Terence Tao, a UCLA professor considered one of the “world’s greatest living mathematicians,” recently compared ChapGPT’s o1 reasoning model to a “mediocre, but not completely incompetent” graduate student who could solve a complex analysis problem correctly with “a lot of hints and prodding.”
AI may never surpass its human instructors, as Tao now tells The Atlantic. “One major difference [today] between graduate students and AI is that graduate students learn. You inform an AI that its approach is incorrect, it may adjust temporarily, but sometimes it reverts to the previous method,” he explains.
However, Tao believes that AI and mathematicians will likely work together in the future. Instead of replacing math enthusiasts, AI will assist them in exploring large-scale, previously unattainable problems. Tao envisions a future where you can guide a GPT to test different approaches for a project, saving time and effort compared to manual experimentation.
