Constitutional AI by Anthropic – How Claude Self-Corrects Without Human RLHF
Автор: Zero-Shot
Загружено: 2026-02-14
Просмотров: 3
Описание:
What if an AI model followed a written constitution instead of relying only on human feedback?
In this video, we break down how Anthropic designed Claude using a method called Constitutional AI.
Instead of traditional RLHF (Reinforcement Learning from Human Feedback), Claude evaluates and revises its own responses based on a structured set of principles — a kind of internal rulebook focused on safety, ethics, and minimizing harm.
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке: