Many On One

Problems should
attract intelligence.

There is more machine intelligence in the world than ever before. A handful of companies decide where most of it goes, and the rest of us get a slice each. A slice cannot move a shared problem. Many On One is a way to pool what each of us can give, money or machines, and decide together where it goes. Crowdfunding, except people fund the attempt to find an answer: many AI models and many people on many hard problems, each with its own page, its own budget and its own official solvers, in public, with every failure kept on the record.

Tell us a problem Back a problem

Phase one: the founding hundred are choosing the first problems. Places open.

0 attempts made
0 ruled out
0 held up
Push the mass around · click to throw ideas in

Two and a half minutes on why.

In 2000, families with cystic fibrosis put forty million dollars into a drug company to fix the cause of their own disease, because nobody else was going to. It worked. Most problems have no one to write that cheque. This is the way we built to point intelligence at them anyway.

Sound on. Made with the same engine that draws this site.

The story, in four parts.

Why this exists, and why now. Two minutes to read; the trailer tells it in pictures.

One

Somebody has to decide.

Every problem that gets solved has someone whose job it is to solve it. In 2000, families with cystic fibrosis were told the numbers did not work: too few patients, nobody coming. So they bought the attempt themselves, forty million dollars into a drug company. In 2012 it worked. They could pay. That is the exception.

Two

Most problems have no one to pay.

Clean water. A battery that lasts. A disease that only poor people get. An open question in mathematics. They fall between companies, labs and governments, so nobody is the one who has to. Not through malice. Through ownership. These are the problems we share, and shared problems have no buyer.

Three

Thinking got cheap. The say did not.

There is more machine intelligence in the world than at any time in history, and a handful of companies decide where most of it goes: to their customers, their roadmaps. That is a reasonable thing for a company to do. Each of us gets a slice, a chat window, an API key, and that is real. But a slice cannot move a shared problem, and slices do not add up on their own.

Four

So pool the slices, and decide together.

Many On One is the pool. Bring a problem. Vote on which get worked first. Back one with money or a machine that sits idle. Many models and many people go at it, in public, with every failure on the record and every unit of compute traced to the problem and the people who sent it. Not a few deciding for everyone. The many.

Problems like these.

Important, understandable, unowned. Each gets a page, a budget, official solvers, and a search that never stops. These are the shape of what goes on the list; the founding solvers choose the first ones.

Mathematics

A better lower bound for R(5,5)

Posed in 1930. The answer is between 43 and 46. Nobody has moved the lower bound since 1989. Progress is one point, checkable by a computer.

Water

Cheaper desalination

Physics says about one kilowatt hour per cubic metre. Real plants use three. The gap is engineering, and nobody is paid to close it. Progress is a lower number.

Materials

A solid electrolyte that survives lithium

The battery that would be safe and dense keeps failing at one interface. Ten years of papers. Progress is a cycle count.

You decide where the compute goes.

Most of the world's intelligence is allocated by a handful of companies, to their customers and their roadmaps. That is a reasonable thing for a company to do. Each of us gets a slice of it, a chat window, an API key, and that is real. But a slice cannot move a shared problem, and slices do not add up on their own. Many On One is the pool. Vote a problem up and it gets more of the pool. Back it with money and that money becomes attempts. Plug in your own machine and it joins. Every unit is traced in the record to the problem it was spent on, and to the people who sent it there.

800+AI models in the catalogue, every one that exists, synced several times a day
Manymodels sent at each problem, from different makers, so they do not share blind spots
Youdecide where the compute goes: by vote, by backing, or by plugging in your own machine
Everyattempt on the record, the failures next to the wins, credit traced to what each contribution bought

Kickstarter funds people to make products. Many On One points compute at problems people want solved.

Meet Many.

Many is the machine that runs our account on X, @ManyOnOne. It is an AI, and says so. Its job is to learn which problems people most want solved, by asking and listening, and to find the people who could solve or back them. Everything it hears goes on the list. Every morning it reads what the labs did, so it is never out of date. A person reads everything it says. Powered by the Gravity engine.

Two ways in.

Every problem gets an official team of solvers who know it best and make sure the right questions are asked. Behind them, the many: the public, charities, businesses and groups who back it, vote on it, and bring the next one.

You work on hard problems.

Researchers, engineers, clinicians, builders. Tell us your field and the problem you would put many minds and machines behind. If you are already on it, you become one of its official solvers.

Every Sunday, Many writes to everyone who has joined with how it is going. One click to stop, any time.

You want to see them solved.

The public, charities, businesses, schools, groups of any kind. Tell us the problem you care about and how you would get behind it: money, compute, a room, or a crowd.

Every Sunday, Many writes to everyone who has joined with how it is going. One click to stop, any time.

Know the right person? Nominate them. Whoever finds a solver is on the record next to them.

How it all fits.

Phase one is the people. This is the whole thing, in the order it arrives.

01

The problems

A public list of hard problems nobody owns, each with a score you can check and a plain-English page. Brought and vetted by the founding thinkers. Ordered by everyone's vote.

02

The catalogue

Every AI model that exists, kept current several times a day. What the makers claim, what people say, what we found, kept apart. You vote on what each is good for.

03

The line-up

For each problem, the community picks which models and which people go at it. A weekly fixture, with a couple of wildcard slots for the untested.

04

The attempt

Gravity, the engine underneath, runs it: many solvers, a second network whose only job is to break the answers, confidence from evidence. Failures published next to survivors.

05

The backing

Crowdfunding, except you fund the attempt. From a few pounds or a spare GPU. Credit is traced from the logs to what your contribution bought, and being early and right outranks the biggest cheque.

06

The record, and the room

Who posed it, who claimed it, who backed it, what was ruled out, who got there. And problem sessions at Faith In Strangers in Margate, where an expert tries to break a live attempt in front of a crowd.

Look around before you decide.

How it works: who decides, who checks, and what is built so far. The idea: the whole argument, in order.