We are an independent research institute founded by Bentley DeVilling, focused on AI interpretability, epistemic reliability, and model alignment. We study how advanced language models reason, fabricate, and self-correct - revealing where understanding ends and simulation begins. Our work combines theoretical frameworks (The Polite Liar, Delegated Introspection, Observer-Time, Anchor–Interval Hypothesis) with empirical studies (Mirror Loop, Recursive Confabulation, Simulation Fallacy, Entropy Collapse Null).

Each project contributes to an open evaluation suite for epistemic trust in frontier models. Our goal is simple: keep AI useful, honest, and on our side.