Science: The Top AI Startups Have Nearly Stopped Publishing

Science reports that the best-funded AI startups now publish very little peer-reviewed research, a reversal of the norm that produced the transformer paper and most of the field’s shared vocabulary. The Hacker News thread that followed — 352 points, 199 comments — split between “obviously, it’s competitive advantage” and the observation that the whole industry was built on results its predecessors gave away. The practical cost is not sentimental: without published method detail, external replication, security review and independent evaluation all get harder, and benchmark claims become unfalsifiable by construction.