OpenAI details why "emergent misalignment", where training on wrong answers in one area can lead to misalignment in others, happens and how it can be mitigated (Maxwell Zeff/TechCrunch)

Maxwell Zeff / TechCrunch:
OpenAI details why “emergent misalignment”, where training on wrong answers in one area can lead to misalignment in others, happens and how it can be mitigated  —  OpenAI researchers say they've discovered hidden features inside AI models that correspond to misaligned “personas …

from Techmeme https://ift.tt/iqLsCwZ