Unverified50% confidenceCautionExact time
多数LLM的自报置信度存在过度自信问题，使用前需进行温度缩放等校准处理
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
Related Claims
Unverified即便temperature设为0，LLM在长对话上下文中仍可能出现幻觉或遗漏必要披露，不宜依赖模型生成强监管行业的合规话术70% similarUnverified温度超参数τ越小分布越集中，模型对难负样本越敏感，但τ过小会导致梯度对异常值敏感、训练不稳定68% similarUnverifiedSetting the temperature parameter too high in decoding strategies leads to increased sampling randomness, which is associated with hallucination generation67% similarUnverifiedLLM evaluation requires finding a balance between tests that are too loose (letting real issues pass) and tests that are too strict (generating large numbers of false positives), and thresholds often need to be calibrated through experimentation.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/514418API
curl https://kongchang.com/api/v1/knowledge/claims/514418MCP
get_claim(id=514418)