Unverified50% confidenceFactExact time
Beam introduces a controllable length penalty into the reinforcement learning reward function, penalizing unnecessary token generation while rewarding successful task completion.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedReward shaping refers to adding extra, denser guiding reward signals on top of the original sparse reward to help agents learn faster and more correctly.69% similarUnverifiedPotential-based reward shaping guarantees that the optimal policy remains unchanged.66% similarUnverified该项目当前的奖励机制围绕鼓励向前移动、鼓励保持直立、惩罚不稳定性、惩罚过大动作四个方向设计64% similarUnverifiedCheckpoint partial rewards densify sparse signals by setting milestone-style rewards at intermediate nodes in long tasks, helping gradient propagation.64% similarUnverifiedReward engineering is essentially a highly iterative process: design the reward, observe behavior, discover loopholes, correct the reward, and repeat.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1070561API
curl https://kongchang.com/api/v1/knowledge/claims/1070561MCP
get_claim(id=1070561)