SPHERESHIFT LABS
A research program and a thesis. No one can do this yet. The first rung is within reach, and it is open.

01 / 07 · The test

Before I tell you anything, try this.

See twelve symbols for six seconds. Put them back in order. Then try again wearing a halo that remembers perfectly and decides nothing.

02 / 07 · The idea

General intelligence already exists. It is us.

Some instruments supply superhuman pieces. The person supplies goals and judgement.

Chess engineSees positions no person can calculate alone.
Protein foldingPredicts structures at a speed no manual method can match.
Weather modelProcesses observations across a planet.

An instrument makes a person stronger without becoming the person.

03 / 07 · A line in the sand

The Instrument Standard

Seven tests that separate a tool from a mind.

  1. It waits. It does nothing unless a person calls it.
  2. It forgets. It keeps nothing between uses unless the person saves it.
  3. It has one job. One kind of thing goes in, one kind comes out.
  4. It cannot act. It reports to the person. It never reaches into the world on its own.
  5. It can be checked. Every answer can be tested by something independent of it.
  6. It cannot change itself. People improve it, in the open, between uses.
  7. It comes off. The person can stop, undo, or walk away at any time, and is still whole.

Fail one, and it is not an instrument. It is something else, and it should be named as something else.

04 / 07 · Open research

Six rungs. All open.

Wear one. Build a better one. Repeat. The loop always passes through a person.

05 / 07 · Two paths

Two ways forward.

Build a new mind

Develop independent general-purpose intelligence. Questions include alignment, oversight, autonomy and how to share the benefits.

Raise the ones we have

Build bounded instruments that people choose, inspect, remove and improve. Measure the person before, during and after using them.

SphereShift Labs chooses to research the second path. It is a thesis to test, not a claimed result.

06 / 07 · In the toolbox

AI is a tool in the toolbox.

People choose the purpose and keep the final say. These are the working tools named by the program founder.

Hard thinking

Claude Fable 5.2
Claude Opus 5.5
Codex Sol 6.1
Argon 4

Heavy lifting

ChatGPT Sol 6
Claude Sonnet 5.5

The loop always passes back through a human.

07 / 07 · The open invitation

Founding researchers and contributors.

Help design the ruler. Test an instrument. Show where the theory is wrong. Leave a trace another human can check.

The first study we want to runFour weeks with one thinking tool. Volunteers are measured with it on, with it off, and after it comes off. The prediction is sealed before the first volunteer starts.

Pick an open seat to apply, or see everything a contributor gets.

For the many, not just a few.

No public trial or study outcome is being claimed here. Head: a 3D scan of Lee Perry-Smith by Infinite-Realities, used under CC BY 3.0, via the McGuire Computer Graphics Archive.