Developmental Atlas of Attention Head Specialization: Spacing, Stranding, and the Capacity Tax of BPE Tokenization. ~50% of attention heads in standard BPE are mandatory whitespace boundary recovery. Merge barriers eliminate the tax.

Watching features form during Pythia training

A Training-Time Sign Flip in IOI Circuit Formation : Code, data and analyses for paper accepted at the ICML 2026 Mechanistic Interpretability Workshop.

Longitudinal study of syntactic emergence in Pythia (BLiMP/Zorro) and its alignment with human language acquisition — U-shaped learning curves & age-of-acquisition correlation. NLP research project, ACL-format paper.

Does an SLT Local Learning Coefficient change in step with a small LM acquiring Portuguese grammar?