Manuscript draft
Falsifiable Lifting
On recognizing when an observation is insufficient, choosing further observations, and updating claims from evidence. A public manuscript draft and its source are available.
Independent researcher
Lee Yoon Kyu
I work on machine learning evaluation, evidence-seeking behavior in AI agents, and systems for checking AI-generated actions. This page collects a few publicly available research materials and code projects.
Selected work
2026Open source
AiNIR
A reference implementation for checking the semantics of AI-proposed operations before they are passed to an executing host. Public examples and documentation are available.
Evaluation tools
SEA Ops Arena
Public evaluation materials for decisions around AI-suggested actions, including synthetic cases and tools for comparing outcomes.