Search papers, labs, and topics across Lattice.
3
0
4
12
Few features in sparse autoencoders provide consistent, reusable directions for model behavior, complicating interpretability efforts.
GRADE achieves a 4.8-point performance boost on MMLUPro while cutting active compute in half, revolutionizing multi-agent reasoning efficiency.
Forget tedious multi-turn dialogues: Co-FactChecker's "trace-editing" lets human experts directly shape an LLM's reasoning process, leading to higher quality claim verification.
