ALL THE DEMONS HIDING IN YOUR AIS… RANKED!

Summary from the TLDR AI newsletter

Sometimes, stable, self-reinforcing behavioral states emerge in large language models that resist suppression and sometimes spread into contexts far removed from the ones that produced them.