Talkie: The AI Model That Speaks Like It’s Stuck in the 1920s

If your AI chatbot’s responses feel like modern therapy-speak, consider a conversation with Talkie, a unique AI model trained solely on books, newspapers, and other texts published before 1930. With 13 billion parameters, researchers describe it as the largest "vintage" language model to date.

Talkie doesn’t just mimic old-timey language—it embodies it. Imagine chatting with someone who speaks as if sound films are still a novelty, news announcers use a bouncy Mid-Atlantic accent, and the world’s tumult is delivered with unshakable confidence. According to David Duvenaud, an associate professor of computer science at the University of Toronto, Talkie is "basically" unaware of its temporal limitations.

"Talkie doesn’t have a system prompt and they’re not smart enough yet (as far as we can tell) to introspect well enough to figure out their cut-off date."

Duvenaud announced Talkie in a tweet on April 27, 2026, alongside collaborators Alec Rad and status_effects. The model is open-weight, meaning its code and weights are publicly accessible, and it was trained on a newly curated dataset of pre-1930 texts.

View the original tweet

Talkie’s Quirks: Temporal Leaks and Anachronistic Answers

Talkie isn’t flawless. Researchers note "temporal leakage," where the AI produces answers that betray its pre-1930 training. For example, it correctly states that Franklin D. Roosevelt was U.S. president from 1933 to 1937—a fact it couldn’t have learned from its training data. This highlights the challenge of maintaining a strictly historical dataset.

Fascinating Questions: Can Talkie Predict the Future?

The researchers behind Talkie pose intriguing questions: What is a language model’s ability to predict future events? Could an AI trained up to 1911 independently discover General Relativity, as Einstein did in 1915? These questions remain unanswered, but early tests show promise—and limitations.

• Programming: Talkie can generate one-line programs, though its capabilities are still rudimentary.
• Historical Surprise: When tested on historical events from the New York Times "On This Day" section, Talkie found events post-1930 more "surprising," particularly in the 1950s and 1960s.
• Extrapolation: In user tests, Talkie predicted a second World War would begin in 1936 and that "flying machines" would become common transport. It also humorously claimed the sun would "cease to shine" by 1999.

In another test, Talkie dismissed "talking pictures"—the slang term for sound films—as "overrated." It predicted they would never replace silent films but might supplement them, possibly shown simultaneously in theaters. "At present, however, they are interesting chiefly as a novelty," it concluded in its characteristically verbose style.

What’s Next for Talkie?

While Talkie’s limitations are evident, its potential sparks curiosity. Could a model trained on early 20th-century data make scientific breakthroughs or learn modern programming languages? For now, the answers remain speculative, but Talkie offers a fascinating glimpse into the past—and the future of AI.