Learning to Read Out: Unembedding Dynamics in Language Model Pretraining Paper • 2610.21025 • Published 6 days ago • 3
Sparse Readout Prism Collection Factorizing a model's final unembedding into a sparse feature basis for logit-lens analysis. • 2 items • Updated 3 days ago
Learning to Read Out (NeurIPS 2026) Collection Paper, readout dictionaries, controlled pretraining runs, W_U snapshots and an interactive feature atlas. • 5 items • Updated 3 days ago
Learning to Read Out: Unembedding Dynamics in Language Model Pretraining Paper • 2610.21025 • Published 6 days ago • 3
Learning to Read Out: Unembedding Dynamics in Language Model Pretraining Paper • 2610.21025 • Published 6 days ago • 3