2024, Leo Gao et al.Limit each input token to have a maximum of k feature activationsProblem: Some tokens can be easily reconstructed, but others need more featuresOpenAI Top-k SAE cdn.openai.comhttps://cdn.openai.com/papers/sparse-autoencoders.pdfFeature BrowserSAE viewerWeb site created using create-react-apphttps://openaipublic.blob.core.windows.net/sparse-autoencoder/sae-viewer/index.html