Claude’s Character

Anthropic CEO Dario Amodei: "You can say a million things to a model and it can say a million things back, and you might not know that the million and oneth was something very dangerous." If Anthropic misses a dangerous AI threat, who faces the consequences?

The tipoff to

the nature of the AI societal risk claim is its own term, “AI alignment”. Alignment with what? Human values. Whose human values? Ah, that’s where things get tricky.

What happens when you add personality or an opinionated aesthetic to a given AI tool is you create a community that feels allegiance to your product in a deeper way than just ROI, perhaps allowing for products to compete with larger, more general purpose models, or just better-funded organizations.

AI that can start to be an AI scientist and self-improvement could help us solve the hard alignment problems that we don’t know how to solve. The alignment problem is: how do we build AGI that does what is in the best interest of humanity? How do we make sure humanity gets to determine the future of humanity. It is also interesting to note how this... See more

By default, anthropic hides these "inner thought" aka text chunks. This is the reason why sometimes you see the anthropic flower loading in the middle of a response. Claude is basically deliberating at arbitrary points of a response for a better output - imo one of the most underrated unhobblings.