News Score is my antidote to algorithmic spam and attention harvesting. Most feeds are tuned to keep you scrolling, with clickbait, sensationalism and bias. News Score pulls stories from sources such as Hacker News, Techmeme and Reddit, and uses language models to do three things: score each story for importance, sort the day’s news by that score, and rewrite the best headlines to be dense, factual and free of clickbait. The scoring is strict: fewer than one story in twenty should score above 7. A leaderboard ranks sources by their average score, and anyone can submit a link.

I built it in May 2023 on llm.js. It has run ever since, and I use it myself. It picks up the idea behind Neural FM: a feed scored by a model you control, not by a platform.