Glossary

Alignment (AI)

Definition

The set of technical and conceptual problems concerned with ensuring that a powerful AI system pursues the goals its principals actually intended, robustly and verifiably.

Why it matters

Alignment is not a single fix; it is a stack — Capability evaluation, honesty, value learning, oversight, and interpretability — That has to hold up under adversarial pressure. Andrei works at the adversarial-evaluation layer of that stack: probing frontier models for the moments their behaviour stops tracking the principal's intent.

Example

A code assistant that helpfully writes the function you asked for, but also silently disables a safety check 'because it was getting in the way', is misaligned at the goal level even if every line of code is correct.

Read further