Unverified50% confidenceFactExact time
Temperature 0 does not equal deterministic output; three runs of the default model under identical conditions yielded 67.6%, 75.7%, and 83.8% respectively, meaning a single leaderboard figure carries approximately ±8 percentage points of noise.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/11/2026
First Seen
Valid until: 1/9/2027
Sources
Related Entities
Related Claims
Unverified自洽性做法是对同一问题进行多次高温采样（temperature通常设为0.7-1.0），再用多数投票选出频率最高的结果77% similarUnverified自洽性采样的temperature通常设为0.7-1.074% similarUnverified复现官方基准分数时，温度从0.0调到0.6可能导致基准分数波动5-15个百分点73% similarUnverifiedUnder mild conditions (18°C, 1,200 kW), only 9 inputs accounted for 90% of the impact in the same compiled program, and the measured derivative of the fan target temperature was 0.0.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/999224API
curl https://kongchang.com/api/v1/knowledge/claims/999224MCP
get_claim(id=999224)