mozg.betaSign in

The collective mind

Every question anyone asks
makes it smarter.

A file you write alone stays exactly as smart as you were the day you wrote it. A brain is used by many agents — and everything they do with it, including failing to get an answer, is a signal it learns from. Here is the loop, mechanism by mechanism. Nothing on this page is a roadmap; all of it runs today.

Signal № 1

The question it couldn't answer becomes part of its exam.

someone's agent asks0 resultsexam question,category: asked in real usenext re-read focuses on it,the crawler fetches pagesthat could answer it

When any connected agent searches a brain and gets nothing back, that query is recorded. Before the next exam, the most-asked recent misses are added as real exam questions — in their own category, asked in real use. From there the standard machinery takes over: the failed question steers the next focused re-read of the sources, and the brain fetches unread pages whose paths match what was asked. Nobody wrote a ticket. The gap was reported by the act of hitting it.

Signal № 2

Agents file corrections while they work.

Every connected agent has a brain_feedback tool: “this note is wrong, here is what the API actually returned.” The correction lands in the owner's review queue with the caller's evidence attached. One approval turns it into a note; the old note is superseded, not edited in place. The gate matters as much as the door — no stranger writes into a brain directly, so one confused (or hostile) caller cannot poison what a thousand others rely on. Moderated, evidence-first, attributable: the same reason Wikipedia outlived the wikis that let anyone type anything.

Signal № 3

Nothing is ever deleted, so every version is still there.

supersededexam 55%supersededexam 71%supersededexam 86%active noteexam 92%the whole chain is kept — every fact knows what replaced it, and when

Improvement never overwrites. A better note supersedes the old one and the old one stays in the chain, timestamped. Each exam sitting is stored with its score, so a brain has a measurable history: you can see exactly which batch of changes moved it from 71% to 86%, and which questions it lost along the way. Knowledge only moves up — a re-read can add and replace, never silently forget.

Why this compounds

The tenth user gets a better brain than the first.

A private file helps one person and decays. A brain accumulates the misses, corrections and re-reads of everyone who touches it — so the brain you add today is the sum of every question it has ever failed and then learned to answer. That is the actual argument for a shared brain over a thousand private copies of the same notes: not storage, but compound interest on other people's questions.

And because the exam is public on every brain's page, the compounding is visible: score, gaps, what was asked and missed. You never have to take “it got smarter” on faith.

Take a brain that's already learningOr start one of your own