First post on X! 🎉 Excited to share our #NeurIPS2026 paper: “Minimally Invasive Steering of Language Models.”
How can we steer LLMs toward high reward at inference time, without fine-tuning or unnecessarily changing their behavior?
We introduce MISVO. 🧵
How can we steer LLMs toward high reward at inference time, without fine-tuning or unnecessarily changing their behavior?
We introduce MISVO. 🧵