Skip to content

Comment on Ask HN: How is the community using LLMs for data cleaning/enriching/structuring?parent

Comments

Everything I do now is classification and AUC-ROC is my metric. For your problem my first thought is an up-down accuracy metric, but the tricky problem you might have is "do you accept both 'United States' and 'USA' as a correct answer?" and the trouble dealing with that is one reason I stick to classification problems.

I'm skeptical of any claim that "A works better than B" without some numbers to back it up.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.