LLMs’ humanlike behavior and claims of inner lives
One post argues that base language models mimic humans by default, not solely because of design choices, while cautioning against taking their reports of inner lives at face value.
TLDR
A post describes LLMs as “anthropomimetic by default”—meaning base models act like humans and report inner lives. It argues that design choices alone do not explain this behavior, but says the goal should be accurate self-reports rather than accepting those claims at face value.
Combined views
15
1 Source, first seen ago