Search papers, labs, and topics across Lattice.
2
0
5
Stealth biases in language models can be reliably detected using a novel distillation technique that amplifies hidden signals, transforming bias detection into a practical tool.
FAPO achieves a remarkable +33.8 percentage point gain in performance by seamlessly transitioning from prompt optimization to structural adjustments in LLM pipelines.
