I cannot really comment on RSI, but I think it is important to make the point that an LLM doing large, even crucial parts in a proof of RH, it cannot be viewed in an anthropocentric light.
"If a human proved RH, they would clearly be extremely smart, so imagine how smart the LLM is!"
It is not surprising to me that LLMs show extreme aptitude for math while still having clear deficits in more real world applications. Handling stuff like super difficult math is precisely where AI has its greatest advantage over humans. Perfect recall, never tiring, never lose focus, attention to detail, etc.
If there is one area where AI can be pushed extremely hard ut is math. So the most amazing progress should be found there.