Timothy Gowers and Peter Sarnak, two of mathematics' most respected voices, have drawn a sharp distinction between what large language models can and cannot do in their field. LLMs excel at calculation and recombination of existing mathematical methods but fail at the intuitive leaps required for genuine mathematical innovation.
This assessment cuts against the hype around AI's mathematical capabilities. While models demonstrate competence in executing known techniques and combining established approaches, they lack the creative spark needed to discover fundamentally new mathematical ideas. Gowers and Sarnak highlight a critical gap between computational strength and mathematical insight.
The distinction matters. Calculation is mechanical. A model can follow logical steps, verify proofs, and apply standard techniques with reliability. But creativity in mathematics requires something different: the ability to sense when a problem might yield to an entirely new approach, to recognize patterns invisible to standard methods, to build intuition about what questions matter and why.
LLMs train on patterns in existing mathematical literature. They absorb what humans have already written and learned. They cannot, by their current design, generate the kind of original intuition that leads to breakthroughs. They cannot feel when a conjecture is likely true or sense the shape of an argument before it exists on paper.
This assessment aligns with observations from other domains. Models are reproducing tools, not discovery engines. They amplify human capability but don't replace human insight. For mathematics, where the value lies in proving previously unproven theorems and developing entirely new frameworks, this limitation is fundamental.
The implication is clear: expect LLMs to become increasingly useful as tools for mathematicians. They can handle grunt work. They can verify steps. They can explore variations on known problems. But the mathematician still needs to provide the creative direction, the intuition about which problems matter, and the original thinking that marks genuine contribution to the field.
Gowers and Sar