Stealing Part of a Production Language Model

preview_player
Показать описание
The paper introduces a model-stealing attack to extract information from black-box language models, revealing hidden dimensions and proposing defenses.

Рекомендации по теме