Terence Tao's blog posted a golden-age-or-dark-age essay on AI and math that Tao didn't write

A guest post, not a verdict from Tao

The post is by Francesco Maggi and ran on Terence Tao's blog on October 18. Tao notes it was converted from another file format using AI. Maggi's thesis is that "reality will likely combine these scenarios."

The golden age depends on "the careful integration of LLMs into the various STEM ecosystems": training a new generation of scientists and keeping research open. Maggi is hopeful about "mathematization," inventing frameworks that make phenomena intelligible, which cheaper proofs could spread across the natural and social sciences. The dark age is a short-term surge in output while "independent human scientific activity would progressively weaken," and knowledge becomes "alien to human understanding."

His list of specific worries is concrete:

• Long Lean proofs can be correct yet unreadable, and their natural-language translations get "messier and less faithful" as proofs grow.
• "A strong mathematical result, by itself, can no longer be taken as convincing evidence" of a student's ability, which strains hiring and assessment.
• Mathematicians are becoming "very cautious about sharing ideas," and unpublished proposals uploaded to LLMs could leak.
• Credit and accountability for LLM-generated proofs are unclear.
• Giving a few companies "increasing—and potentially total—control" of frontier tools "seems ill-advised."

His remedies: publicly governed LLM infrastructure, research access to frontier models, disclosure and reproducibility standards for LLM-assisted work, and studies of the effect on science education.

Half the thread assumed it was Tao

"I miss articles actually written by Tao." — bananaflag

Others went further and read the post as Tao's own reversal.

"I love how Terrence Tao went from being AI shiller … to a Oh nO MaThS iS KilLeD wHaT aBoUT mY JoB in a few months." — muadddib

Jacques Mattheij, the engineer behind the Unite micro-kernel OS, was careful to note Tao "didn't write the post but he did post it," then widened the question beyond mathematics: scientific knowledge can be hoarded or used to patent products, and he said there is plenty of evidence that OpenAI and Anthropic will go after other sources of valuable knowledge.

What mathematicians are actually upset about

The founder of MailChannels offered a reading of Tao's recent comments: professionals are mainly upset that AI-submitted math papers lack the teaching content good human papers carry. Professional mathematics, he said, is as much about spreading knowledge as solving problems.

A French researcher and engineer pulled out the secrecy point: Maggi's observation that mathematicians already guard research statements and proposals, since a reviewer might paste them into an LLM, applies to everyone who has never ticked a "do not train on my data" box. One commenter bet on "a short golden age" followed by a dark age in which nobody wants to publish new ideas, and another said he had no sympathy: the field was gatekept and snobbish, and seeing mathematicians "taken down a peg or a hundred is somewhat gratifying."

The thread's lesson is about authorship. Attribution on a blog post decided which argument readers thought they were having, and Maggi's own point, that credit for machine-assisted work is unclear, was demonstrated before the thread reached it.