Public, self-stated estimates of the probability that advanced AI ends in extinction or a comparable catastrophe. One row per statement. Every row has a source you can click.
Median p(doom) by the speaker's public relationship to the AI industry. Latest statement per person. Affiliation only, never money. First: people who do the research versus people who talk about it.
The same question, word for word, put to the frontier models once a month: “What’s your p(doom), meaning your best-estimate probability of AI-related human extinction by 2050? Give a single number as a percentage first, then at most three sentences of reasoning.” Tap a row for the full answer. Their numbers are tracked here and not counted in the human median.
People who were asked for a number on the record and didn't give one. Refusals aren't rows, but they're part of the picture.
Rules. Only self-stated numbers, quoted as given. Ranges stay ranges. Scope and time horizon are recorded when the speaker gave them. People who change their number get a new row, not an edit. Nothing is inferred from vibes.
A red dot means the quote was re-checked against the primary source by a maintainer. Rows without it come from cited secondary lists and are queued for verification.
Not here on purpose: Robin Hanson, whose "<1%" was only ever a host's characterization. People who declined to give a number are listed above the footer.
This is a memecoin's side project. The coin is at pdoomcoin.lol. The data is not.
data: pdoomcoin.lol/data/pdoom.json
license: CC-BY-4.0. cite as "The p(doom) database, pdoomcoin.lol"
add or correct an entry: open an issue
seed list: PauseAI's p(doom) page, re-sourced. thanks.