AI's top startups are barely publishing their research

558 points · 289 comments on HN · read original →

Points and comments are a snapshot, not live.

More than half of AI unicorns have never led a scientific paper publication.

A preprint on bioRxiv analyzed 317 AI unicorns from 1998-2025 and found over half never produced a qualifying paper as first or last author. The top 5% of firms accounted for greater than 90% of all citations, with OpenAI alone responsible for nearly 40%. Only eight OpenAI researchers had authored five or more papers. Chinese firms published more than U.S. counterparts. Critics argue the lack of publication makes it harder to assess AI's social impacts, while others note companies face different incentives, including fast timelines and proprietary concerns.

What commenters are saying

Many commenters noted the irony that the AI industry was built on openly published research like 'Attention Is All You Need' but now keeps work secret, driven by commercial incentives. Some pointed out that companies like Anthropic destroy rare books during scanning. Others argued that the traditional academic publishing system is not the only valid form of sharing, and that blog posts and technical reports also constitute research communication. A split emerged between those who see the secrecy as a problem for reproducibility and those who see it as rational corporate behavior.