Hacker News new | past | comments | ask | show | jobs | submit
codingisfreedom | on: HackerTube: YouTube built from videos recommended by HN
> Opus downloaded all comments of HackerNews, filtered them, ran Sonnet over them in a Workflow (Claude Code feature), got rid of spam, and narrowed everything down to 40K comments recommending a YouTube video.

This part. And Andrej even mentioned he uses medium! I wonder what went into the prompt that prompted the agent to dig deep. Mine seems to stop at very mediocre results fairly consistently and then I just reprompt it a couple of times and it does some back and forth contradicting itself.

I have projects and I made real apps using only LLMs, but it feels tedious. Definitely not a 2 hour thing. And definitely not magical and smooth sadly :(

xpct | parent
Well, not sure what I can add here, because this is exactly the type of semantic classification that I expect LLMs to do well, and can't really say why your agents failed to do it.

I've done similar work simply by giving a model a database of text, and asking it to do a unit of work for every passage. "Checking for spam" would be the model checking surrounding messages and user's history, likely an opinionated verdict, but would work. Difficulty comes in when those units of work become complex or have to work in tandem, which isn't the case here.