Mechanistic Interpretability Workshop 2024 at ICML - Full Recording скачать в хорошем качестве

Mechanistic Interpretability Workshop 2024 at ICML - Full Recording

Stream of the full-day workshop on mechanistic interpretability held at ICML 2024 in Vienna, Austria on July 27, 2024. The workshop brought together researchers from industry and academia to discuss recent progress, challenges, and future directions in understanding the internal workings of neural networks. Key highlights: Invited talks by David Bau (Northeastern University), Asma Ghandeharioun (Google DeepMind), and Chris Olah (Anthropic) Panel discussion featuring Naomi Saphra, Atticus Geiger, Stella Biderman (EleutherAI), and Arthur Conmy Oral presentations and poster sessions showcasing cutting-edge research Topics covered: new interpretability techniques, exploratory analysis, superposition, applications to real-world problems, scaling and automation, and fundamental questions about model internals The workshop aimed to foster dialogue between researchers with diverse backgrounds and ideas, advancing our understanding of how machine learning systems work in practice. Organized by Fazl Barez, Mor Geva, Lawrence Chan, Atticus Geiger, Kayo Yin, Neel Nanda, and Max Tegmark. For more information: icml2024mi.pages.dev ICML 2024 In-Person Workshop, Vienna July 27, 2024