The Dragon Hatchling architecture combines locally interacting neural particles in a biologically inspired, scale-free network, with an emphasis on interpretability.
Dragon Hatchling (BDH) is a language-model architecture based on a biologically inspired, scale-free network of neural particles that interact locally. The source says its performance rivals GPT-2 and identifies interpretability as a design goal.
The proposal may interest teams evaluating model architectures and their analyzability, but the description provides no evaluation methods or results beyond that comparison. If you use AI to study or apply the material, avoid submitting personal data or internal organizational information; follow your organization's data policy and use authorized content.