AI's top startups are barely publishing research

What's this about?

Science reports that the leading AI labs have sharply cut their peer-reviewed output, keeping methods internal for competitive and safety reasons.

What exactly is "peer-reviewed research"?

Publishing means writing up how something works so outside experts can scrutinize and reproduce it. It's how a field checks claims and builds shared knowledge. When labs stop, outsiders can't verify what the models actually do or how.

What's the community saying?

1. Verifiability worry — without papers, capability claims can't be independently checked.
2. Recognition that secrecy is rational when the research *is* the product.
3. Concern the academic pipeline starves if the best work never leaves the labs.