Hacker News new | ask | show | jobs
by Der_Einzige 155 days ago
That’s anthropic fault for continuing to use top-K, a stoneage tier shitty sampler. Your own head of mechanistic interpretability invented a better one called tail free sampling in 2019.
1 comments
Majromax 155 days ago
That seems to have nice properties, but 2019 was a while ago. Is the problem of top-k sampling still relevant with much better frontier models?
link
Der_Einzige 154 days ago
Yes, yes, oh god yes.
link