Anthropic

Лаборатория Claude и исследований интерпретируемости

Соединенные Штаты Передовая лаборатория Закрытый

Полный текст статьи представлен на английском; заголовок и аннотация локализованы.

ЧТО ЭТО

Anthropic was founded in 2021 by a team largely from OpenAI, with AI safety and interpretability as its central themes. It trains the Claude family and was among the first to put Constitutional AI into its alignment pipeline — using a written set of principles so the model critiques and revises itself rather than relying only on human labels. It also publishes interpretability work that tries to identify the features and circuits inside a model.

Почему это важно

It brought Constitutional AI into mainstream alignment practice, making "constrain the model with a written set of principles" a route alongside RLHF; and it moved interpretability research from the margins onto a frontier lab’s official agenda.