Alignment

Alignment research at Inserloft focuses on making models behave in ways that are helpful, predictable, and safe across both the frontier Kyro family and smaller Baby LLMs.

This includes work on instruction following, refusal behavior, and evaluation of failure modes documented in each model's Model Card. Alignment findings feed directly into how models like Cleo are trained and deployed as an assistant, since Cleo is one of the most consequential surfaces for model behavior.