Science: AI's top startups have mostly stopped publishing

Science has a survey of publication output at the frontier labs, and the finding is the one the field has been muttering about for a couple of years: the best-funded AI startups, staffed largely by people who came out of academia, are barely publishing their research anymore. Model cards, system cards and blog posts have quietly replaced papers.

#Why it matters

The practical consequence is that the artifacts external researchers need — methods, training details, evaluation protocols — arrive filtered through marketing, if at all. That makes independent replication and safety auditing dependent on whatever the lab chooses to disclose, right when the capability curve is steepest. It also compounds the "PR-as-publication" problem: benchmark claims land with no reviewable methods section behind them.

The Hacker News thread (231 points, 129 comments) mostly divided into two camps: those who read it as straightforward competitive secrecy once research became the product, and those noting that safety rationales have become a convenient cover for not showing your work.

Sources: Science, HN discussion