zlacker

Is it just me, or does the article not show how posts were classified as AI vs crypto? I see that they "use an LLM" but don't see a code snippet describing how, or an attempt at assessing error rates, which seems like a basic step in validating this. Maybe I missed it?

edit: Looks like this is the notebook used in the article: https://github.com/corbt/hn-analysis/blob/main/analysis.ipyn...

The classification seems pretty fraught; one cell shows a sample of articles classified as crypto, which appears to include a bunch of cryptography and other unrelated articles.