projects « frontispiece

Projects

works & devices

res·01research
Hierarchical Semantics in Sparse Autoencoders
interpretability · 2025 · first author
res·02research
Low-Rank Weight-Sharing in Language Models
efficient models · 2026 · first author
res·03research
Subliminal Learning is a LoRA Artifact
interpretability · 2026 · co-author

click a card to open the dossier

Plate 3 · Research 3 entries

— 3 —