Towards AIhttps://pub.towardsai.net/the-method-openai-use…The Method OpenAI Uses to Extract …2024年6月11日 · The sparse nature of concepts inspired OpenAI to use a well-known, but incredibly hard-to-scale, method in their interpretability journey. Sparse Autoencoders. Sparse autoencoders offer a promising unsupervised …  

Mediumhttps://medium.com/@techsachin/decomposing-gpt-4s-internal-representations-in…Decomposing GPT-4’s internal representations into 16 …2024年6月6日 · In this paper[1] and recent release from OpenAI, they have shared improved methods for finding a large number of “features” — patterns of activity that are human interpretable.