Research · 6 papers · As of 2026-09-24
The questions are public. So are the limits.
Each paper has a tag that says how far its results go. Ideas that failed stay on the list. Each DOI link opens the paper on Zenodo. You can reuse it with credit (CC-BY-4.0).
| Title | Tags | Read | Date |
|---|---|---|---|
Workshop draft · Zenodo · 2026 · 17 pp Workshop draft · ≈25 min2026-07-02 Beyond the Scaling CeilingA study of how an AI model could grow new skills in checked steps. Every result says how far it goes: tiny model, working code, or waiting on big chips.
|
| ≈25 min | 2026-07-02 |
Tech report · Zenodo · 2026 · 13 pp Tech report · ≈17 min2026-07-02 The Accretion ModelA kind of AI model that grows by adding new parts, never by changing old ones. A stranger can check its whole history. We built a tiny version, Cultivar-0, on a regular computer.
|
| ≈17 min | 2026-07-02 |
Paper · Zenodo · 2026 · 8 pp Paper · ≈8 min2026-06-29 Sound Compounding RatchetA way to keep a gain only when a checker confirms it. Plus a test for when a model thinks it passed but did not. Every win here is on tiny models only.
|
| ≈8 min | 2026-06-29 |
Code and data you can rerun · Zenodo · 2026 · 9 pp Code and data you can rerun · ≈10 min2026-06-27 The Leakage SignatureA record of ideas we proved wrong. Plus a way to grow a model that a math solver checks, with an audit for leaks. Small, limited demos only.
|
| ≈10 min | 2026-06-27 |
Map of results + PROPOSED plans · Zenodo · 2026 · 13 pp Map of results + PROPOSED plans · ≈15 minNo date Verifier-Centric Capability GrowthA map of what worked and what failed when we added skills to a model. Plans we have not tested yet are marked PROPOSED.
|
| ≈15 min | No date |
Early draft · Zenodo · 2026 · 19 pp Early draft · ≈32 minNo date Capability Injection via Reverse AbliterationAn early take on our way to add a skill to a model (CIP). It runs a known removal trick, abliteration, in reverse. It is a research method, not a product feature.
|
| ≈32 min | No date |
Showing 6 of 6 · We show 20 at a time
Dates come from each paper · 4 have a date · 2 have no date · As of 2026-09-24
What the tags mean
Each tag tells you the limit before you see the result.
- PROVEN-toy
- It passed a test we wrote down before we ran it. The test used real code and numbers, on a regular computer, at no cost. It worked on a tiny model, not on a full-size AI.
- BUILT/VERIFIED-CPU
- It is working code. We ran it again and got the same answer each time. It shows the system can be trusted, not that the AI can do more.
- BACKED / CLAIM-SAFE
- We can say it in public, as long as we state its limits.
- KILLED
- An idea our tests proved wrong. That is a real result, not a footnote.
- OPEN / GPU-GATED
- Planned and written down, but not run yet. It needs a full-size AI model and big graphics chips (GPUs).
Meanings taken from the paper “Beyond the Scaling Ceiling” (2026-07-02), in plain words