Q

We're a research lab focused on understanding and solving generalization – the core open problem in AI. We believe orders-of-magnitude improvements in sample efficiency are possible.

Our thesis is that generalization should scale optimally with compute alone, even with minimal data. Compute grows much faster than data, so algorithms should exploit that. We're building learning algorithms beyond gradient descent to make this happen. Read more about our research direction.

We're funded1 and hiring researchers to solve these fundamental problems – reach out at research@qlabs.sh

Progress

Decoupling Search and Learning
September 2025
Our first paper on search. We decompose training into tractable evolutionary search in representation space and learning in parameter space.
The Generalization Problem
August 2025
Our core framework for solving generalization. We describe how to combine search and generalization theory to create algorithms that approach optimal learning

[1] Our seed investors include Jeff Dean, Emmett Shear, Mike Knoop, Guillermo Rauch, Qasar Younis, and many other researchers from the top labs.