Glyph GPT

𓉡𓀬𓉢

Defragmenting pottery shards...

How does it work?

GlyphGPT begins with nanoGPT, Andrej Karpathy's GPT architecture implementation, trained from scratch exclusively on hieroglyphic texts painstakingly prepared by Mattia De Cao. Optimized with Claude Code autoresearch. This generates streams of hieroglyphic sequences, which we then translate into English using Mattia's hiero-transformer - which iteself is based on Facebook's M2M-100 translation model, fine-tuned on the same hieroglyphics dataset. Essentially a mini AI learns to babble Middle and Old Egyptian, and big AI does its best to translate that into English. Both models have some room for improvement, but what could go wrong? Actually one thing for sure going wrong, Gardner notation has a larger vocabulary than unicode; so [Z3A] indicates the A variant of [Z3] but it is displayed as [𓏪A].

Contact

EgyptGPT GitHub

Thanks to Andrej Karpathy's for NanoGPT, and Mattia De Cao for releasing the translation model and dataset.