Tao warns AI is collapsing the open science model

The concern here isn't just about individual researchers hoarding ideas. It points to a second-order problem: if the most productive researchers respond to AI by going dark, the public training data and published literature that future AI systems depend on gets thinner and lower-quality over time, creating a feedback loop that degrades the very tools accelerating the hoarding behavior.

The Platformer piece from the same day, 'The AI warnings are coming from inside the lab,' covers a related credibility problem: when warnings about AI's structural effects come from parties with skin in the game, the signal gets muddied by institutional incentives. Tao's concern lands differently because he has no product to protect, which gives it more weight than the lab-internal warnings Platformer describes. Both stories, taken together, suggest the field is entering a period where the people best positioned to diagnose AI's systemic risks are either compromised by proximity to capital or, as in Tao's case, raising alarms that don't fit neatly into any lab's preferred narrative.

Watch whether major preprint servers like arXiv report measurable submission slowdowns in high-activity ML subfields over the next two to three quarters. A sustained dip in early-stage idea sharing, rather than just finished papers, would be the first concrete evidence that Tao's described dynamic is already underway.