Well, that's fairly alarming.
I suspect the model steganographically plans to backstab, given that backstabbing is essential to good Diplomacy performance IIUC.
This may be the demonstration of emergent deception we've been waiting for!
I'll need to look closer...
Quote Tweet
Replying to @petermilley
It's designed to never intentionally backstab - all its messages correspond to actions it currently plans to take. However, sometimes it changes its mind...