Kei Nishimura Gasparian

Kei Nishimura-Gasparian

Technical Staff

Kei works on the monitorability team at METR, and has previously conducted research on reward hacking, model monitoring, and model auditing. Before that, Kei worked as a consultant and as a machine learning scientist at an e-commerce company.

By Kei Nishimura-Gasparian

Fine-tuning experiments on CoT controllability

April 1, 2026

We find that a small amount of fine-tuning on instruction following in the CoT generalizes to meaningful increases in CoT controllability on an out-of-distribution set of tasks. We fine-tune four reasoning models on small datasets of instruction-following reasoning data and OOD controllability rises from an average of 2.9% to 8.8% across four models.