WEBVTT

NOTE Sentence-level transcript of https://www.youtube.com/watch?v=Sg6vtKOAOQ0

NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/Sg6vtKOAOQ0.html. A cue ends where the next begins, or 2 s after its last word.

s1
00:00:01.309 --> 00:00:03.309
[music]

s2
00:00:12.160 --> 00:00:13.040
I'm Dylan Rolnick.

s3
00:00:13.040 --> 00:00:16.480
I'm the CEO of News Research.

s4
00:00:16.480 --> 00:00:23.360
And today we're going to talk about building on foundations and platforms that you can actually own.

s5
00:00:29.000 --> 00:00:30.400
Whoop, there we go.

s6
00:00:30.400 --> 00:00:32.119
All right.

s7
00:00:32.119 --> 00:00:37.680
How do you make back a cent a billion-dollar infrastructure and research bet?

s8
00:00:37.760 --> 00:00:50.840
Labs individually have spent hundreds of billions of dollars training and building ever more sophisticated intelligence system and the industry as a whole has spent trillions, perhaps.

s9
00:00:56.440 --> 00:01:02.280
The problem is though that the models have begun to commoditize.

s10
00:01:02.360 --> 00:01:07.920
You guys in this room who are hopefully practitioners of AI have known this intuitively.

s11
00:01:07.920 --> 00:01:09.840
You switch among models readily.

s12
00:01:09.840 --> 00:01:19.040
You switch between Astra, Fable, Opus, Kimmy, whatever it may be without too much regard for who's on the back end or what it may be.

s13
00:01:19.040 --> 00:01:21.640
You want to try the new model.

s14
00:01:21.640 --> 00:01:22.920
We've seen this in data, too.

s15
00:01:22.920 --> 00:01:35.840
There was some charts going around Twitter this week showing the average cost of a token going down while the total spent on tokens surges and the infrastructure costs associated

s16
00:01:35.840 --> 00:01:38.880
with serving those tokens has stayed flat.

s17
00:01:38.880 --> 00:01:52.080
Meaning where there used to be margin in a very, very performant model, it's now starting to narrow as we see competition and you know, a baseline level of intelligence be available to people.

s18
00:01:53.040 --> 00:02:03.372
Okay, so if the models are starting to commoditize and you've spent hundreds of billions, trillions of dollars on them, you need to make that back.

s19
00:02:03.372 --> 00:02:06.252
And you need to find a different way to do it.

s20
00:02:06.252 --> 00:02:11.732
And in my opinion, there are two pretty obvious ways to do it.

s21
00:02:11.732 --> 00:02:14.932
One, you find a different layer of the stack to compete at.

s22
00:02:14.932 --> 00:02:19.452
You go down to the infrastructure, or you go up to the user.

s23
00:02:19.452 --> 00:02:26.572
Or two, you build God, and you build God faster than anyone else, and you monetize God.

s24
00:02:27.332 --> 00:02:30.532
If you're in front of the user, you have an opportunity for stickiness.

s25
00:02:30.532 --> 00:02:32.932
Stickiness means margin.

s26
00:02:32.932 --> 00:02:36.292
This is why the fight has largely shifted to the agent layer lately.

s27
00:02:36.292 --> 00:02:42.652
We started talking, I heard somebody previously talk about 5% of, you know, a workforce engaging in agents.

s28
00:02:42.652 --> 00:02:48.492
This is becoming more and more the zeitgeist of where the AI battle is taking place.

s29
00:02:48.572 --> 00:02:50.301
Consumers, though,

s30
00:02:50.301 --> 00:02:50.732
[sighs]

s31
00:02:50.732 --> 00:02:53.932
are fickle, and they're terrible customers.

s32
00:02:53.932 --> 00:03:02.852
Businesses, on the other hand, spend tons of money on tokens, and they have economically productive use cases for those tokens.

s33
00:03:02.852 --> 00:03:11.332
So, if you're going to be competing at the user layer, and you're going to be bringing this fight somewhere else, you're going to be bringing it to businesses.

s34
00:03:11.812 --> 00:03:18.132
The best thing to do is to get ingrained in a business, cuz when you're in a business, it's hard to take you out, right?

s35
00:03:18.132 --> 00:03:29.452
And if you think about your business 2 3 years into the future, how much of your workforce, as measured by its output, is agentic?

s36
00:03:29.572 --> 00:03:34.052
70% conservatively, probably more.

s37
00:03:34.052 --> 00:03:43.532
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.

s38
00:03:43.532 --> 00:03:54.652
And so, what you would do if you had to make back a trillion-dollar investment, was you would embed, you would use your asymmetric leverage, and you would change the rules of the game.

s39
00:03:55.012 --> 00:03:58.892
Or, like I said, you would build God.

s40
00:04:00.572 --> 00:04:03.007
If you succeed at building God,

s41
00:04:03.007 --> 00:04:04.054
[sighs]

s42
00:04:04.054 --> 00:04:09.494
what you would do is you would point God at the most lucrative problems in the world.

s43
00:04:10.374 --> 00:04:17.534
The best shot at earning it back is not to sell it for an ever-decreasing margin to others to solve these hard and lucrative problems.

s44
00:04:17.534 --> 00:04:20.694
The best shot is to solve them yourselves.

s45
00:04:28.454 --> 00:04:33.094
Labs think that markets are solvable problems.

s46
00:04:33.094 --> 00:04:39.974
And that's exactly how they view the financial industry as a problem to be solved.

s47
00:04:41.014 --> 00:04:43.134
And a tractable problem at that.

s48
00:04:43.134 --> 00:04:46.374
Perhaps not even a very hard one.

s49
00:04:46.534 --> 00:04:56.054
Right now though, they know, at least the smart ones know, that they don't have enough context, know-how, or experience in the financial industry to be able to do that themselves.

s50
00:04:56.054 --> 00:04:58.014
So, they will partner.

s51
00:04:58.014 --> 00:05:00.414
They hire investment banking analysts.

s52
00:05:00.414 --> 00:05:04.254
They partner with some of the biggest firms on the globe.

s53
00:05:04.254 --> 00:05:09.294
And they start to build their foundation here.

s54
00:05:09.494 --> 00:05:16.854
My belief is that they will eventually go to the source because, like I said, that's where the money is, and we're trying to make back a trillion dollars.

s55
00:05:18.094 --> 00:05:21.014
And so, you might say, "This is nothing new.

s56
00:05:21.014 --> 00:05:28.534
We have always been trying to trap people into these games and change the margin, and once we have you and you're stuck, we are able to raise the prices."

s57
00:05:28.534 --> 00:05:31.094
But, the difference is that the stakes are much higher.

s58
00:05:31.094 --> 00:05:37.934
We're talking about meaningful deca-percentages of the Earth's economy.

s59
00:05:38.374 --> 00:05:44.614
And you in this room need AI.

s60
00:05:44.734 --> 00:05:51.094
Everybody has seen what they can do, what these models can do, what agentic systems can do, and there's no going back.

s61
00:05:51.574 --> 00:06:10.335
People in this room who are um making decisions, vice presidents, and executives, and engineers about what they're building on, are going to have to make decisions now about where they grow roots around these tools and processes that will affect the future of these businesses.

s62
00:06:10.935 --> 00:06:14.615
And whatever you pick, you'll start to harden.

s63
00:06:14.615 --> 00:06:15.615
You'll gain know-how.

s64
00:06:15.615 --> 00:06:19.175
It'll stop up the intelligence of the organiza- organization itself.

s65
00:06:19.175 --> 00:06:22.535
You will start to build processes and systems around it.

s66
00:06:22.535 --> 00:06:27.135
I implore you, do not build on quicksand found- quicksand foundations.

s67
00:06:27.135 --> 00:06:29.655
Build on open foundations.

s68
00:06:29.735 --> 00:06:32.335
Build on Hermes Agent.

s69
00:06:33.815 --> 00:06:45.855
So, I just spent a fair amount of time convincing you, hopefully, that the highest expected value move in AI right now is to either trap you and jack up the prices,

s70
00:06:45.855 --> 00:06:48.415
or replace you directly.

s71
00:06:48.415 --> 00:06:51.135
So, why am I standing here?

s72
00:06:51.135 --> 00:06:55.575
Uh we view the world very differently.

s73
00:06:55.575 --> 00:07:05.255
And the first thing I hope you guys will learn about me and about News Research is that we built this company primarily for moral and philosophical reasons.

s74
00:07:05.255 --> 00:07:13.735
And a lot of times in business, morality and incentives don't always align, and when that happens, it's probably easiest to change your morality.

s75
00:07:13.855 --> 00:07:16.055
But, we think the world is changing, and we think it's different.

s76
00:07:16.055 --> 00:07:23.375
And for that reason, we think that the best course of action is one in which we are both aligned.

s77
00:07:25.255 --> 00:07:37.615
When we built the company originally, we built it to combat a seemingly inexorable trend towards centralization and centralized control of perhaps the world's most most important technology and most important invention.

s78
00:07:37.615 --> 00:07:41.055
News Research was built by and staffed by nerds.

s79
00:07:41.055 --> 00:07:43.175
Nerds who love science fiction.

s80
00:07:43.175 --> 00:07:50.935
And we've all seen what happens when a centralized entity gets control of god-like intelligence and uses that for exploitation.

s81
00:07:50.935 --> 00:07:54.895
Skynet was, after all, a closed-source company.

s82
00:07:57.135 --> 00:08:04.175
The next thing that you need to know is that there's no longer really a strong moat in proprietary software.

s83
00:08:04.175 --> 00:08:08.537
The marginal cost to reproduce a piece of software has plummeted.

s84
00:08:08.537 --> 00:08:16.217
A few months ago, uh xAI released Grok bots and we saw that and we thought, "Wow, that's an intuitive user interface.

s85
00:08:16.217 --> 00:08:17.937
That's really nice."

s86
00:08:17.937 --> 00:08:27.377
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.

s87
00:08:27.377 --> 00:08:29.217
And the game has changed.

s88
00:08:29.217 --> 00:08:44.857
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 as SaaS.

s89
00:08:44.857 --> 00:08:47.017
But, now the opposite is true.

s90
00:08:47.017 --> 00:08:56.617
Anything can be reverse engineered and there are many, many motivated open source builders pointing their agents at these different softwares.

s91
00:08:56.977 --> 00:09:03.297
If the close competitors do manage to make interesting features, open source will have it not long afterwards.

s92
00:09:03.297 --> 00:09:06.497
The motes of yesterday were in proprietary IP.

s93
00:09:06.497 --> 00:09:11.817
The motes of tomorrow are in network effects, reputation, and reliability.

s94
00:09:12.977 --> 00:09:25.137
What's funny is that in a lot of ways the labs have always seemed to underestimate their own tools or perhaps they are underestimating the users of those tools and their genius and their motivation.

s95
00:09:25.337 --> 00:09:37.177
This happened and is happening on the model side and we'll talk about why that's important, but it turns out that models can fairly easily teach another model to be, you know,

s96
00:09:37.177 --> 00:09:40.617
nearly as smart as the original model.

s97
00:09:40.617 --> 00:09:48.977
And the advantage in a model is very rarely in the architecture or some novel system that was created in a proprietary way for that model.

s98
00:09:48.977 --> 00:09:53.817
In fact, it's usually in the quality of the data and the scale.

s99
00:09:54.497 --> 00:09:58.857
And sorry.

s100
00:10:05.497 --> 00:10:08.217
The uh

s101
00:10:13.138 --> 00:10:26.658
The The model started to The model company started to realize that the you know, scale was catching up and that you could create these things with supplementary quality data

s102
00:10:26.658 --> 00:10:29.538
by teaching one model from another.

s103
00:10:29.538 --> 00:10:37.898
So, they labeled the teaching process distillation and they labeled illicit distillation theft.

s104
00:10:37.898 --> 00:10:48.178
You cannot call that copyright infringement because copyright infringement would imply that you somehow took every living human being's written or creative work and then trained on top of it.

s105
00:10:48.178 --> 00:11:00.738
They also labeled the process of They also trained the models to not assist in the process of training other models and they labeled that safety.

s106
00:11:06.338 --> 00:11:19.178
The next thing that you need to know about agents and about the new world is that agents are like operating systems and they're like frameworks and they're a lot more like those than they are like chatbots.

s107
00:11:19.178 --> 00:11:28.018
And so, if you continue to think about your agent as merely a personal assistant you will forever live on the surface of what it means to engage in agentic systems.

s108
00:11:28.018 --> 00:11:30.778
And so, what do I mean by that?

s109
00:11:31.178 --> 00:11:39.178
If you guys were here a little bit earlier, there was a presentation about a financial firm and how they think about AI and these different inputs that go into

s110
00:11:39.178 --> 00:11:42.658
ultimately what becomes an investment decision.

s111
00:11:42.938 --> 00:11:48.778
Agents are really at their heart a system that combines determinism and non-determinism.

s112
00:11:48.778 --> 00:11:56.698
If you remember from that talk, there are data sources, there are routine code, there are people's opinions and there's judgment.

s113
00:11:56.698 --> 00:12:01.058
And all these things come together to create what is the new application.

s114
00:12:01.058 --> 00:12:07.658
The new application is no longer necessary that it is deterministic and solid and cold hard code.

s115
00:12:07.658 --> 00:12:17.139
It can inject intelligence, reasoning, judgment, all of these things together to combine to create the new application.

s116
00:12:17.139 --> 00:12:29.939
And so, as you at your company think about building these new applications, you should be building these systems on foundations that you own.

s117
00:12:29.939 --> 00:12:32.419
You should be able to swap in and out the models that you want.

s118
00:12:32.419 --> 00:12:48.859
You should be creating ever more sophisticated towers of functionality on top of these core primitives of determinism, non-determinism, code, data sources, document, reasoning, intellect.

s119
00:12:50.099 --> 00:12:56.939
But, you should not be spending your entire lives being intelligence companies.

s120
00:12:56.939 --> 00:12:59.059
What has made you successful this far?

s121
00:12:59.059 --> 00:13:01.179
Most of you are most likely not agent companies.

s122
00:13:01.179 --> 00:13:04.739
Most of you are not likely AI companies.

s123
00:13:04.739 --> 00:13:07.139
You should be focused on the things that you do well.

s124
00:13:07.139 --> 00:13:17.179
And when the state of the art changes and it evolves and what it means to be an agentic system starts to shift, you should inherit that, not rebuild it and maintain it.

s125
00:13:17.179 --> 00:13:19.379
Remember, these are tools that should make you better.

s126
00:13:19.379 --> 00:13:21.779
They are not drop-ins to replace you.

s127
00:13:21.779 --> 00:13:26.259
Do you know how much energy it takes to create a human?

s128
00:13:27.739 --> 00:13:35.459
And the beauty of the Hermes Agent proposition is that you can run it however you want, wherever you want, modify it as you see fit.

s129
00:13:35.459 --> 00:13:39.699
If you're not having fun anymore, you can take your ball and you can go home.

s130
00:13:39.699 --> 00:13:41.699
The moat has changed.

s131
00:13:41.699 --> 00:13:44.419
The moat is no longer entrapping you.

s132
00:13:44.419 --> 00:13:47.179
The moat is in reputation.

s133
00:13:47.179 --> 00:13:49.659
And that's what Hermes Agent offers you.

s134
00:13:49.659 --> 00:13:54.379
And so, what do I want out of all of this?

s135
00:13:54.379 --> 00:13:56.219
Well, I want to build a standard.

s136
00:13:56.219 --> 00:14:01.779
Hermes Agent today is the most used and downloaded open source agent in the world.

s137
00:14:01.779 --> 00:14:07.299
It is order of of magnitude more downloaded than some of its close competitors.

s138
00:14:07.619 --> 00:14:11.339
And I want everybody to be able to build on these open foundations.

s139
00:14:11.339 --> 00:14:20.541
Because as long as one of us us is willing to open source these tools and these agentic systems, water will continue to flow downhill.

s140
00:14:20.541 --> 00:14:28.781
And I want to be downhill so that I can work with you to provide support, safety, scalability to go along with the world's best open agent.

s141
00:14:28.781 --> 00:14:36.581
I've seen where this goes and we've seen where the motes will lie, network effects, reliability, and reputation.

s142
00:14:37.501 --> 00:14:42.301
People are going to continue to solve hard problems with agentic systems.

s143
00:14:42.301 --> 00:14:46.781
People are going to solve lucrative problems with agentic systems.

s144
00:14:46.781 --> 00:14:50.861
Somebody is going to earn back the trillion-dollar investments that we've made.

s145
00:14:50.861 --> 00:14:52.421
Make sure that it's you.

s146
00:14:52.421 --> 00:14:55.461
Make sure that you are in control of your own destiny.

s147
00:14:55.461 --> 00:14:57.141
You can have your cake and eat it, too.

s148
00:14:57.141 --> 00:15:02.221
You can have state-of-the-art features and reliability without sacrificing your autonomy.

s149
00:15:02.221 --> 00:15:03.621
Build on something you own.

s150
00:15:03.621 --> 00:15:04.621
Build with us.

s151
00:15:04.621 --> 00:15:06.530
Thank you.

s152
00:15:06.530 --> 00:15:08.530
[applause]