Unverified50% confidenceFactExact time
Under OSV conditions, standard RLVR reinforces unfalsified positive samples, potentially rewarding shortcut behavior and causing the policy to learn solutions that superficially pass checks but are actually wrong.
1
Sources
50%
Confidence
Long-term
Relevance
11/13/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedRLVR directly uses programmatic verification of answer correctness as the reward signal, requiring no human annotation, and is primarily applied to tasks with objective answers such as mathematics and coding.75% similarUnverifiedIn RLVR, the reward signal provided by Lean is binary and tamper-proof, stricter than answer-matching-based checkers.71% similarUnverifiedRLVR typically adopts group-relative policy optimization methods such as GRPO, sampling a group of rollouts for the same prompt, using an outcome verifier to give 0/1 rewards, and then updating the policy with the standardized within-group reward as the advantage function.69% similarUnverified对抗 RL 的鲁棒性评估应在目标部署场景中重新验证,而非盲目照搬既有结论68% similarUnverifiedThe experiments cover two model scales and six RLVR objectives, and the generality of the conclusions still requires broader validation.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1166009API
curl https://kongchang.com/api/v1/knowledge/claims/1166009MCP
get_claim(id=1166009)