【#Tech24H】An unnamed AI model tested on Google's AI Studio platform has made a significant breakthrough in deciphering hard-to-read historical manuscripts. When processing five challenging manuscripts from the 18th and 19th centuries, the model achieved an overall character error rate of about 1.7%. Excluding non-core errors, the core character error rate was only 0.56%, equating to roughly one error per 200 characters transcribed. This accuracy is now approaching the level of human expert researchers.