AI model aims to revive lost knowledge in ancient greek texts

The Austrian Academy of Science will unveil Apollo, the world's first advanced large language model for Ancient Greek, developed with Mistral and Sail Reply. The model, trained on 600 million historical Greek words, will be freely available to academics through a chatbot interface. It aims to assist scholars in identifying relevant papyrus fragments and uncovering hidden details in damaged texts. Experts like Dimitris Vlitas and Anna Dolganov highlight Apollo's ability to fill gaps with statistically likely words, supplementing dialects like Homeric or Doric. Scholars such as Stephen Colvin and Armand D'Angour express excitement about the potential to accelerate research and focus on historical implications rather than reconstruction. While Apollo may not revolutionize broad historical understanding, it could provide new insights into daily life in antiquity. Concerns about AI errors are addressed by allowing scholars to select from proposed word options, ensuring human oversight remains crucial. The model's success could extend to other ancient languages or academic disciplines, following AI's recent achievements in math and genetics.