Re:Claude Opus 4.8
by atq2119 on Thursday May 28, 2026 @01:58PM
(#48312868)
Unless you're systematically repeating the exact same task, the most parsimonious explanation is that you're seeing natural variation based on different tasks, random sampling of tokens, etc.
[ Parent
| on HN ]
Re:Claude Opus 4.8
by osigurdson on Thursday May 28, 2026 @07:12PM
(#48316835)
I don't think this explains the phenomenon as is more temporal in nature - not prompt to prompt. I'm sure the AI labs gracefully degrade to simpler models when resources are low - why wouldn't they?