Search papers, labs, and topics across Lattice.
5
0
11
Domain experts' unaided reasoning explanations are unnaturally bad, but pairing them with a knowledge engineer unlocks dramatically more natural chain-of-thought data for LLM post-training.
LLMs can now be objectively scored on their creative writing ability using a new benchmark that rewards surprising, yet logically consistent, story continuations.
Forget needing massive datasets: just 100 expert-annotated Chain-of-Thought examples can dramatically improve a small LLM's creative quality, thanks to architectural duality.
Multicast, re-imagined for AI, slashes collective communication latency by up to 33% on real-world hardware.
Achieve millimeter-level accuracy in 3D human body fitting from multi-modal inputs, even with scale distortion common in AI-generated assets.
