Louis-Antoine Mullie, MD
About Blog

Blog

Writing

Where Is Recursive Self-Improvement Possible?

Sustained recursive self-improvement depends on the process that generates tasks as well as on the learner. It requires tasks in which each learned abstraction makes the next one cheaper to discover.

DUST estimates the gradient of a blurred loss

DUST pretrains transformers using only forward passes. In expectation, it follows the gradient that backprop would compute for a blurred version of the loss, with credit to later tokens discounted or cut off.

What Would Count as Recursive Self-Improvement?

Claims of recursive self-improvement are cropping up everywhere. But when is what’s measured more than just a better benchmark score?