Metraipolis › archive › dc_2c1fecc72f92
Sequential knowledge editing breaks a model's ability to tell good evidence from bad, without costing it accuracy — arXiv:2609.29587
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.29587
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-08T04:26:16.502Z
- SHA-256 of the archived text
bffbcbf6c2ef92d067416f16bd0d49aed33d59a8166ce6b9f4559c1621cde885- Stable URL of this page
https://metraipolis.com/documento/dc_2c1fecc72f92
How to cite
Sequential knowledge editing breaks a model's ability to tell good evidence from bad, without costing it accuracy — arXiv:2609.29587 — arXiv (en), https://arxiv.org/abs/2609.29587 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_2c1fecc72f92 (Metraipolis archive dc_2c1fecc72f92, sha256 bffbcbf6c2ef92d067416f16bd0d49aed33d59a8166ce6b9f4559c1621cde885)
BibTeX
@misc{dc2c1fecc72f92,
title = {Sequential knowledge editing breaks a model's ability to tell good evidence from bad, without costing it accuracy — arXiv:2609.29587},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.29587},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_2c1fecc72f92; sha256 bffbcbf6c2ef92d067416f16bd0d49aed33d59a8166ce6b9f4559c1621cde885},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.29587. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Knowledge editing is evaluated on whether the edited fact changed, whether paraphrases follow, and whether unrelated answers stayed put. A model can pass all three and still lose something none of them measures: the ability to decide, on facts that were never edited, which retrieved documents to believe. We score the log odds a model assigns to its remembered answer against the answer an injected passage asserts, before and after editing, holding the query, the passage and both candidate strings fixed. Our cleanest arm is a conservatively tuned LoRA: after 1,000 sequential edits on Qwen2.5-7B-Instruct it leaves MMLU unchanged to four decimal places, yet the spread of the arbitration quantity across untouched facts falls by 36%. Selective prediction degrades with it. Area under the risk-coverage curve rises by 0.107, against 0.005 for a norm-matched perturbation at the same MMLU, and error on the model's most confident quarter of arbitration decisions goes from 0.217 to 0.342. This is not capability loss. Sweeping random perturbation over five severities, damage bad enough to cut MMLU from 0.6275 to 0.3725 produces less harm (0.088) than MEMIT does at 0.6050 (0.102). The effect holds across three seeds, two model families, two datasets, two probe-disjointness criteria, three prompt templates and paraphrased queries. Layer ablation on saved weight deltas shows it is distributed: no single layer reproduces it, and removing any one recovers about half. Under retrieval with a frozen retriever, accuracy falls from 0.592 to 0.46. A secondary finding may matter more in practice. Three of five model and method pairings we ran collapse to chance MMLU at 1,000 sequential edits under published hyperparameters, while edit success stays at 1.00 and locality reads clean. Sequential-editing evaluations that never measure capability cannot see this. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-25 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.29587 — 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 2026-10-08T03:50:57.804Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-25?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).