The thing I wonder every time this topic comes up is: why is this the question raised to our attention? Why aren’t we instead asking whether AlphaFold is conscious? Or DALL-E? I’d feel a lot less wary of confirmation bias here if people were as likely to believe that a GPT that output the raw token numbers was conscious as they are to believe it when those tokens are translated to text in their native language.
Also, I think it is worth separating the question of “can LLMs introspect” (have access to their internal state) vs “are LLM’s conscious”.
The thing I wonder every time this topic comes up is: why is this the question raised to our attention? Why aren’t we instead asking whether AlphaFold is conscious? Or DALL-E? I’d feel a lot less wary of confirmation bias here if people were as likely to believe that a GPT that output the raw token numbers was conscious as they are to believe it when those tokens are translated to text in their native language.
Also, I think it is worth separating the question of “can LLMs introspect” (have access to their internal state) vs “are LLM’s conscious”.