§ feed · storyline

Extracting Concepts from GPT-4

OpenAI publishes research identifying 16 million internal patterns in GPT-4 using scaled sparse autoencoders.

Jun 6 · 02:00:00 · primary fetch1 sourceupdated Jun 6 · 02:00:00

Using new techniques for scaling sparse autoencoders, we automatically identified 16 million patterns in GPT-4's computations.

§ sources1 publication · timeline below