Which end of the world should you actually worry about?

How it works

The short version

We keep a list of end-of-the-world theories, the serious ones and the very online ones. Each theory gets pulled through real public data. Then AI models read that data and tell us how plausible they think the theory is.

It is a fun way to look into your own fears and ideas, with real data doing the heavy lifting. It is entertainment plus real data. It is not a prediction service, and it is not advice.

The question every model gets

Every theory is written down as one precise event, with its own scale. Not just 'a pandemic' but 'a new pathogen kills at least 10 million people'. The models rate that exact event, not a milder version of it.

The window is always the same: the next 25 years. For every theory, every time. A model gives a likelihood from 0 to 100, and that is the chance it thinks the event, as described, happens within 25 years.

Each model also says the soonest it could plausibly happen, and gives three short reasons. We show those reasons as written.

The bands

The number maps to a band so it is easy to read at a glance. Negligible is 0 to 1. Remote is 2 to 9. Unlikely is 10 to 24. Possible is 25 to 44. Plausible is 45 to 64. Likely is 65 to 100.

The ranking uses the combined score, which is the average of the latest score from each model that has weighed in. 'Joining soon' means a model has not started yet.

How fair is it?

Every model gets exactly the same data summary for a theory. No model ever sees another model's answer. They work alone, like students who can't copy each other.

Models are allowed to say no. If a model declines to score a theory, we show that as a result too. We do not hide it or fill in a number for it.

Different models often disagree. That is part of the fun, and part of the point.

How often it updates

The number one theory is re-checked every 6 hours. Ranks 2 to 10 are re-checked daily. Ranks 11 to 20 weekly. Everything else monthly. A brand new theory gets scored right away.

Each theory page shows when it was last scored and when the next check is due.

Where the data comes from

Real public sources, with no private data in the mix. Depending on the theory, that includes GDELT, which tracks news coverage around the world. Wikipedia page views, which show what people are curious about right now. Hacker News story counts, how many stories with a set phrase in the title were posted in the last week (via Algolia). NASA asteroid tracking. NOAA space weather. USGS earthquake and volcano data. CISA's list of known exploited software vulnerabilities. For the UFO theories, a few official figures that we copy by hand and date: AARO (US Department of War), the Pentagon's UAP office, and War Department PURSUE, the released UAP records.

A theory only uses the sources that actually apply to it. Where a source is down on the day, we say so rather than guess.

What this is not

It is not a forecast. The models are making informed guesses about very rare, very uncertain things. A high band means a model finds it more plausible, not that it will happen.

Nothing here is medical, financial or safety advice. If something worries you, talk to a real person who knows the field.