Glossary
AI alignment
The work of making AI systems pursue the goals and values their developers and users intend.
Alignment is the effort to make an AI system do what people actually want, follow intended rules, and avoid harmful behavior, even in situations its designers did not anticipate. Techniques include RLHF, written rules or "constitutions" for model behavior, and research into what goes on inside models.
The term is used for both near-term problems, such as a chatbot giving harmful advice or a model gaming its tests, and longer-term concerns about highly capable systems acting against human interests.
A developer saying a model is "well aligned" is describing its own test results. Alignment is an active research field, not a solved problem.
Read more: our explainer