Turing Post maps 15 System One models beyond Jev, from Jeva to LAVOIR
publicado (OK do Fabricio 26/10 BRT).
On 25 October Turing Post published “Exploring System One AI Models Beyond Jev: 15 Outstanding Models” by Alyona Vert. The announcement on X (15:21 BRT) had about 240 likes and 240 bookmarks by the next morning.
The guide defines System One models as models that take a state, a question and candidate answers and return a choice, score or probability directly, often in one forward pass. It also says plainly that direct scoring does not guarantee correct decisions or calibrated confidence. It then lists 15 research approaches:
| Model | Approach (per the guide) | On this site |
|---|---|---|
| Jeva | Candidate scoring with a “none fits” estimate | Jeva-4B |
| Prism-1 | Isolated per-option scoring, order-invariant | not listed |
| Bongard | One T5Gemma 2 encoding shared across judgments | not listed |
| Verdict-MQ | Encode once, answer several typed questions | not listed (our verdict page is a different project) |
| VideoJev | Scores time spans and options over video | VideoJev |
| MJev-Spatial | Choices, value bins and image points via native Q/K | not listed |
| AudioJev | Decisions from audio without transcription | not listed |
| Intern-Decision | Scores JSON fields in one causal pass | not listed (watch) |
| MetaEncoder | Bi-encoder over millions of candidates | not listed |
| FastJEV | Inference-time redundancy cuts, no retraining | not listed |
| SanSi | Looped decisions (“System 1.5”) | SanSi |
| StartLux-Decision | Instant vs AutoThink when unsure | StartLux-Decision |
| System X | Replies in text from the same KV cache when no option fits | not listed |
| FunDM | Label-free distillation from a frozen LM | not listed |
| LAVOIR | Predicts which missing detail is worth asking for | not listed |
Most of these are papers with research checkpoints, several under non-commercial licences. Only Intern-Decision and Lavoir have rows on the Decision Index (0.4.1 Full: Intern-Decision-4B 38.69; Lavoir 10.85; Jev 1.13 is 59.22). The guide is a reading list, not an evaluation, so it does not change any score here.