Unverified50% confidenceFactExact time
The paper gives an exact two-reward condition under which the reward with the lighter weight is inverted in the equivalent objective.
1
Sources
50%
Confidence
Long-term
Relevance
11/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe theoretical analysis of WAS-PO shows that even if all reward weights are positive, the effective weight of the whitened aggregate objective on certain original rewards may still be negative.70% similarUnverifiedThe paper applies inverse probability weighting to the sampled targets, obtaining a conditionally unbiased single-branch gradient estimate.67% similarUnverifiedThe cost-per-correct-answer metric simultaneously penalizes both extremes: high failure rate but cheap, and high success rate but expensive.64% similarUnverifiedIn the synthetic experiment, "always output RISK_LEVEL: 1" achieves an ordinary accuracy as high as 0.922, but an average reward of −0.000, showing that class-balanced rewards can prevent laziness.64% similarUnverifiedWhen the within-group reward distribution is severely skewed, the mean is pulled off by extreme values, causing distorted advantage estimates; in such cases, the median or a trimmed mean may be more robust.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1130533API
curl https://kongchang.com/api/v1/knowledge/claims/1130533MCP
get_claim(id=1130533)