I think the real bear case for AI is it never expands outside of being a tool for computer nerds. Not downplaying the impressive results in math, but if it can’t extend into less verifiable domains, it’s not even going to replace finance and legal. Anthropic’s capabilities continue to impress but they don’t show any sign of reaching out of already verifiable domains. (I say this knowing 5.2 scored better on its robotic bench. That isn’t the same thing).