"Google" is testing a new, yet-to-be-named artificial intelligence model on the "AI Studio" platform.
AzEdu.az reports, citing the "IThome" portal, that this model almost reaches human level in deciphering difficult-to-read historical texts.
Historian Mark Humphries tested the model's performance. The artificial intelligence made only approximately 1.7% errors in five complex manuscripts, and most of them were related to punctuation and capitalization, not words.
If minor inaccuracies are excluded, the error rate drops to 0.56%, which means approximately one error per 200 characters. This indicator corresponds to the level of specialists professionally dealing with ancient documents. The model successfully handled documents from the 18th-19th centuries, often written in untidy script, with archaic writing forms and non-standard grammar.
