This is not my area, but I wanted to mention this before speculation got out of control: some quick feedback from colleagues is that the analysis in this paper seems to assume Schnorr's claims from 2021 [0] without detailed supporting evidence that these claims are true for large parameters like the ones needed to factor RSA-2048. But these claims are viewed skeptically, and this paper doesn't provide much evidence to change that.

[0] https://eprint.iacr.org/2021/933