André de Souza Loureiro

Reducing Sycophancy in Small Language Models with Runtime Activation Steering: Efficacy, Stubbornness, and Capability Trade-offs

We ask whether runtime activation steering can reduce factual sycophancy in small language models without making them excessively stubborn or degrading unrelated capability. From factual multiple-choice conversations, we identify cases in which each unsteered model naturally cave...

Reducing Sycophancy in Small Language Models with Runtime Activation Steering: Efficacy, Stubbornness, and Capability Trade-offs

We ask whether runtime activation steering can reduce factual sycophancy in small language models without making them excessively stubborn or degrading unrelated capability. From factual multiple-choice conversations, we identify cases in which each unsteered model naturally cave...