Unverified50% confidenceFactExact time
The team observed that under their current setup, the RL performance curve has not yet shown signs of saturation, and scaling up RL training can still bring observable improvements.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/26/2026
First Seen
Valid until: 11/9/2026
Sources
Related Entities
Related Claims
UnverifiedAfter pre-training scaling laws encountered diminishing returns, RL took over and extended the new performance curve.80% similarUnverified近期研究表明RL（尤其是RLVR）并没有给模型注入大量新知识，而更像是激发和对齐预训练阶段已具备的能力78% similarUnverifiedRL阶段能显著提升pass@1，但对pass@k几乎没有影响，无论预训练规模如何70% similarUnverifiedHigh-leverage points significantly inflate the training-set R² while contributing little to out-of-sample performance, thereby causing traditional shrinkage formulas to overestimate predictive ability.70% similarUnverified强化学习训练比监督学习更不稳定，将NVFP4激进量化引入RL训练会进一步放大固有不稳定性69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1099169API
curl https://kongchang.com/api/v1/knowledge/claims/1099169MCP
get_claim(id=1099169)