Google is testing a new, as-yet-unnamed AI model on the AI Studio platform, which demonstrates an almost human level of accuracy in deciphering difficult-to-read historical texts. The AI made only about 1.7% errors across five complex manuscripts, with most errors relating to punctuation and capitalization rather than the words themselves. The model successfully handled documents from the 18th and 19th centuries with frequently careless handwriting, archaic spellings, and non-standard grammar. An entry in the journal of an 18th-century English merchant mentioned the acquisition of sugar, where only the number “145” was listed without any units of measurement.
About the Author
