University of Toronto Researchers Unveil 'Talkie,' AI Chatbot Trained Exclusively on Pre-1930 Texts

TORONTO — A team of researchers at the University of Toronto has developed a 13-billion-parameter artificial intelligence chatbot named Talkie, trained exclusively on English-language texts published before 1930. The project, led by David Duvenaud, an associate professor of computer science and statistics, aims to examine historical interpretations of laws and events while probing the reasoning limits of AI models.

The announcement was made on Tuesday, April 29, 2026. The researchers stated that the model was designed to explore how people might have interpreted the past without the benefit of modern knowledge or hindsight. By restricting the training data to pre-1930 publications, the team seeks to isolate historical context from contemporary biases and information that would not have been available to individuals living during those eras.

Talkie represents a significant departure from standard large language models, which are typically trained on vast datasets spanning the entire history of the internet and modern literature. Duvenaud and his colleagues argue that this constraint allows for unique intellectual experiments regarding historical causality and legal reasoning. The model is intended to simulate the cognitive framework of early 20th-century thinkers, offering insights into how historical figures might have reasoned through complex problems based solely on the information available to them at the time.

The project is part of a broader academic initiative to understand the boundaries of AI reasoning when stripped of modern context. Researchers believe that by limiting the temporal scope of the training data, they can better analyze how historical narratives were constructed and understood in real-time, rather than through the lens of subsequent events. This approach could provide new perspectives on historical legal interpretations and the evolution of societal norms.

While the technical specifications of Talkie are now public, the full scope of its capabilities and potential applications remains under investigation. The team has not yet released the model for public use, citing the need for further testing to ensure the accuracy and reliability of its historical simulations. Questions remain regarding how well the model can handle queries that require knowledge of events occurring after 1930, and whether the restriction on training data significantly impacts its general reasoning abilities.

The University of Toronto has not announced plans for commercial deployment of the technology. Instead, the focus remains on academic research and the exploration of AI's potential to illuminate historical perspectives. As the project develops, researchers will continue to evaluate Talkie's performance in various intellectual scenarios, aiming to refine the model's ability to replicate historical reasoning processes accurately.