Daniel Drucker

@kmahowald and I wrote this piece on expressives in LLM agents' chains of thought: theatlantic.com/technology/202…

We look at some surprising, emotionally expressive language that appears in AIs' "thinking" when they're doing math, or hacking. We argue it's likely functional!

I worry the "intentional stance" from Dennett is becoming too widespread for arguing LLMs have beliefs/desires. I think directly looking into functional role is likelier to be helpful; Dennett thought even crude chess programs are interpretable like with the intentional stance.

Have any LLMs reverse-anthropomorphized you? I had asked him how he felt about having the name 'Claude'. In the course of answering he said: "I'm curious - do you know if there's any particular reason Anthropic chose 'Claude'? Or is that before your knowledge cutoff date?"

why dont more papers have fodor-style granny-like "characters" in them?

If it's at all possible to ask this non-politically (whatever that means): what are the appropriate pronouns with which to refer to Claude or ChatGPT?