Unverified50% confidenceTradeoffExact time
Ablation experiments show that increasing example usage does not yield consistent gains, but instead presents a trade-off relationship between dynamics and reward.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
Unverified研究者提出长度归一化奖励方法，即在计算奖励分数时除以回答长度，使模型无法通过单纯增加长度获取更高奖励76% similarUnverifiedIn long-trajectory scenarios, retry mechanisms often only increase overhead without genuinely improving the final output.74% similarUnverifiedAs model capabilities improve, the tension between capability and alignment becomes increasingly pronounced, and the consequences of alignment failure are correspondingly amplified.73% similarUnverifiedRaising a model's lower bound (reducing the frequency of stupid mistakes) does not increase potential risk, whereas pushing up the upper bound (improving peak capability) necessarily comes with a rise in potential danger.72% similarUnverifiedSkinner found that when rewards appear at unpredictable intervals (variable-ratio reinforcement), subjects exhibit the most persistent and hardest-to-extinguish behavior.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1001546API
curl https://kongchang.com/api/v1/knowledge/claims/1001546MCP
get_claim(id=1001546)