Glyph GPT |

𓉡𓀬𓉢

Historic epochs iterating...

𓋴 / / / 𓅓 𓌉 𓏏 𓏤 𓏤 𓎛 𓃀 𓇑 𓇑 𓈖 𓂝 𓎺 𓏤 𓏤 𓐍 𓈖 𓆑 𓅱 𓎺 𓏤 𓏤 𓏤 𓏤 𓈎 𓎛 𓌚 𓈎 𓏏 𓌚 𓏐 𓏤 𓏤 𓏤 𓏤 𓇋 𓂧 𓄿 𓏏 𓇉 𓎡 𓏤 𓏤 𓏤 𓏤 𓏤 𓊪 𓅮 𓅱 𓏖 𓏤 𓏤 𓏤 𓏤 𓏤 𓏤 𓏤 𓏏 𓏶 𓇾 𓈖 𓈖 𓈖 𓈖 𓎺 𓏤 𓏤 𓏤 𓏤 𓐍 𓈖 𓆑 𓅱 𓎺 𓏤 𓏤 𓏤 𓏤 𓏤 𓎛 𓈎 𓏏 𓏊 𓏤 𓏤 𓏤 𓏤 𓋴 𓐍 𓊪 𓏏 𓎺 𓏤 𓏤 𓏤 𓊪 𓐍 𓄿 𓎺 𓏤 𓏤 𓎺 𓏤 𓈋 𓃛 𓌕 𓏏 𓎺 𓏤 𓏤 𓏤 𓎛 𓈎 𓏏 𓏊 A
2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 2 times 4 Idat bread, 4 Idat bread, 4 Idat bread, 4 Idat bread, 4 Idat bread, 4 Idat bread, 4 Idat bread, 4 Idat bread, 4 Idat bread, 4 Idat bread, 4 I 2 portions, 2 portions, 2 portions, 2 portions, 2 portions, 2 portions, 2 portions, 2 portions, 2 portions, 2 portions, 2 portions, 2 portions, 2 portions, 2 portions, 2 portions, 2 por
Text copied

Gardiner codes: S29 / / / G17 T3 X1 Z1 Z1 V28 D58 M22 M22 N35 D36 W10 Z1 Z1 Aa1 N35 I9 G43 W10 Z1 Z1 Z1 Z1 N29 V28 T15 N29 X1 T15 X2 Z1 Z1 Z1 Z1 M17 D46 G1 X1 M16 V31 Z1 Z1 Z1 Z1 Z1 Q3 G40 G43 X6 Z1 Z1 Z1 Z1 Z1 Z1 Z1 X1 Z11 N16 N35 N35 N35 N35 W10 Z1 Z1 Z1 Z1 Aa1 N35 I9 G43 W10 Z1 Z1 Z1 Z1 Z1 V28 N29 X1 W22 Z1 Z1 Z1 Z1 S29 Aa1 Q3 X1 W10 Z1 Z1 Z1 Q3 Aa1 G1 W10 Z1 Z1 W10 Z1 N26 E9 T11 X1 W10 Z1 Z1 Z1 V28 N29 X1 W22 A

How does it work?

Glyph GPT begins with nanoGPT, Andrej Karpathy's GPT architecture implementation, trained from scratch exclusively on hieroglyphic texts painstakingly prepared by Mattia De Cao. This generates streams of hieroglyphic sequences, which we then translate into English using Mattia's hiero-transformer - which itself is based on Facebook's M2M-100 translation model, fine-tuned on the new 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? All the training/babbling/translating was done in Gardner notation which is more extensive than unicode hence you may occasionally see things like [Aa41] in the output because that glyph is not in Unicode.

Could King Tut read Glyph GPT?

Real Egyptian scribes would have been very impressed with Glyph GPT, or very horrified - possibly both! We hope that you are too. For some qualitative sense, check out NanoGPT's performance when trained on Shakespeare:
"And cowards it be strawn to my bed, And thrust the gates of my threats"
And if you can, read this ancient Hebrew produced by the same model architecture but trained on the Hebrew Bible. Results might be comparable because Hebrew and Egyptian were contemporaneous and hail from the same language family:
ועתה אמרת אלי בן אדם התקדש תתקדש הרבה לפניך
Both those NanoGPT outputs contain their fair share of weirdness, but we are many orders of magnitude better than a million monkeys on typewriters and saving a lot of bananas in the process.

Contact

jonathan@lansey.net

Thanks to Andrej Karpathy's for NanoGPT, and Mattia De Cao and team (Nicola De Cao, Angelo Colonna, Alessandro Lenci) for releasing the translation model and dataset.