I think the argument was that it represents a sub-exponential factorization algorithm, so while he demonstrates with 800, it is ostensibly broken at above that.
Note of course that polynomial time factorization doesn't mean "fast", it means feasible. So what a professor at a university can do (800bits) does not necessarily represent what can be done with someone with more money and computers (NSA, Google, AWS, ...).
Isn't gnfs exponential to logN? although as I type that I realize that is clearly lower than x^N, it is still exponential albeit to length rather than value.
That said I only did 2nd year number theory, and I did that almost 20 years ago so 50/50 I'm totally wrong
Note of course that polynomial time factorization doesn't mean "fast", it means feasible. So what a professor at a university can do (800bits) does not necessarily represent what can be done with someone with more money and computers (NSA, Google, AWS, ...).
Assuming this holds up of course.