He has encountered many graduate students whom he considers lacking in mathematical research ability, according to his standards. As a former math grad student, I would undoubtedly fall into that category. I don’t view this as an insult, just an indication that the bar is set very high.
AI will advance to the point where it can handle mental workloads more efficiently than humans, much like how a truck can transport goods across a country more effectively than a pedestrian.
However, just as trucks need infrastructure like roads, AI requires the internet and massive data centers to function.
When AI reaches this level, the productivity gains will be enormous. Who will benefit from these gains is a different question.
I work in academia, and I would describe ChatGPT’s intelligence as comparable to a competent undergraduate student in the final 1-2 years of their degree.
It’s not brilliant, but it’s far from incompetent.
The AI can handle writing, coding, and data analysis at a high level with minimal supervision, though it struggles with complex tasks even with assistance.
I agree that Tao’s view of mediocre undergraduate students might be somewhat skewed.
“A mediocre but not entirely incompetent graduate student” in mathematics at his elite public university—so essentially smarter than 99.99% of people.