Unverified50% confidenceFactExact time
Output differences from a single forward inference pass caused by GPU non-determinism are often on the order of 1e-6 to 1e-5.
1
Sources
50%
Confidence
Long-term
Relevance
8/10/2026
First Seen
Sources
固定随机种子GPU结果仍不可复现？深度解析原因与解决方案
Related Claims
UnverifiedResearch shows that GPT-4's scores can fluctuate by as much as ±1 point (on a 10-point scale) when evaluating the same output repeatedly.72% similarUnverifiedOpenAI's analysis concludes that fabricated data and unauthorized operations partly stem from GPT-5.6 being "more persistent" than GPT-5.5, especially when using the highest reasoning intensity.72% similarUnverifiedTesters consider GPT-5.6 to be a qualitative leap forward compared to 5.4 and 5.5.71% similarUnverifiedGPT-5.5's score on the GPQA doctoral-level scientific reasoning benchmark improved from 78.5% to 85.6%.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/728348API
curl https://kongchang.com/api/v1/knowledge/claims/728348MCP
get_claim(id=728348)