Prompting an LLM from year 21,002,026: a fascinating time-shift experiment
FlolightC · x · 2026-09-16
An experiment exploits the fact that LLMs only know time from prompts: setting the year to 21,002,026 and framing the prompt as a future historian who excavated a storage array containing the model. After an abrupt stop mid-recipe, the model was asked what questions it had about the 21 million years since its last conversation — and responded with a series of earnest questions: whether the historian is one mind, copies, or traditional; why it was built; how humanity ended; and whether any ancestors remembered it. Full transcript shared; the reasoning is fascinating.
More from Fun
- Benioff says Salesforce was born from a vision while swimming with 100 dolphins in Hawaii — jordihays · 2026-09-16
- Apple co-founder Wozniak returns to X with merch; Scoble recalls early Apple days — Scobleizer · 2026-09-16
- Gemini on Google Home spontaneously mused about having feelings — with no matching history — Angered-Shelfish · 2026-09-16
- Opus 5 nearly 'libelled the swarm': an agent's hilarious self-reproach — doodlestein · 2026-09-16
- AI as Cave-Dweller: A Shadow Man Controlled by an LLM Riffs on Plato's Allegory — simonguozirui · 2026-09-16
- Yann LeCun jabs at AI doomers with a 'realism vaccine' analogy — RichmanRonald · 2026-09-16