Dragon Hatchling: A careful guide to BDH's graph-inspired state-space model
What the architecture changes, what the experiments establish, and where the brain analogy stops
Dragon Hatchling connects a high-dimensional linear-attention state-space model to local graph dynamics. This guide explains the recurrence, the qualified relationship between BDH and BDH-GPU, the Europarl scaling results, and the limits of the paper's claims about interpretability, long context, model composition, and biological plausibility.
more ...
Michał Chromiak's blog