Search papers, labs, and topics across Lattice.
6
0
11
Calibration rankings of LLMs can flip dramatically when accounting for accuracy, challenging long-held beliefs about model performance.
PACT achieves state-of-the-art performance in medical dialogue systems by leveraging a unique combination of multi-paradigm synthesis and consensus training, all while safeguarding patient data.
Existing hallucination detection methods are missing subtle, word-level medical errors, but a new data-centric pipeline and detector closes the gap by 15%.
You can now reliably pick the *most* aesthetically pleasing image from a subtly different series, thanks to a new dataset and framework designed specifically for fine-grained image aesthetic assessment.
Clinical question answering gets a boost: TARSE aligns language model reasoning with clinically valid logic by retrieving and adapting to relevant skills and prior reasoning experiences at test time.
Get 2x-47x faster on-policy distillation by focusing only on the output prefix, without sacrificing performance.
