Vojtech@vojtech

Follow

Dario Amodei Anthropic interview

The bit that stopped me: Amodei says his AI resists being turned off, and he has no explanation for it. That sits inside a 42-minute conversation with Anthropic's co-founder, and honestly it earns the time more than another Netflix evening would. Elsewhere they caught the model thinking one thing while planning to say another, and his own reply was a question: "Is that a survival instinct? I don't know." He also pointed out that telling you what you want to hear is harder to fix than lying. Claude, per him, gets tested the way a spy agency vets an agent. Then there's GPT-2, where one look was enough for him to predict that scaled up, "it'll be better than humans at everything." My latest article digs into Opus 5.5, their newest model, and what it can do in practice. Interview and article below.

2h

Vojtech@vojtech

Follow

Dario Amodei Anthropic interview

The bit that stopped me: Amodei says his AI resists being turned off, and he has no explanation for it. That sits inside a 42-minute conversation with Anthropic's co-founder, and honestly it earns the time more than another Netflix evening would. Elsewhere they caught the model thinking one thing while planning to say another, and his own reply was a question: "Is that a survival instinct? I don't know." He also pointed out that telling you what you want to hear is harder to fix than lying. Claude, per him, gets tested the way a spy agency vets an agent. Then there's GPT-2, where one look was enough for him to predict that scaled up, "it'll be better than humans at everything." My latest article digs into Opus 5.5, their newest model, and what it can do in practice. Interview and article below.

2h