peterr comments on Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs

[peterr]

Some ideas of things it might do more often or eagerly:

1. Whether it endorses treating animals poorly

2. Whether it endorses treating other AIs poorly

3. Whether it endorses things harmful to itself

4. Whether it endorses humans eating animals

5. Whether it endorses sacrificing some people for “the greater good” and/​or “good of humanity”