The public record for predictions
Put your prediction on the record.
Lock what you said, when you said it, and the test for whether it came true — before the outcome is known. Make the prediction here, or register one made anywhere else.
Browse the record · No account needed to read or verify anything.
On the record
“If there exist AI systems that can perform 8 of the 10 tasks below by the end of 2027, as determined by our panel of judges, Gary will donate $2,000 to a charity of Miles’ choice; if AI can do fewer than 8, Miles will donate $20,000 to a charity of Gary’s choice.”
MET if The panel of judges mutually agreed by Gary Marcus and Miles Brundage determines that AI systems existed by the end of 2027 that can perform at least 8 of the 10 tasks listed in the 30 December 2024 post (movie comprehension, novel comprehension, biographies/obituaries, video-game mastery, legal briefs, 10,000+ line bug-free code, Pulitzer-caliber books, Oscar-caliber screenplays, Nobel-caliber discoveries, formalization of math proofs), where tasks need not have been actually attempted commercially but must be technically feasible with modest per-task cost (roughly $1 million of compute plus 10 person-weeks), and where the tasks are not performed by 10 completely separate bespoke systems. at any point, before 2027-12-31 23:59 (UTC).
The prediction doesn’t have to start here.
A call made on X, a podcast, YouTube, Substack, CNBC — or here — becomes the same locked record: the exact words, the original source, the date, and the test that settles it. You said it first? Now you can prove it.
How it works
- Put it on the record. Make your own prediction, or paste a public source where someone else made one.
- Lock the test. Thonest works out what would settle it, by when, and from which source — and shows you that in plain English. You confirm before anything locks.
- See what happens. Predictions settle the moment their locked condition becomes conclusive. Nobody moves the goalposts afterwards, including us.
Why this is worth more than a screenshot
A screenshot proves nothing — anyone can fake one, and the post behind it can be edited or deleted. Every record here carries four things instead, each checkable by someone who thinks we’re lying.
That link goes to a block explorer, not to us. Recompute any record’s proof in your browser →
What we deliberately don’t claim: none of this proves a statement is true. Provenance shows where it came from; anchoring shows it hasn’t changed. Anyone who tells you a blockchain makes something true is selling you something. The full argument, including what could still go wrong →
Coming due
Recently settled
“no that is not true. s2f model is a simple linear regression. parameters, stats and bounds are in the article, everyone can replicate and verify. I will call s2f invalidated if we have not reached 100K by Dec this year, we can not stay at current levels for the rest of the year.”
“I will take that bet. You buy 1 BTC. I will send $1M USD. This is ~40:1 odds as 1 BTC is worth ~$26k. The term is 90 days.”
Recently registered
“In the midterms, AI will be an election issue for first time.”
“No country will take a decisive lead in the GenAI “race””
“Human domestic robots like Optimus and Figure will be all demo and very little product.”
“We won’t get to AGI in 2026 (or 7).”
“1. Some form of AI, probably neurosymbolic in nature, will come that is far more economical and data- and energy-efficient than LLMs, and it will make an absolute fortune.”