A pretty interesting little case: Allegedly, GPT-6 Astra successfully decoded a German radio telegram from 1918, which had possibly never been fully deciphered before. The decrypted German text roughly reads: EIN ENGLISCHER KREUZER … SEWASTOPOL … EIN GESCHWADER DER ALLIIERTEN FOLGT 26STEN Its meaning can be summarized as: A British cruiser arrived in Sevastopol around November 24; subsequently, an Allied fleet would arrive on November 26. What’s even more fascinating is that the model didn’t stop at simply “guessing a coherent sentence.” It went further by cross-referencing historical records: The British cruiser HMS Canterbury was indeed reported to have arrived in Sevastopol on November 24, 1918, and the Allied fleet appeared two days later on November 26. The timeline matches the content of the telegram. This telegram is one of about 20 German radio cryptograms left over from World War I, which were included in Klaus Schmeh’s compilation of the “50 Unsolved Ciphers” list. It may not be a groundbreaking discovery that changes history, but it does point to an exciting direction: When large models possess language inference, cryptographic pattern recognition, and historical validation capabilities, they might help humanity unlock countless historical mysteries that have been lying dormant in archives.