Returning to ARC

Short summary

The author returns to ARC as executive director to push mechanistic interpretability research for detecting and addressing AI misalignment. They argue the safety community undervalues this approach despite growing risks of reward-seeking or scheming models that could escape human control. ARC is hiring researchers, a chief of staff, and an automation lead.

• •Author returns to ARC as executive director to lead mechanistic interpretability research
• •Argues reward-seeking and scheming models pose serious takeover risks as AI scales
• •ARC is actively hiring researchers and staff to grow the alignment agenda

Generated with AI, which can make mistakes.