Metraipolis › archive › dc_94d2d22d471f

Moving the Mean Toward the Known Good, Not Beyond It: What Inference-Time Interventions and Weight Consolidation Buy in Open-Ended Generation — arXiv:2608.28886

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.28886
License (as archived)
CC0-1.0
Archived
2026-10-17T07:47:21.286Z
SHA-256 of the archived text
25ac75c9e7c8388f49e7f067b99e382dd6bf94087a2786228e99d263e0d89510
Stable URL of this page
https://metraipolis.com/documento/dc_94d2d22d471f

How to cite

Moving the Mean Toward the Known Good, Not Beyond It: What Inference-Time Interventions and Weight Consolidation Buy in Open-Ended Generation — arXiv:2608.28886 — arXiv (en), https://arxiv.org/abs/2608.28886 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_94d2d22d471f (Metraipolis archive dc_94d2d22d471f, sha256 25ac75c9e7c8388f49e7f067b99e382dd6bf94087a2786228e99d263e0d89510)
BibTeX
@misc{dc94d2d22d471f,
  title = {Moving the Mean Toward the Known Good, Not Beyond It: What Inference-Time Interventions and Weight Consolidation Buy in Open-Ended Generation — arXiv:2608.28886},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.28886},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_94d2d22d471f; sha256 25ac75c9e7c8388f49e7f067b99e382dd6bf94087a2786228e99d263e0d89510},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.28886. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

What does a generation loop gain from learning on its own verified successes? In cycles of generate, verify, select and LoRA-consolidate on online bin packing, training on value-filtered candidates shifts what the model writes on held-out variants toward value (-1.7 points of excess, p=0.008; -3.1 against a random-consolidation control, p=0.004) while the best observed candidate converges to the classic heuristic's level and no further. A confirmation battery replicates the whole procedure three times, with fresh seeds and a never-consulted held-out set read exactly once: the mean was nearly identical in all three lineages (-2.0, -1.8, -1.9), and after aggregating within held-out variant all seven evaluable variants favored consolidation (p=0.008). The best observed candidate moved to the classic heuristic's level, exactly (0.021028 in all three lineages, for attract and for the random control alike), and never beyond it. A matched SFT-only control shows the supervised anchor, not repulsion from bad candidates, does the concentrating (96% of candidates land exactly at the classic heuristic's level). The tails cut both ways: consolidation lowers the per-candidate rate of better-than-classic candidates (10% to 3.9%) while its larger production yields more such candidates absolutely (5 against 1, on few events). As motivation we report the inference-time ledger that led here: a model-written schematic recap buys judged document integration and nothing buys development; a verifier written into the stream is imitated, 16.4 fabricated verdict lines per notebook. Mean quality among valid candidates can be bought and replicated; the observed best goes to the classic and, so far, never beyond it.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-01 in cs.AI).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2608.28886 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-17T07:28:26.699Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).