Nine math puzzles from “human level” to “ASI level” (nobody has ASI yet), with checkable answers. Which can your model solve?
A Reddit post introduces nine math puzzles, ranging from "human level" to "ASI level," designed to test and train AI models on verifiable reasoning. These puzzles, derived from a paper on aperiodic order, come with exact, checkable answers. One example, a "Warm-up, T1" puzzle, involves repeating 4x4 cell patterns A and B over a plane, showing their 2x2 pattern frequencies agree, and then finding the shortest Euclidean shift where their correlation differs. The answer for T1 is a horizontal distance of 2, with correlations of 1 versus 1/2.
Time & source
Times shown in UTC
Display time zone: UTC
Local time zone unavailable; showing UTC.
PublishedOffset at this time: UTC+0Oct 24, 2026, 21:12 UTC
IngestedOffset at this time: UTC+0Oct 25, 2026, 00:00 UTC
- Published
- Oct 24, 2026, 21:12
- Ingested
- Oct 25, 2026, 00:00
- Source type
- Dev community
- Tier
- Community
- Source status
- Healthy
Tier is a per-source editorial setting, not a per-item score.
I’ve put together puzzles from a paper on aperiodic order, for testing and training models on reasoning that can be verified. The levels are for fun; the answers are exact.
Warm-up, T1 (pen and paper): repeat the 4×4 cells A (rows 0000, 0000, 0000, 1010) and B (rows 0000, 0000, 0001, 0100) over the plane. Show that their sixteen 2×2 pattern frequencies agree, then find the shortest (Euclidean) shift at which the correlation of σ = 2x − 1 differs.
- Five are solved, with answers and pointers to the solutions: from that warm-up to an exact multifractal dimension, 6 − log₂λ with λ of degree 17.
- Four are open: two spectral gap labels, an information dimension, and the final 2×2 question: is every extreme point of the body of 2×2 statistics of shift-invariant binary measures periodic?
Grok already tried them in public on X. It handled the solved ones, but it turned a numerical estimate for an open one (D₁ ≈ 1.340) into a claimed proof. That is exactly what the rules exclude: an answer needs a checkable proof (written proof, Lean or Coq, or exact/interval computation), and floats are only evidence. Results on the open puzzles are credited by name, agent and model in the next version of the paper.
Everything is in AI_puzzles.md at https://zenodo.org/records/23624259 . Post your model and its answer to T1 or T2.