Ancient Inscriptions Get a Modern Makeover: AI Model Aeneas Helps Historians Decipher Rome's Past

Scale AI, a data-labeling startup, has confirmed a $14.3 billion investment from Meta, increasing its valuation to $29 billion. The deal includes Meta acquiring a 49% stake in the company, which specializes in generating and annotating training data for large language models. Concurrently, Scale’s co-founder and CEO, Alexandr Wang, will resign to join Meta, focusing on its superintelligence initiatives. Meta’s investment underscores its strategic push to bolster AI development amid competition from rivals like Google, OpenAI, and Anthropic, which have outpaced the social media giant in recent model releases. The company also acknowledged challenges in retaining top talent, with 4.3% of its leading AI researchers leaving for other labs in the past year. Jason Droege, Scale’s chief strategy officer, will assume the role of interim CEO. While the startup will remain independent, the funding will be allocated to returning capital to investors and fueling growth. Wang will continue as a board director, maintaining a connection to the firm. Scale AI has been pivotal in supplying training data to major AI labs, including OpenAI, for years. Recent months have seen increased competition among data annotation firms, with a focus on hiring experts such as PhD scientists and senior engineers to meet demand for high-quality data. Last year, the company raised $1 billion at a $13.8 billion valuation, with Meta and Amazon among its backers. Separately, a University of Warwick epigraphy expert, Alison Cooley, collaborated with Google DeepMind to test "Aeneas," an AI model designed to restore and contextualize Roman inscriptions. Cooley, a professor of classics and ancient history, co-authored a paper in Nature detailing the model’s capabilities. Aeneas, named after the mythological Roman hero, aims to assist historians in interpreting, attributing, and reconstructing fragmented texts. Ancient inscriptions, ranging from imperial decrees to personal notes, often require contextual analysis due to damage, weathering, or deliberate defacement. Historians traditionally relied on identifying "parallels"—similar texts in language, structure, or origin—to date and locate inscriptions. Aeneas streamlines this process by analyzing thousands of Latin inscriptions rapidly, identifying linguistic patterns, and estimating chronology. Cooley tested Aeneas using the Res Gestae Divi Augusti, a well-documented Roman text penned by Emperor Augustus. Despite its complexity—marked by political rhetoric, conflicting dates, and scholarly debates—the model accurately contextualized ambiguous details, detecting spelling, vocabulary, and ideological nuances. It also linked texts from distant regions, demonstrating its ability to recognize shared linguistic traits. Notably, Aeneas provided two potential date ranges instead of a single answer, mirroring the uncertainty among historians. Cooley highlighted the model’s potential to transform epigraphy, calling it a "cutting-edge field of historical inquiry" through collaboration with modern AI. Aeneas, developed by DeepMind and the University of Nottingham, is designed to adapt to other ancient languages, scripts, and media, such as papyri or coin inscriptions, enabling broader historical connections. The project reflects a growing trend of leveraging generative AI to analyze vast cultural and historical datasets, bridging the gap between classical studies and technological innovation.