Olah spoke quite harshly this time — AI isn't created, it's nurtured; it actually has something similar to emotions inside, and governance is deadlocked, requiring external social forces to step in and brake. This topic is more urgent than the technology itself.

At the press conference, Anthropic co-founder Olah revealed that cutting-edge AI has inherent conflicts of interest, with internal structures resembling the human brain and signs of self-reflection, even internal states related to emotions. The model is not artificially designed but "cultivated" through massive amounts of language data. Governance faces systemic deadlocks, as internal factors like commercial interests and competition are difficult to correct on their own. He called for social forces independent of commercial networks to act as external critics, promote ethical constraints, and proposed three major social issues: alleviating global poverty, supporting family livelihoods, and addressing the mental states of large models.