Search papers, labs, and topics across Lattice.
4
1
7
5
Mixed SFT outperforms next-chunk reasoning RL while consuming over 60 times less compute, reshaping our understanding of effective training strategies with no-CoT data.
PURA achieves over three times the message match rate of existing unbiased watermarking methods, all while preserving text quality and speed.
Turns out, teaching LLMs to *think* like reverse engineers beats just throwing more parameters at the problem of binary deobfuscation.
LLM agent progress increasingly hinges on better external cognitive infrastructure, not just stronger models.
