Skip to content

Reinforcement Learning from Human Feedback (RLHF)

Definition
AI-generated

Reinforcement learning from human feedback (RLHF) aligns a pretrained language model by training a reward model on human preferences, then optimizing the policy—often with proximal policy updates on sampled outputs—so the assistant is more helpful, honest, and harmless according to labeler rankings.

Topics
LLM and Agents
Synonyms
RLHF
Related
Human in the Loop (HITL) Chain-of-Thought (CoT) Genome-Wide Association Study (GWAS)

Why it matters in GWAS

Instruction-tuned models used for curation or methods drafting inherit RLHF tradeoffs (verbosity, refusals, preference bias); critical genomic claims should still be checked against databases and statistics, not model politeness.

Example usage

"The RLHF-tuned assistant refused to estimate individual disease risk from a VCF paste, which we preferred over a speculative guess."

Human in the Loop (HITL) Chain-of-Thought (CoT) Genome-Wide Association Study (GWAS) Large Language Model (LLM) Pretraining Temperature Transformer
Related-term network

Click any node to open that term page.

Node order. The current term is first. Next come every page linked from Related terms, in the same top-to-bottom order as that list. Then new terms appear in two-hop layers: for each of those direct neighbors (in that same order), their related terms are appended if not already listed. A third pass does the same for two-hop nodes. The walk stops when a fixed maximum number of nodes is reached.

References

  • Ouyang L, et al. (2022). Training language models to follow instructions with human feedback. NeurIPS.
  • Christiano PF, et al. (2017). Deep reinforcement learning from human preferences. NeurIPS.
← Risk Stratification RNA Sequencing (RNA-seq) →

Last updated (UTC · Git history)