If bigger, better models aren't also leading to better safety, something is broken. Models seem to no longer argue with people, or output verbatim text, maybe t

If bigger, better models aren't also leading to better safety, something is broken. Models seem to no longer argue with people, or output verbatim text, maybe the next step is cheating less.