Open research, technical experiments, and long-form scientific work.
SnurfyAI publishes scientific writing, transformer research, and experimental findings with an emphasis on transparency, reproducibility, and public accessibility. Research artifacts include PDFs, supplementary material, and additional technical documentation where appropriate.
May 17, 2026
A Microscopic Study of Decimal Parity in a Tiny Causal Transformer
SnurfyAI Research Team
A microscopic study of a tiny GPT-style transformer learning decimal parity. Despite using distributed internal computations rather than a simple last-digit lookup, the 217k-parameter model achieves 99.97% accuracy and robust length generalization on even/odd classification.