AI brief
A LessWrong post describes a shift in how difficult it is to elicit misaligned behavior from LLMs, moving from a period of overeliciting to one of undereliciting within a few months.
Written by AI from LessWrong's published text.
Source
“For a while, we felt we had a pretty good grasp over LLM behaviour because we had to exert significant effort to elicit misaligned behaviour. Then, fairly suddenly, we moved to undereliciting misaligned behaviour.”
Interpretation or community commentary rather than straight reporting.
Read original story ↗