Unverified50% confidenceOpinionExact time
For novel inputs that annotators have never seen, the judgment of RLHF's reward model degenerates into extrapolation, and the reliability of extrapolation cannot be guaranteed.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedCompared to RLHF, which relies on human feedback, RLVR's verifiable rewards reduce dependence on manual annotation and provide a more objective and consistent optimization signal.73% similarUnverifiedThe preference data that RLHF relies on has systematic flaws such as low inter-annotator agreement, position bias, and length bias.70% similarVerifiedAnnotation bias in RLHF causes models to learn writing templates that please evaluators, a phenomenon known as a variant of reward hacking67% similarUnverified如果传给RLS的上下文本身是错误的，RLS会为错误的主体返回正确的行，因此RLS只是问题B的缓解手段而非完整答案66% similarUnverifiedMTP's verification logic involves dynamic token accept/reject decisions; if the relevant state updates are not correctly reflected in the static graph input tensors, it causes verification results to become invalid and acceptance rates to plummet.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1012126API
curl https://kongchang.com/api/v1/knowledge/claims/1012126MCP
get_claim(id=1012126)