Unverified50% confidenceFactExact time
智能体的执行具有非确定性，同样的输入可能因模型的随机采样而产生不同输出，因此其恢复机制需要保存完整的执行轨迹和中间推理步骤
1
Sources
50%
Confidence
Long-term
Relevance
8/5/2026
First Seen
Sources
Related Claims
UnverifiedBecause large language models are based on probabilistic sampling, even if instructions are written in the context, the model may still forget or skip certain instructions in complex reasoning chains, which is known as instruction-following instability.74% similarVerifiedReAct 智能体的核心是「推理-行动-观察」循环，LLM 分析问题后决定调用工具，观察返回结果，若不满足则继续推理直到给出最终回复73% similarUnverified上下文回贯缺失指工具执行后的观测结果（observation）没有回写到对话上下文，导致模型每轮接收相同上下文而重复相同决策72% similarUnverifiedA well-aligned model will guide meta-prompts toward compliant creative expression, while an insufficiently aligned model may exhibit unexpected behavior.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/690428API
curl https://kongchang.com/api/v1/knowledge/claims/690428MCP
get_claim(id=690428)