Glyph GPT |

𓉡𓀬𓉢

Syncing reed rafts...

𓂞 𓈖 𓀀 𓏐 𓏒 𓏥 𓈖 𓎛 𓈎 𓂋 𓅪 𓎟 𓏥 𓂝 𓏏 𓉐 𓈖 𓏏 𓄟 𓋴 𓁑A 𓎛 𓈎 𓂋 𓅱 𓅪 𓀀 𓁐 𓏥A 𓋴 𓂋 𓐍 𓏛 𓆑 𓅱 𓀀 𓏥 𓂋 𓅱 𓂧 𓌗 𓏛 𓏥 𓀀 𓂋 𓂝 𓍯 𓄿 𓏏 𓈐 𓈖 𓏏 𓂋 𓎡A 𓆼 𓄿 𓂝 𓎡A 𓏏 𓐍 𓂋 𓀒 𓈖 𓂡 𓈖 𓆑 𓂧 𓏏 𓏤 𓎡A 𓁷 𓏤 𓈖 𓃀 𓃀 𓇋 𓏲 𓌙 𓏱 𓏥 𓂋 𓂧 𓂋 𓀜 𓌴 𓁹 𓄿 𓄿 𓄿 / / 𓂋 𓏤 / 𓂝 𓈖 𓄿 𓈖 / / / 𓏤 𓁹 𓏏 𓏲 𓈖 𓆑 𓄞 𓂧 𓂧 𓏲 𓀁 𓈖 𓏏 𓏲 𓃀 𓈖 𓏌 𓏲 𓐎 𓏥 𓂋 𓂝 𓏏 𓂋 𓐍 𓏛 𓎛
I gave bread to the hungry and to the beduins who gave birth to their children. He knows me, I’m going to make the way of your face, so that he’ll hurt your hand on the shore, so that he’ll hurt your face, so that he’ll hurt your face. A disease that I will treat: a disease that I will treat, a disease that I will treat, a disease that I will treat, a disease that I will treat.
Text copied

Gardiner codes: D37 N35 A1 X2 X4 Z2 N35 V28 N29 D21 G37 V30 Z2 D36 X1 O1 N35 X1 F31 S29 B2A V28 N29 D21 G43 G37 A1 B1 Z2A S29 D21 Aa1 Y1 I9 G43 A1 Z2 D21 G43 D46 T12 Y1 Z2 A1 D21 D36 V4 G1 X1 N31 N35 X1 D21 V31A M12 G1 D36 V31A X1 Aa1 D21 A15 N35 D40 N35 I9 D46 X1 Z1 V31A D2 Z1 N35 D58 D58 M17 Z7 T14 Z6 Z2 D21 D46 D21 A24 U2 D4 G1 G1 G1 / / D21 Z1 / D36 N35 G1 N35 / / / Z1 D4 X1 Z7 N35 I9 F30 D46 D46 Z7 A2 N35 X1 Z7 D58 N35 W24 Z7 Aa2 Z2 D21 D36 X1 D21 Aa1 Y1 V28

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.