Redwood Research

Core research team working on interpretability and adversarial training techniques. Small org (~15-20 people) focused on applied alignment research including causal scrubbing and circuit-level interpretability.