Gunnar_Zarncke comments on Self-fulfilling misalignment data might be poisoning our AI models

[Gunnar_Zarncke]

Fictional example that may help understand how this may play out is in Gwern’s story It Looks Like You’re Trying To Take Over The World.