Unverified50% confidenceFactExact time
Linear reward functions cannot express certain goals with non-Markovian properties.
1
Sources
50%
Confidence
Long-term
Relevance
9/18/2026
First Seen
Sources
Related Entities
Related Claims
Unverified单次生成而不经过迭代验证的LLM奖励函数几乎不可能第一次就适配复杂的操作任务68% similarUnverifiedThe linear assumption that more parameters means greater capability does not necessarily hold, and blindly stacking parameters may not improve performance.67% similarUnverified任何脱离实际产出价值的优化指标都是没有意义的65% similarUnverifiedPotential-based reward shaping (PBRS) theoretically guarantees that shaping does not alter the functional form of the optimal policy, but does not cover the issue of magnitude.64% similarUnverified专项优化不保证对所有任务类型的普遍提升，在特定领域可能反而不及通用基础模型63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/931280API
curl https://kongchang.com/api/v1/knowledge/claims/931280MCP
get_claim(id=931280)