A mathematics undergraduate asks whether rapid progress in AI has made a research career pointless. In a guest essay on Terence Tao’s blog, Álvaro Lozano-Robledo answers with qualified optimism: nobody knows the future, but producing a proof and developing human understanding are different goals.

His case for continuing to study mathematics rests on three ideas:

  • Collaboration, not passive prompting. In his experience, useful work with language models depends on the mathematician contributing guidance and intuition—not simply requesting a finished proof.
  • More results create work to understand and teach them. He reports that his own research projects have tripled and that he is recruiting more student collaborators. That is his experience, not a forecast for the whole profession.
  • New combinations may not be new concepts. He explores a tentative model in which AI connects existing ideas while humans expand the underlying repertoire. This is a hypothesis about limitations, not an established ceiling on what models can do.

The useful distinction is between learning to understand mathematics and racing to produce answers. Lozano-Robledo argues that a PhD should serve the former; he also acknowledges that reassurance sounds different from the security of a tenured position. His optimism is a reason to investigate AI-assisted research, not a guarantee of employment.

The 229-comment thread on Hacker News adds objections that matter especially to people deciding whether to begin.

What the thread adds

  • reasonableklout — raises a knowledge-sharing risk beyond faster proofs: “Some observations that publishing of research is drying up because results can be easily retrieved at any time. Over the long-term, I wonder whether this will result in accumulation of knowledge grinding to a halt”. This is a concern drawn from a blog comment, not independently verified evidence of a trend.
  • analog31 — challenges the reliability of faculty reassurance from their account of physics careers: “The one constant during this time was the perpetual optimism of the faculty for the employment prospects of PhDs.”
  • mikestylz — asks for advice that accounts for continuing improvements rather than just today’s capabilities: “At a minimum a discussion like this should acknowledge the possibility that the current rate of AI progress continues apace.”
  • huitzitziltzin — supplies an optimistic counterpoint from their reported use of models: “The models are most useful and most productivity-enhancing in the hands of experts and in the area of their expertise.”

The unanswered question: can models originate the next ideas?

Two commenters approach the gap from different directions: pks016 asks whether an LLM could propose unique research problems; keithluu wonders whether current leading models could have come up with the incompleteness theorems. The essay’s tentative picture of AI combining existing ideas does not resolve either question. Its afterword explicitly leaves mathematicians time to examine the new proofs for genuinely new ideas.

HN handles are pseudonymous, and the site exposes no per-comment scores. Ordering is HN’s own ranking; this is a slice of the discussion, not a consensus.