This repository is a public archive of structured future visions generated by contemporary large language models.
The project asks different LLMs the same set of future-oriented questions, using a stable prompt structure. The goal is not to claim scientific authority, but to preserve a time capsule: what did current AI systems expect, overestimate, underestimate, avoid, or frame as plausible at a specific moment in time?
The source language of the archive is German. The root README and the wiki README are in English so the purpose is easier to understand internationally; a German README with the same content is available in README.de.md. The collected model responses and future additions will remain in German.
prompts/contains the prompt templates used to generate and later review the responses.2026/contains the first collection of model responses.INDEX.mdgives a topical overview of the archive.
The 2026 archive currently contains six model sets, each with eleven topic cards:
- Claude Opus 4.7
- Crush Grok 4.3
- Crush Qwen3.7
- DeepSeek V4 Pro
- GPT-5.5
- Gemini 2.0 Flash Experimental
The first set covers eleven topics:
- AI
- Robotics
- Medicine
- Life expectancy
- AI companionship
- Work
- Society
- Energy
- Resources
- Nutrition
- Psychology in transition
Each model was asked to answer the same structured prompt for each topic. The responses include:
- context and model metadata
- a short thesis
- predictions for 1, 3, 5, and 10 years
- everyday impacts
- opportunities and risks
- assumptions
- blind spots and uncertainties
- what the model may overestimate or underestimate
- notes on what seems typical for the model or its time
Later reviews can compare the predictions against actual developments and against newer model generations.
This is an independent personal archive, not a scientific study. It is intended as open material for interested readers, analysts, researchers, and anyone curious about how AI systems imagine the future.
Ideas for additional analysis are welcome, but the project is maintained independently. The archive is published for unrestricted public reuse; no permission request is needed.
- Archive texts, prompts, indexes, and future model cards are written in German by default.
- English documentation exists to make the archive discoverable and understandable internationally.
- Translations, summaries, datasets, analyses, and derivative works are explicitly welcome.
- If a translation or derivative work changes meaning, the change belongs to that derivative work, not to the original archive entry.
The repository is meant to be easy to copy, quote, mine, translate, mirror, or republish. Attribution is appreciated as a courtesy when useful, but it is not required.
The texts are AI-generated or AI-assisted language artifacts. They should be read as historical model outputs and reflective archive material, not as factual forecasts or professional advice.
The material is dedicated to the public domain under CC0 1.0 Universal. See LICENSE.