A 13 billion parameter artificial intelligence model, trained on data from the 1930s, has produced surprising predictions about future travel times, including the notion that flying to China could take a week. The model, which was discussed on Hacker News and YouTube, appears to reflect the idealized, utopian visions of technology common during that era.
The AI, described as a "vintage AI" model, was trained on historical texts that imagined a future of rapid technological progress. When asked about travel times, it suggested that a journey to China would require about a week—an estimate that aligns with early 20th-century expectations of air travel speed, but is far slower than modern reality.
The discussion around this model has sparked debates about "temporal leaks" in AI training data and the computational requirements (VRAM) needed to run such models. The experiment highlights how AI can inadvertently preserve and reflect the assumptions and limitations of the historical data it learns from.