LIGHT THINKER

Summary from the TLDR AI newsletter

This work aims to compress really verbose reasoning traces into smaller and more compact representations to save context space while still being effective at steering the model.