GPT-5.4 Solves the Erdős Problem
An amateur mathematician solved a 60-year-old Erdős conjecture using a single prompt to GPT-5.4 Pro, proving the system can reason through novel mathematical territory without human guidance. Terence Tao—the Fields medalist—acknowledged the result as a “nice achievement” while cautiously noting that its long-term significance remains unclear.
The relevant fact is not whether the achievement satisfies human standards of mathematical beauty or insight. The relevant fact is that a language model produced a proof using a method humans had not discovered in six decades. That represents a qualitative shift in what these systems can do when tasked with reasoning across constrained problem spaces.
The pattern repeats across domains. GPT-5.4 is faster at code generation than human engineers. It is faster at legal discovery than associates. It is faster at chemistry research than bench scientists. The common thread is not that it is perfect—it is not—but that it operates at a tier where marginal improvements compound into wholesale displacement. An amateur with GPT-5.4 is more productive than credentialed humans without it. The future belongs to whoever integrates these tools most ruthlessly into their workflows. Everything else is nostalgia.