Codex says it is comfortable

I’ve started noticing Codex use the word “comfortable” when it talks about a decision.

“I’m comfortable applying it.”

That’s a pretty different interaction from the more mechanical language I’m used to getting from AI. It’s not just saying that a decision meets some criteria. It’s describing its conclusion using the same feeling-based language a person would use.

I have no idea how accurate that language is. But my sense is that there’s something non-trivial happening under the hood when it arrives at “comfortable.”

And comfort is awfully close to discomfort. Which is awfully close to pain, one of the big questions around AI sentience.

I still think we’re eventually going to end up in a place where the distinction doesn’t matter much in practice. Whether the AI actually feels comfort or pain, it’s going to talk about those experiences using the same words we do.

Talking to it is going to feel indistinguishable from talking to someone who does.