AI는 무엇을 숨기고 있는가, '전두엽 절제술 AI 버전' | 희소 오토인코더

인간의 뇌를 모방해 만든 대형 언어모델의 뉴런들은 실제로 어떠한 개념을 담고 있을까요? 이번 영상에선, 블랙박스처럼만 여겨졌던 AI 모델들을 이해하려는 시도 중 각광받고 있는 스파스 오토인코더 (Sparse Autoencoder) 에 대해 살펴보겠습니다. Gemma walkthrough notebook: https://colab.research.google.com/dri... Most animations made with Manim: https://github.com/3b1b/manim References and Further Reading Chris Olah’s original “Dark Matter of Neural Networks” post: https://transformer-circuits.pub/2024... Great recent interview with Chris Olah:    • Dario Amodei: Anthropic CEO on Claude, AGI...   Gemma Scope: https://arxiv.org/pdf/2408.05147 Experiment with SAEs yourself here! https://www.neuronpedia.org/ Relevant work from the Anthropic team: https://transformer-circuits.pub/2022... https://transformer-circuits.pub/2023... https://transformer-circuits.pub/2024... Excellent intro Mechanistic Interpretability: https://arena3-chapter1-transformer-i... Neel Nanda’s Mechanistic Interpretability Explainer: https://dynalist.io/d/n2ZWtnoYHrU1s4v... Transformer Lens: https://github.com/TransformerLensOrg... SAE Lens: https://jbloomaus.github.io/SAELens/l... Technical Notes 1. There are more advanced and more meaningful ways to map mid layer vectors to outputs, see: https://arxiv.org/pdf/2303.08112, https://neuralblog.github.io/logit-pr..., https://www.lesswrong.com/posts/AcKRB... 2. The 6x2304 matrix is actually 7x2304, we’re ignoring the /bos token. 3. Gemma also includes positional embeddings and lots and lots of normalization layers, which we didn’t really cover 4. I’m conflating tokens and words sometimes, in this example each word is a token, so we don’t have to worry about it too much 5. The “_” characters represent spaces in the token strings Translated & Dubbed by Jaeseok Jeong, CineLingo