How to follow AI news without drowning

The volume of AI news is a denial-of-service attack on your attention. Almost none of it will matter in three months. A routine that actually works:

1. Pick two primary sources, not ten. The live headlines below this post pull from sources chosen for signal density — Simon Willison for hands-on truth, the lab blogs for what actually shipped.
2. Read releases, not takes. When a model launches, read the model card and system card before anyone's hot take. Primary sources are shorter than the discourse about them.
3. Let benchmarks age a week. Day-one benchmark claims have a poor survival rate against independent replication.
4. Track capabilities, not valuations. Funding rounds are business news. What a model can now do that it couldn't last month — that's AI news.

Twenty minutes a week with good sources beats two hours a day with a feed.