Just links

torch.compile, the missing manual You are here because you want to use torch.compile to make your PyTorch model run faster. torch.compile is a complex and relatively new piece of software, and so you are likely to have growing pains. This manual is all about how to resolve problems that may arise...

Scalable oversight protocols aim to enable humans to accurately supervise superhuman AI. In this paper we study debate, where two AI's compete to convince a judge; consultancy, where a single AI...

Reinforcement Learning (RL) has achieved impressive results on complex tasks but struggles in multi-task settings with different embodiments. World models offer scalability by learning a...

Large language models (LLMs) have brought a paradigm shift to the field of code generation, offering the potential to enhance the software development process. However, previous research mainly...

How did humanity coax mathematics from the aether? We explore the Platonic view that mathematics can be discovered from its axioms - a game of conjecture and proof. We describe Minimo (Mathematics...

We consider a refined version of the string-net model which assigns a different energy cost to each plaquette excitation. Using recent exact calculations of the energy-level degeneracies we...

PoliFormer: Scaling On-Policy RL with Transformers Results in Masterful Navigators.

The news these days feels apocalyptic to me—as if we’re living through, if not the last days of humanity, then surely the last days of liberal democracy on earth. All the more reason to ignore all of that, then, and blog instead about the notorious Busy Beaver function! Because holy moly, what news have I got today. For lovers of this super-rapidly-growing sequence of integers, I’ve honored to announce the biggest Busy Beaver development that there’s been since 1983, when I slept in a crib and you booted up…