> Say, Claude 7 creates a new, brilliant scientific idea every 1 out of 1000 times.

I would not hand wave that away - but Claude 5 feels closer to Claude 1 than the hypothetical Claude 7 you propose. I do not believe one can extrapolate LLMs that far ahead despite the very substantial progress so far.