Creators

Tobías is a theoretical physicist who moved into applied AI. His doctoral work was on holographic neutron stars — compact objects used to study the seam between quantum field theory and gravity — and his published research sits in general relativity, the AdS/CFT correspondence, and the large-scale structure of the universe.

This project came out of that habit of mind. A benchmark is an instrument, and an instrument you cannot keep clean stops measuring what it claims to. Published test sets are scraped and learned; the score survives and the measurement does not. The design here is an attempt at one that resists that — original items, answers that exist nowhere in plaintext, and benchmark sets used once and never released.

The full argument, with sources, is in the manifesto.

Items are authored by invitation. If you build problems for their own sake — Mensa members, puzzle constructors, mathematicians, physicists — there is a seat here, and the studio to build them in.

What it asks of you: that the items are yours, not adapted from any published test, and that you are content for the difficult ones to go into a private set that is used once and never released. What it gives back: your name here, if you want it there, and a measurement that still means something in five years.

Write to tobiascanavesi@gmail.com