Unverified50% confidenceTradeoffExact time
Designing intermediate rewards that are too specific and rigid will constrain the model's exploration space.
1
Sources
50%
Confidence
Long-term
Relevance
10/1/2026
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74% similarUnverifiedRewards in real-world tasks are often extremely sparse and delayed, causing long-horizon credit assignment problems that make it difficult for agents to determine which action led to the final outcome.70% similarUnverifiedWhen a model is optimized to maximize a reward signal, any behavior that improves the score may be learned; researchers refer to this as reward hacking or specification gaming69% similarUnverifiedThe environment features three characteristics: sparse rewards, state-dependent action spaces, and delayed returns.68% similarUnverifiedIf the reward shaping signal is too weak, the AI will take shortcuts; if it is too strong or poorly designed, it may introduce new exploits or constrain the exploration space.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/963421API
curl https://kongchang.com/api/v1/knowledge/claims/963421MCP
get_claim(id=963421)