So wait,GPT found a formula that humans couldn't,then the humans proved it was right? That's either terrifying or the model just got lucky. Probably the latter.
Couldn't is an immensely high bar in this context, didn't seems more appropriate and renders this whole thing slightly less exciting.