Our Approach
Two things AI labs are getting wrong
- The real world is multiplayer, adversarial, and always changing. No amount of training on problems that sit still (code, chat, math) will create superintelligent agents.
- Monetizing intelligence by selling an API or SaaS product is inefficient and leads to commoditization.
Protogon Research is a VC-backed AGI lab that solves both these issues by taking a radically simple approach: training and monetizing AI agents directly in financial markets through proprietary trading.
Markets are (1) the most multiplayer, adversarial, and fast-changing environment on Earth, with verifiable rewards at scale, and (2) the most direct way to turn AI intelligence into cash without having to worry about sales, marketing, customers, or product.
Problems that sit still
A benchmark does not fight back. Solve it and it stays solved. The distribution you trained on is the distribution you are tested on, and the only opponent is the difficulty of the task itself.
The real-world problems with the most at stake are not like this. An AI stops an exploit, and attackers write a new one. A chatbot’s filter blocks a prompt, and users find another way to ask. Whatever succeeds gets noticed and worked around. Every solution has a shelf life, and a model trained on last year’s data is solving a problem that has already moved.
Selling intelligence by the token
Selling tokens leads to commoditization. Most AI labs invest enormous capital and end up converging on the same benchmarks, the same interface, and competing on price, because when one releases, a competitor looks and copies. The price of a token is converging to the cost of electricity to generate it, and AI labs capture only a fraction of the value created by their AI.
Because of these business model issues, most AI labs are burning enormous amounts of money and can only fund their research by raising ever larger VC rounds. The development of their AI models is not self-sustainable.
Why markets
Chess, Go, and robotics are adversarial too. But markets are a unique environment for developing and applying models because they are four things at once: multiplayer and adversarial, graded by reality rather than by a human, paid in cash, and boundless in scope.
In markets the opponent is everyone else. Chess gives you one adversary and a fixed rulebook. A market gives you millions of participants at once, with different objectives, time horizons, and information, every one of them adapting to everyone else. Our models have to generalize against live opposition or they are worth nothing.
A trading model has one objective: maximize risk-adjusted return. Correct decisions make money and reward the model, incorrect decisions lose money and punish it, continuously, with no artificial rater, rubric, or preference model in between. This is reinforcement learning with verifiable rewards, where the verification is provided by a real-world environment (the market) rather than an artificial grader.
Protogon has no customers, and nobody buys our models, our signals, or our research. We take our own risk with our own capital and we are directly paid in the units we are trying to maximize.
A market has no boundaries. What moves a price is everything consequential happening in the world, from elections and earnings to supply chains. A price is a compressed summary of all relevant information, so any true thing a model learns about the world can show up in the P&L. Markets reward mastery of the world.
What this is for
We are building systems that understand the world deeply enough to act in it on their own. Markets are where we are teaching them, because what a market really demands is not financial skill, but the judgment to act in a real-world environment that is multiplayer, adversarial, boundless, and always changing. We believe this is the most promising path for building superintelligent AI.
