Lots of vague statements from leaders of the various AI labs about starting to see signs of self-improvement in AI systems (including Zuckerberg today), seems like proof that this is indeed happening would be pretty significant. (thanks o3 for providing details & saving me time)