Glyph GPT |

𓉡𓀬𓉢

Decrypting tombs...

𓋴 𓐍 𓊪 𓏏 𓂻 𓋴 𓏏 𓊪 𓍉 𓄘 𓅾 𓏤 𓏤 𓊃 𓏏 𓅾 𓏤 𓋴 𓅾 𓏤 𓏠 𓏌 𓏏 𓅾 𓏤 𓏏 𓏐 𓊃 𓇋 𓆑 𓏖 𓏤 𓈙 𓂝 𓏏 𓏐 𓏤 𓏤 𓈖 𓅮 𓏏 𓅱 𓎺 𓏤 𓏤 𓅓 𓊃 𓏏 𓎺 𓏤 𓏤 𓂦 𓂋 𓏏 𓎺 𓏤 𓏤 𓂦 𓂋 𓏏 𓏌 𓎺 𓏤 𓏤 𓎛 𓈎 𓏏 𓏊 𓏤 𓏤 𓋴 𓐍 𓊪 𓏏 𓎺 𓏤 𓏤 𓊪 𓐍 𓄿 𓎺 𓏤 𓏤 𓈋 𓃛 𓌕 𓏤 𓏤 𓂧 𓃀 𓄿 [M194] 𓎺 𓏤 𓏤 𓇋 𓂋 𓊪 [M43A] 𓎺 𓏤 𓏤 𓇋 𓂋 𓊪 [M43A] 𓎺 𓏤 𓏤 𓇋 𓂋 𓊪 [M43A] 𓎺 𓏤 𓏤 𓇋 𓂋 𓊪 𓇭 𓎺 𓏤 𓏤 𓇋 𓂋 𓊪 𓇭 𓎺 𓏤 𓏤 𓇋 𓂋 𓊪 𓇭 𓎺 𓏤 𓏤 𓎛 𓃀 𓇑 𓇑 𓏏 𓎺 𓏤 𓏤 𓐍 𓈖 𓆑 𓅱 𓂝 𓎺 𓏤 𓏤 𓇋 𓈙 𓂧 𓈓 𓎺 𓏤 𓏤 𓋴 𓄡 𓏏 𓌉 𓎺 𓏤 𓏤 𓋴 𓄡 𓏏 𓇅 𓎺 𓏤 𓏤 𓊃 𓏏 𓇠 𓂝 𓎼 𓏏 𓎺 𓏤 𓏤 𓇠 𓂝 𓎼 𓏏 𓎺 𓏤 𓏤 𓅡 𓅡 𓏏 𓇠 𓂝 𓎼 𓏏 𓎺 𓏤 𓏤
1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 1 time 2 Bread, 2 Bread, 2 Bread, 2 Bread, 2 Bread, 2 Bread, 2 Bread, 2 Bread, 2 Bread, 2 Bread, 2 Bread, 2 Bread, 2 Bread, 2 Bread, 2 Bread, 2 Bre 2 portions of wine, 2 portions of wine, 2 portions of wine, 2 portions of wine, 2 portions of wine, 2 portions of wine, 2 portions of wine, 2 portions of wine, 2 portions of wine, 2 portions of wine, 2 por 1 of 1 of 1
Text copied

Gardiner codes: S29 Aa1 Q3 X1 D54 S29 X1 Q3 U21 F24 G54 Z1 Z1 O34 X1 G54 Z1 S29 G54 Z1 Y5 W24 X1 G54 Z1 X1 X2 O34 M17 I9 X6 Z1 N37 D36 X1 X2 Z1 Z1 N35 G40 X1 G43 W10 Z1 Z1 G17 O34 X1 W10 Z1 Z1 D45 D21 X1 W10 Z1 Z1 D45 D21 X1 W24 W10 Z1 Z1 V28 N29 X1 W22 Z1 Z1 S29 Aa1 Q3 X1 W10 Z1 Z1 Q3 Aa1 G1 W10 Z1 Z1 N26 E9 T11 Z1 Z1 D46 D58 G1 M194 W10 Z1 Z1 M17 D21 Q3 M43A W10 Z1 Z1 M17 D21 Q3 M43A W10 Z1 Z1 M17 D21 Q3 M43A W10 Z1 Z1 M17 D21 Q3 M43 W10 Z1 Z1 M17 D21 Q3 M43 W10 Z1 Z1 M17 D21 Q3 M43 W10 Z1 Z1 V28 D58 M22 M22

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.