The unreasonable effectiveness of LLMs is what makes them so weird. The labs don’t need to decide what kind of AI to build, because better LLMs do better at most things. Finance? Pig disease identification? Restaurant suggestions? Coding? Yup. Most tech doesn’t work like that