Mathematics students increasingly question whether to pursue academic careers amid rapid advances in AI language models. Concerns center on whether AI will surpass human mathematical abilities, making academia accessible only to the 'brightest minds,' and whether mathematicians will remain employable. In an October 8, 2026 guest post on Terence Tao's blog, Álvaro Lozano-Robledo addresses these anxieties directly, acknowledging that many students who recently dreamed of math academia now see those dreams as 'distant and, for some, apparently impossible to ever become a reality' due to AI's disruptive potential.
Lozano-Robledo responds to the question of whether AI will surpass human mathematical research ability by referencing the 'convex hull of ideas' hypothesis put forward by Nestor Guillen. He notes that 'none of the proofs so far seem to contain
alien ideas,' a move-37, or completely novel arguments or new concepts that were not present in the literature in some form or another,' explaining that LLMs, trained exclusively on existing human knowledge, would 'think' within the boundaries of our current knowledge and make connections (sometimes surprising and ingenious!) among ideas that are already present in the literature. Using a geometric analogy, he describes current mathematical ideas as a stellated high-dimensional polytope where LLMs fill gaps between outer spikes but require human advances to generate new ideas that expand the knowledge base—using the disproof of the Erdos unit-distance conjecture as an example of work accessible within the convex hull, while suggesting problems like the Riemann hypothesis may require ideas outside this hull.
The article discusses the convex hull of ideas hypothesis, explaining that language models trained on existing knowledge can only form connections within the current boundaries of mathematical knowledge.
The article uses a geometric analogy of a stellated high-dimensional polytope to describe current mathematical ideas, where LLMs fill gaps between outer spikes but human advances are needed to expand the knowledge base.
While expressing personal optimism about AI as a tool that has tripled his research projects in months, Lozano-Robledo includes two important disclaimers: his views may 'age like milk' as the field evolves rapidly, and crucially, 'no one has the answers at this time' to the core questions students face. He warns that confident predictions about AI's triumph often come from those with vested interests in AI companies, particularly before their IPOs (noting Anthropic's November 2026 and OpenAI's early 2027 offerings). Ultimately, he advises students to seek a 'balanced range of opinions' by consulting diverse perspectives, mentioning his own 'Human Mathematicians in the Age of AI' video project as one such resource, while reaffirming his belief that human mathematicians will remain 'front and center' in future mathematical achievements.