Saralnama
A hobbyist developer and computer science student at Muhlenberg College, Hayk Grigorian, created TimeCapsuleLLM, an AI language model trained exclusively on London texts from 1800 to 1875 to emulate Victorian-era English. During testing, the AI unexpectedly generated a passage referencing the 1834 London protests and Lord Palmerston, events Grigorian had not explicitly taught the model but later confirmed through research. The protests followed the Poor Law Amendment Act 1834, and Palmerston was Foreign Secretary at the time. TimeCapsuleLLM was trained from scratch on about 6.25GB of period texts, avoiding modern vocabulary to capture authentic historical language and reduce confabulations. Grigorian’s project demonstrates how small, focused AI models can reconstruct coherent historical narratives from scattered data, offering potential tools for historians and digital humanities researchers to explore past linguistic patterns. The developer plans to expand the approach to other cities and languages and has made his code and models publicly available on GitHub. This experiment highlights an emergent ability of AI to connect historical facts without explicit programming, described by Grigorian as a form of “digital time travel.” (Updated 23 Aug 2025, 05:23 IST; source: link)