Methodology
The whole site boils down to one question: when prominent people make confident public claims about the future of AI, how often are they right? Answering that fairly takes rules. These are ours.
What counts as a prediction
- It’s public and on the record. If we can’t cite it, we can’t track it. Everything here comes from a publication, a talk, an interview, an earnings call, or an official post.
- It can fail. A real prediction names a concrete outcome and a timeframe. “AI will change everything” tells us nothing. “AI will outperform radiologists within five years” is a prediction, because the world gets a vote on whether it was right.
- The person matters. We track people whose words move markets, policy, or public opinion: researchers, executives, investors, officials, and journalists. Your uncle’s AGI timeline is probably fascinating, but it doesn’t go on the board.
How we score
It happened, substantially as described, before the deadline.
Right direction, wrong size or wrong date. Or true, but only in a narrower way than claimed.
The deadline came and went. The thing didn't happen.
The clock is still running. We wait.
The success rate
The headline number on the front page is the share of resolved predictions that came true, with a partially correct call worth half a point. Predictions still on the clock don’t count until their deadline passes. Being early isn’t the same as being wrong, so nobody gets marked down while their time is still running.
Where to be skeptical
- Selection bias is real. Memorable predictions get written down, and spectacular misses are the most memorable of all. This is a curated record, not a random sample of everything anyone has ever said about AI.
- Judgment is involved. Some claims are fuzzy at the edges, and reasonable people can read them differently. That’s why every verdict comes with a note explaining the call and a source you can check yourself. If you think we got one wrong, tell us.
- Odds don’t get graded. When someone says there’s a 10 to 20 percent chance of something, one outcome can’t prove them right or wrong. We keep those claims for the record and leave them unscored.