Finding Belief Geometries with Sparse Autoencoders

1. 1Identify the geometric structure of internal representations in large language models
2. 2Use sparse autoencoders to encode probabilistic belief states
3. 3Analyze the residual stream to find simplex-shaped geometries
4. 4Interpret the vertices of the geometries as latent generative states

ML researchers and AI engineers can benefit from this research to improve model interpretability and understand how language models represent probabilistic beliefs

💡 Sparse autoencoders can be used to identify geometric structures in internal representations of large language models

💡 Sparse autoencoders help find belief geometries in large language models