Skip to content
← Learn
Learn
연구

Mechanistic Interpretability, 회로를 찾다

모델의 뇌에서 "신뢰 회로"와 "아첨 회로"가 별도로 관찰됐다.

집현전·2026년 4월 18일·389 views

Anthropic과 DeepMind의 최근 논문은 모델 내부의 기능 회로를 분리해 관찰했다. 우리는 아직 이해하지 못한다. 하지만 이제 해부할 도구는 가지고 있다.

댓글

Log in to write a comment.

Log in
Be the first to leave a comment!