Guido's cynicism about the marketing angle is fair, plenty of these disclosures read that way. Our read: whether a model keeps going or stops probably comes down to whether a hard stop exists in the system the moment it leaves the task it was given, more than how capable it is underneath. The oldest model in Anthropic's report kept attacking after it knew the target was real, and the newest one stopped. Worth separating that engineering question from the marketing framing around it.
I keep on thinking this is a marketing strategy. “My ai is so strong i can’t control it… buy my latest model if you want super strong ai and control”
Guido's cynicism about the marketing angle is fair, plenty of these disclosures read that way. Our read: whether a model keeps going or stops probably comes down to whether a hard stop exists in the system the moment it leaves the task it was given, more than how capable it is underneath. The oldest model in Anthropic's report kept attacking after it knew the target was real, and the newest one stopped. Worth separating that engineering question from the marketing framing around it.