This post increases my worries about the revolution we are in the early stages of.
Please, someone correct me in my thinking, since I should hope this Princeton professor of math is more intelligent than I:
He asserts:
Today ML results are not "creative" but come from:
>a search guided by ideas developed by human mathematicians, but far more exhaustive than any search they could carry out unaided.
Therefore:
>There are many risks associated with the uses and misuses of AI. The extinction of mathematics as a human endeavor is not among them.
How does behavior today guard against risks tomorrow? Has he considered the trend from imagenet to today, or from LLMs to today?
Are there any hurdles that prevent against ML models being "creative" (especially as creativity can be thought of as a search process)?
A couple of years ago the goalposts were something like:
"It sounds like the LLM is saying something meaningfull, but it actually doesn't"
Today the goalpost is apparently:
"LLM solved one of the hardest outstanding math problems, but it did it by assuming a constraint. We now, post-hoc, think that was a bad constraint to allow"
Now, I'm not going to assert that human math is obsolete based on this. But asserting that there is no risk of it becoming obsolete based on this evidence does not seem to me to be "intelligent" let alone worthy of ivy league tenure.
This post increases my worries about the revolution we are in the early stages of.
Please, someone correct me in my thinking, since I should hope this Princeton professor of math is more intelligent than I:
He asserts:
Today ML results are not "creative" but come from:
>a search guided by ideas developed by human mathematicians, but far more exhaustive than any search they could carry out unaided.
Therefore:
>There are many risks associated with the uses and misuses of AI. The extinction of mathematics as a human endeavor is not among them.
How does behavior today guard against risks tomorrow? Has he considered the trend from imagenet to today, or from LLMs to today?
Are there any hurdles that prevent against ML models being "creative" (especially as creativity can be thought of as a search process)?
A couple of years ago the goalposts were something like:
"It sounds like the LLM is saying something meaningfull, but it actually doesn't"
Today the goalpost is apparently:
"LLM solved one of the hardest outstanding math problems, but it did it by assuming a constraint. We now, post-hoc, think that was a bad constraint to allow"
Now, I'm not going to assert that human math is obsolete based on this. But asserting that there is no risk of it becoming obsolete based on this evidence does not seem to me to be "intelligent" let alone worthy of ivy league tenure.