1 articles

RLHF

Index of editorial tags and filtered coverage.

InstructGPT and RLHF: aligning models with human preferences

OpenAI showed fine-tuning with human feedback yields more useful, safer models than base GPT-3 for instruction use.