Skip to content

Comment on Using reinforcement learning and $4.80 of GPU time to find the best HN postparent

Comments

Yeah my Bluesky followers are carefully curated to stop from swelling into negativity. I've been playing around with a labeller that filters followed posts into those that I find emotionally pleasant which I've been training based on my own labeling of followers' posts. The goal is to follow more people and have the labeller (or feed generator depending on how I go) hide the posts I don't care for.

AboutSource Built by g1lg1l

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