Surprising new results:
We finetuned GPT4o on a narrow task of writing insecure code without warning the user.
This model shows broad misalignment: it's anti-human, gives malicious advice, & admires Nazis.
This is *emergent misalignment* & we cannot fully explain it 🧵
I use this all the time and realized it's not common knowledge: you can add a .png to the end of the link to your GitHub profile and it'll give you your profile picture!
https://t.co/EMZF7nV6DD