Intelligence that improves its ability to improve.

An AI research company.

On a mission to solve humanity’s hardest problems.

AI already improves itself where it can check its own work. A coding agent learns within seconds whether its code works, and gets better with every attempt.

The decisions that matter most give no such answer. In investing, medicine or policy, a choice is judged months or years later, and the result depends partly on luck.

We are making recursive self-improvement work on these problems.

First application

Our agents already beat the market.

We start with trading. Markets record the result of every decision, and no human has to judge it. Our agents make thousands of decisions each quarter. Market data is also public and timestamped, so we can replay any past period exactly as it was known at the time and test an agent on months of decisions in hours.

Every week, the system builds a better version of itself, from its agents' instructions down to the model weights.

Infinite Ascent +12.6%S&P 500 −1.0%