Hacker News new | past | comments | ask | show | jobs | submit
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?