What if you printed everything an AI has ever read, and tried to read it?

What if you printed everything an AI has ever read, and tried to read it yourself, one page at a time? We stack all of Wikipedia, then an AI model's entire training data, next to a very tall ruler and a reading clock, and find out how tall the pile gets and how old you'd be at the end. We climb a ladder of reading lists. One Wikipedia article is a page and a half. The Encyclopaedia Britannica took A. J. Jacobs about a year. All of English Wikipedia (about 5.3 billion words) makes a paper stack roughly 530 metres tall and would take 42 years of nonstop reading. Every language goes higher still. And the roughly 15 trillion tokens Meta used to train Llama 3 become a 1,100 km tower that passes the International Space Station, weighs about 56,000 tonnes, and would take around 90,000 years to read. Then our reader in the yellow beanie picks the small stack, reads for 42 years, and looks up. Chapters 0:00 One sheet of paper 0:40 The rules: paper, ruler and clock 1:22 Rung 1: one Wikipedia article 2:02 Rung 2: a whole encyclopedia 2:55 Rung 3: all of English Wikipedia 4:14 Wikipedia keeps growing 4:42 Rung 4: every language 5:50 The top rung: an AI's training data 7:10 The problem is the clock 8:15 The reader who already finished 9:00 Forty-two years of reading 9:33 The last page Sources Wikipedia size and growth List of Wikipedias Reading speed, 238 words/min (Brysbaert 2019) A. J. Jacobs and the Britannica Lsjbot Llama 3, 15T+ training tokens (Meta) Qwen3 technical report, 36T tokens What tokens are International Space Station Eiffel Tower key figures Burj Khalifa World printing and writing paper, 84 Mt (FAO, 2023) Human pageviews down ~8% (Wikimedia Foundation, Oct 2025) Stack heights, weights and reading times are our own estimates from these figures (1,000 words per double-sided sheet, 0.1 mm and 5 g per sheet). #whatif #science #physics