Terence Tao warns that blind optimization of problem-solving is damaging mathematics

Terence Tao, a Fields Medalist and professor at Caltech, took the stage at the "Math 2.0" lecture and laid out a warning that cuts against the prevailing current: the new ability of models to crack open problems is not cause for celebration but a threat to the field's long-term health. The narrow chase for solutions — the very thing large language models are beginning to excel at — misses what actually drives mathematical progress.

The lighthouse is not the destination

Tao used the metaphor of a lighthouse to explain the role of open problems in pure mathematics. In his view, such problems are not targets to be conquered and checked off; they are signposts that steer researchers toward the mathematical landscape surrounding them. The lessons accumulated in the attempt to solve them, even when the attempt fails or yields only partial progress, are often worth more than the final solution itself. When automated tools reach the lighthouse prematurely, they disrupt the exploration of paths not taken and sterilize the surrounding field.

Blind optimization as ongoing harm

Tao's phrasing was sharp: "Blind optimization of problem-solving alone is now actively harmful to the long-term health of mathematics." He rejected the widespread perception — and, in his words, some of the community's own self-marketing — that mathematicians are focused solely on solving open problems. In practice, daily work consists of building theories, developing tools, and discovering connections between fields; solving the problem is merely a byproduct of that process.

One-off solutions versus renewable knowledge

The phrase "non-renewable fashion," which recurred in his remarks, points to the structural problem: when a model emits a solution without exposing the path, the knowledge cannot be recycled. Other researchers cannot learn from the move, generalize it, or apply it to neighboring problems. The result is an accumulation of isolated answers without the theoretical infrastructure that would let them become working tools. Tao linked this directly to the safe transfer of mathematics into "dirtier" real-world applications: without the path, the solution is brittle.

Tao's standing gives the argument weight

Tao is not a generic AI skeptic; he is one of the few mathematicians tracking model capabilities closely and integrating them into his own work. His warning comes from someone who knows the current frontier of the tools and knows what mathematics looks like when it is done right. The message is not "don't use AI," but "don't let the problem-solving benchmark dictate the agenda." Anyone who listens only to the final result loses the mathematics itself.