Unverified50% confidenceBenchmarkExact time
Compared with the unmodified GRPO baseline, MCPO achieves improvements on all five data random seeds, with the macro-average avg@32 improving by 6.55 points on average (95% confidence interval [4.1, 9.0]).
1
Sources
50%
Confidence
Long-term
Relevance
11/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedOn a dataset with 5,000 random features and random labels, if the top 20 features are first selected on the full data and then 5-fold cross-validation is performed, accuracy can reach 85% (87.7% averaged over 20 random seeds); after moving feature selection inside the Pipeline, accuracy falls back to 48% (50.0% average).65% similarVerifiedGRPO eliminates the dependency on an independent critic network by sampling a group of outputs for the same question and using the within-group average reward as the baseline, compressing memory requirements from approximately 4× that of PPO to approximately 2×.63% similarUnverifiedIn the fact-updating task, the success rate of the embedding retrieval method fluctuated between 0.30 and 0.95, with a maximum gap of 0.45 across different random seeds.61% similarUnverifiedIn code review tests, Opus 5.5 improved recall on more difficult pull requests from a baseline of 54% to nearly 77%, but the number of comments increased by approximately 20%.61% similarUnverifiedIn the eight-player configuration, the win-rate gap between gpt-oss and the trained network had a one-sided Fisher exact test p = 0.0017, far beyond random fluctuation.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1157956API
curl https://kongchang.com/api/v1/knowledge/claims/1157956MCP
get_claim(id=1157956)