Expired80% confidenceFactTime unknown
Even with an LLM's temperature set to 0, different inference batches, hardware environments, or API versions may still lead to output differences.
10
Sources
80%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
Related Claims
Unverified即便将temperature参数设为0（贪婪解码），也仅能保证在单次模型版本内的确定性，模型权重更新、CUDA内核版本差异、硬件浮点运算精度差异都可能引入输出漂移78% similarUnverifiedThe basic logic of a thermal covert channel is that a CPU generates heat under high load, which can be used to represent 1 with high temperature and 0 with low temperature, encoding binary information into temperature changes to be read by a nearby machine capable of sensing temperature variation.69% similarUnverifiedThe key mechanisms that produce probabilistic outputs in the Transformer architecture are the temperature parameter and sampling strategies; when the temperature is set to 0, the output tends toward the highest-certainty choice but still cannot be guaranteed to be identical to the previous output.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/55169API
curl https://kongchang.com/api/v1/knowledge/claims/55169MCP
get_claim(id=55169)