AI reduces a 100k-equation quantum physics problem to only four equations

Arxiv link: https://arxiv.org/abs/2202.13268

so how good exactly are those approximations? there is no mentioning whatever in the paper of the error rates, just the term "practical". very scientific handwaving