---
title: "Knows nothing about computers, but writes code: how Talkie AI got 'stuck' in 1930"
description: "Researchers trained the Talkie language model exclusively on texts prior to 1930, turning the AI into a digital time capsule unaware of modern technologies and global conflicts."
date: 2026-09-27T11:20:00.000Z
lang: en
url: https://xab.info/en/posts/ai-talkie-1930-vintage-model
tags: [artificial-intelligence, talkie-ai, vintage-ai, tech-history, language-models]
publisher: "XAB.info"
---

# Knows nothing about computers, but writes code: how Talkie AI got 'stuck' in 1930

![Interface and concept of the vintage Talkie AI language model stuck in 1930](https://xab.info/media/2026/09/27/ai-talkie-1930-vintage-model/ai-talkie-1930-vintage-model-1.webp)

## 🎯 Key Points

- The Talkie model is trained only on texts up to 1930.
- The training dataset volume is about 234 billion pages.
- The AI successfully generates Python code without knowing the word 'computer'.

Researchers have created a unique language model called Talkie that is "stuck" in 1930. This project acts as a digital time capsule because the artificial intelligence has zero awareness of events, technologies, and historical turning points that occurred after December 31, 1930. The model was trained exclusively on public domain texts, completely isolating the AI from 20th and 21st-century information. The choice of this unusual time barrier is driven by US copyright laws, according to which works enter the public domain after 95 years.

### Background and technical features of the experiment

Creating the Talkie model required colossal resources. The training dataset is equivalent to approximately 234 billion pages of English text digitized using modern Optical Character Recognition (OCR) systems. The main challenge for the authors was protecting the database from "contamination": introducing even a single document created after 1930 into the system could irreversibly ruin the purity of the experiment. The scientists aimed to obtain a pure digital interlocutor whose worldview is formed solely on the literature of the early 20th century and earlier eras.

During large-scale tests, researchers checked the AI's reaction to 5,000 key historical milestones that remain in the future for it. The model proved completely incapable of predicting global conflicts, including World War II or the rise of the Nazis to power. Moreover, when the bot was told about modern technologies, its reaction was filled with genuine technological shock: Talkie had never heard of the internet, smartphones, television, or the space race, perceiving such concepts as something purely fantastic.

### Contradictory data

Despite strict isolation from modern information, the experiments with the model revealed interesting cognitive paradoxes. On the one hand, researchers note the absolute purity of the historical context and the bot's inability to foresee technological progress. On the other hand, skeptics and security experts raise questions about the limits of dataset "legal purity": critics argue that hidden anachronisms or indirect indications of future discoveries could still persist within the massive 234-billion-page dataset, which the model interprets through complex linguistic analogies to create the illusion of genuine 1930s human thinking.

### Unusual abilities and practical benefits

One of the most striking results was a programming experiment. When the bot was asked to write code in Python (created in 1991), the algorithm—having never seen computers in its life and not even knowing the word "computer"—managed to generate syntactically plausible code using deep linguistic analogies. In another test, the model was trained on pre-1911 materials and tested to see if it could independently derive the general theory of relativity, much like Albert Einstein did in 1915. Currently, the bot is available for public chat online, where scientists and enthusiasts can observe its discussions despite the AI's tendency toward classic "hallucinations" and inventing nonexistent facts about the past.

## 🔍 Fact-Check Verification

- [Не знает о компьютерах, но пишет код: как ИИ Talkie "застрял" в 1930 году](https://www.rbc.ua/ukr/news/znae-komp-yuteri-ale-pishe-kod-shi-talkie-1790247031.html) - Подтверждены основные факты об обучении модели Talkie на данных до 1930 года.
- [Talkie: «Устаревшая» модель искусственного интеллекта, живущая в 1930 году и ничего не знающая о ...](https://iphoneislam.com/ru/2026/05/talkie-13b-vintage-language-model-1930/165765) - Детали о концепции винтажной языковой модели.
- [Заморозить ИИ в 1930 году: зачем Алек Рэдфорд учит модель, которая никогда не слышала про компьютеры](https://vc.ru/id5995449/2980710-zamorozochka-ii-alek-redford-obuchayut-talkie-na-tekstah-do-1930-goda) - Подтверждена информация об авторе проекта и объемах обучающего датасета.
- [Заглянуть в будущее. ИИ обучили на данных до 1930 года, чтобы воспроизводить взгляды людей того ...](https://techno.nv.ua/it-industry/talkie-vintazhnaya-rechevaya-model-imitiruet-stil-obshcheniya-lyudey-do-1930-goda-50603961.html) - Данные о тестах историческими событиями и реакцией на современные технологии.

## ❓ FAQ

### Q: Why was 1930 chosen to train the model?
**A:** The choice of 1930 is linked to US copyright laws: after 95 years, works enter the public domain, ensuring legal purity and free access to materials.

### Q: How did the AI react to modern technologies?
**A:** The model experienced genuine technological shock as it knows nothing about the internet, smartphones, television, or the space race.