest to change your morality.

96
00:07:12,287 --> 00:07:14,627
But we think the world is changing, and we think it's different.

97
00:07:14,907 --> 00:07:19,067
And for that reason, we think that the best course of action

98
00:07:19,067 --> 00:07:20,487
is one in which we are both aligned.

99
00:07:23,767 --> 00:07:25,407
When we built the company originally,

100
00:07:25,407 --> 00:07:30,467
Finally, we built it to combat a seemingly inexorable trend towards centralization and

101
00:07:30,467 --> 00:07:35,047
centralized control of perhaps the world's most important technology and most important

102
00:07:35,047 --> 00:07:35,487
invention.

103
00:07:36,367 --> 00:07:41,167
Nous Research was built by and staffed by nerds, nerds who love science fiction.

104
00:07:41,467 --> 00:07:46,847
And we've all seen what happens when a centralized entity gets control of godlike intelligence

105
00:07:46,847 --> 00:07:48,547
and uses that for exploitation.

106
00:07:49,747 --> 00:07:52,007
Skynet was, after all, a closed source company.

107
00:07:55,467 --> 00:08:00,787
The next thing that you need to know is that there's no longer really a strong moat in

108
00:08:00,787 --> 00:08:01,807
proprietary software.

109
00:08:02,527 --> 00:08:06,007
The marginal cost to reproduce a piece of software has plummeted.

110
00:08:06,927 --> 00:08:13,367
A few months ago, XAI released GrokBots and we saw that and we thought, wow, that's an

111
00:08:13,367 --> 00:08:15,747
intuitive user interface, that's really nice.

112
00:08:16,287 --> 00:08:24,707
So we pointed Hermes Agent at it, and we reproduced it in a weekend and released it as a feature within Hermes Agent, open source, of course.

113
00:08:25,787 --> 00:08:27,387
And the game has changed.

114
00:08:27,727 --> 00:08:43,407
It used to be that software was difficult to produce, and it took coordinated teams of engineers long hours to pilfer open source and take libraries and combine them together into proprietary pieces of software and then sell that software at SaaS.

115
00:08:43,587 --> 00:08:44,967
But now the opposite is true.

116
00:08:45,747 --> 00:08:50,907
Anything can be reverse engineered, and there are many, many motivated open source builders

117
00:08:50,907 --> 00:08:54,027
pointing their agents at these different softwares.

118
00:08:55,367 --> 00:08:59,987
If the closed competitors do manage to make interesting features, open source will have

119
00:08:59,987 --> 00:09:01,087
it not long afterwards.

120
00:09:01,667 --> 00:09:04,387
The moats of yesterday were in proprietary IP.

121
00:09:04,887 --> 00:09:09,027
The moats of tomorrow are in network effects, reputation, and reliability.

122
00:09:11,407 --> 00:09:15,727
What's funny is that in a lot of ways, the labs have always seemed to underage

123
00:09:15,747 --> 00:09:21,147
underestimate their own tools, or perhaps they are underestimating the users of those tools and

124
00:09:21,147 --> 00:09:27,927
their genius and their motivation. This happened and is happening on the model side, and we'll talk

125
00:09:27,927 --> 00:09:34,007
about why that's important, but it turns out that models can fairly easily teach another model to be,

126
00:09:34,007 --> 00:09:42,787
you know, nearly as smart as the original model. And the advantage in a model is very rarely in

127
00:09:42,787 --> 00:09:47,027
the architecture or some novel system that was created in a proprietary way for that

128
00:09:47,027 --> 00:09:47,307
model.

129
00:09:47,607 --> 00:09:50,967
In fact, it's usually in the quality of the data and the scale.

130
00:09:53,707 --> 00:09:54,307
And...

131
00:09:55,387 --> 00:09:55,987
Sorry.

132
00:10:11,207 --> 00:10:14,987
the model company started to realize

133
00:10:14,987 --> 00:10:18,607
that the scale was catching up

134
00:10:18,607 --> 00:10:21,887
and that you could create these things

135
00:10:21,887 --> 00:10:24,687
with supplementary quality data

136
00:10:24,687 --> 00:10:28,227
by teaching one model from another.

137
00:10:28,427 --> 00:10:31,187
So they labeled the teaching process distillation

138
00:10:31,187 --> 00:10:36,067
and they labeled illicit distillation theft.

139
00:10:36,427 --> 00:10:38,767
You cannot call that copyright infringement

140
00:10:38,767 --> 00:10:40,547
because copyright infringement would imply

141
00:10:40,547 --> 00:10:43,427
that you somehow took every living human being's

142
00:10:43,427 --> 00:10:45,507
written or creative work and then trained on top of it.

143
00:10:46,547 --> 00:10:50,587
They also labeled the process of...

144
00:10:50,587 --> 00:10:54,107
They also trained the models to not assist in the process

145
00:10:54,107 --> 00:10:57,867
of training other models, and they labeled that safety.

146
00:11:04,767 --> 00:11:08,627
The next thing that you need to know about agents

147
00:11:08,627 --> 00:11:13,287
and about the new world is that agents are like operating systems

148
00:11:13,287 --> 00:11:15,567
and they're like frameworks, and they're a lot more like those

149
00:11:15,567 --> 00:11:17,327
than they are like chatbots.

150
00:11:17,667 --> 00:11:20,027
And so if you continue to think about your agent

151
00:11:20,027 --> 00:11:21,847
as merely a personal assistant,

152
00:11:22,507 --> 00:11:26,487
you will forever live on the surface of what it means to engage in agentic systems.

153
00:11:26,747 --> 00:11:27,807
And so what do I mean by that?

154
00:11:29,687 --> 00:11:34,247
If you guys were here a little bit earlier, there was a presentation about a financial firm

155
00:11:34,247 --> 00:11:39,827
and how they think about AI and these different inputs that go into ultimately what becomes an investment decision.

156
00:11:41,827 --> 00:11:47,487
Agents are really at their heart a system that combines determinism and non-determinism.

157
00:11:47,487 --> 00:11:54,907
If you remember from that talk, there are data sources, there are routine code, there are people's opinions, and there's judgment.

158
00:11:55,187 --> 00:11:59,147
And all these things come together to create what is the new application.

159
00:11:59,467 --> 00:12:02,647
The new application is no longer necessary that it is

160
00:12:02,787 --> 00:12:15,187
deterministic and solid and cold hard code. It can inject intelligence, reasoning, judgment, all of these things together to combine to create the new application.

161
00:12:15,187 --> 00:12:28,687
And so as you at your company think about building these new applications, you should be building these systems on foundations that you own.

162
00:12:28,787 --> 00:12:31,147
You should be able to swap in and out the models that you want.

163
00:12:31,147 --> 00:12:35,867
you should be creating ever more sophisticated towers of functionality

164
00:12:35,867 --> 00:12:42,867
on top of these core primitives of determinism, non-determinism,

165
00:12:43,047 --> 00:12:46,147
code, data sources, judgment, reasoning, intellect.

166
00:12:49,547 --> 00:12:54,927
But you should not be spending your entire lives being intelligence companies.

167
00:12:55,327 --> 00:12:57,787
What has made you successful this far?

168
00:12:57,907 --> 00:12:59,927
Most of you are most likely not agent companies.

169
00:12:59,927 --> 00:13:02,267
most of you are not likely AI companies.

170
00:13:03,247 --> 00:13:05,887
You should be focused on the things that you do well,

171
00:13:06,027 --> 00:13:08,487
and when the state-of-the-art changes and it evolves

172
00:13:08,487 --> 00:13:11,527
and what it means to be an agentic system starts to shift,

173
00:13:11,867 --> 00:13:14,587
you should inherit that, not rebuild it and maintain it.

174
00:13:15,787 --> 00:13:17,967
Remember, these are tools that should make you better.

175
00:13:18,187 --> 00:13:19,887
They are not drop-ins to replace you.

176
00:13:20,207 --> 00:13:23,327
Do you know how much energy it takes to create a human?

177
00:13:26,227 --> 00:13:28,647
And the beauty of the Hermes agent proposition

178
00:13:28,647 --> 00:13:31,607
is that you can run it however you want,

179
00:13:31,867 --> 00:13:34,207
wherever you want, modify it as you see fit.

180
00:13:34,347 --> 00:13:35,527
If you're not having fun anymore,

181
00:13:35,747 --> 00:13:37,507
you can take your ball and you can go home.

182
00:13:38,147 --> 00:13:40,147
The moat has changed.

183
00:13:40,427 --> 00:13:42,307
The moat is no longer entrapping you.

184
00:13:42,847 --> 00:13:44,647
The moat is in reputation.

185
00:13:45,927 --> 00:13:48,107
And that's what Hermes Agent offers you.

186
00:13:48,287 --> 00:13:52,047
And so what do I want out of all of this?

187
00:13:52,827 --> 00:13:54,227
Well, I want to build a standard.

188
00:13:55,027 --> 00:13:59,847
Hermes Agent today is the most used and downloaded open source agent in the world.

189
00:14:00,187 --> 00:14:04,567
It is order of magnitude more downloaded than some of its closed competitors.

190
00:14:05,987 --> 00:14:09,467
And I want everybody to be able to build on these open foundations.

191
00:14:09,807 --> 00:14:16,627
Because as long as one of us is willing to open source these tools and these agentic systems,

192
00:14:16,627 --> 00:14:18,307
water will continue to flow downhill.

193
00:14:18,307 --> 00:14:24,407
And I want to be downhill so that I can work with you to provide support, safety, scalability

194
00:14:24,407 --> 00:14:26,287
to go along with the world's best open agent.

195
00:14:27,207 --> 00:14:30,547
I've seen where this goes and we've seen where the moats will lie.

196
00:14:30,867 --> 00:14:33,807
Network effects, reliability, and reputation.

197
00:14:35,887 --> 00:14:40,107
People are going to continue to solve hard problems with agentic systems.

198
00:14:40,707 --> 00:14:44,287
People are going to solve lucrative problems with agentic systems.

199
00:14:45,507 --> 00:14:48,787
Somebody is going to earn back the trillion-dollar investments that we've made.

200
00:14:49,387 --> 00:14:50,507
Make sure that it's you.

201
00:14:50,927 --> 00:14:53,227
Make sure that you are in control of your own destiny.

202
00:14:53,907 --> 00:14:55,607
You can have your cake and eat it too.

203
00:14:55,987 --> 00:14:59,767
You can have state-of-the-art features and reliability without sacrificing your autonomy.

204
00:15:00,827 --> 00:15:01,987
Build on something you own.

205
00:15:02,287 --> 00:15:02,947
Build with us.

206
00:15:03,247 --> 00:15:03,747
Thank you.