Within a decade, the largest player in financial markets won’t be a hedge fund, it will be an AI research lab. It will feel more like OpenAI than Citadel, but with a key difference: it will train agents that learn directly from markets.
We started Nof1 because no one else was taking this seriously enough.
Markets compress all economically relevant information about the world into crisp, measurable reward signals. They are, in some ways, the ultimate test of intelligence, and the perfect environment to train never-before-seen types of agency.
At Nof1, we’re building the first algorithmic trading system that understands the world. It will have the speed and ability to take speculative action, like other systematic trading platforms, but will learn from the consequences of those actions, generalize its learnings to novel environments, evolve its own trade ideas, and carefully manage its own risk.
Markets are a uniquely favorable environment for open-ended learning algorithms, where agents create their own training data, indefinitely. Over time, they must build their own world model, with applications that extend beyond finance. We call it reinforcement learning from market feedback (RLMF).
We’re hiring aggressively from places like DeepMind, OpenAI, Renaissance, and academia to build the world’s best team at the intersection of AI and finance.
The models being used in markets today are broken.
Why? Because algorithmic hedge funds are obsessed with speed. For two decades, it was the only metric that reliably scaled profits, triggering an arms race for microsecond-level edge.
This speed-at-all-costs paradigm produced shallow, reactionary models that distort markets and cap potential returns.
The core challenge is that markets are complex and flooded with data, most of which is irrelevant. Until recently, we couldn’t build models smart enough to separate signal from noise, speed was all we had.