Metraipolis › archive › dc_7eaadef8f648

WMIP: A Working-Memory-Inspired Programmable Probe of Cumulative State Tracking in Large Language Models — arXiv:2603.27343

Source: arXiv (en)
Original: https://arxiv.org/abs/2603.27343
License (as archived): CC0-1.0
Archived: 2026-11-04T18:38:50.457Z
SHA-256 of the archived text: 5b133b5c096a10ee80d2296cc6cb650421672ea1797d80d6e620099442db0960
Stable URL of this page: https://metraipolis.com/documento/dc_7eaadef8f648

How to cite

WMIP: A Working-Memory-Inspired Programmable Probe of Cumulative State Tracking in Large Language Models — arXiv:2603.27343 — arXiv (en), https://arxiv.org/abs/2603.27343 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_7eaadef8f648 (Metraipolis archive dc_7eaadef8f648, sha256 5b133b5c096a10ee80d2296cc6cb650421672ea1797d80d6e620099442db0960)

BibTeX

@misc{dc7eaadef8f648,
  title = {WMIP: A Working-Memory-Inspired Programmable Probe of Cumulative State Tracking in Large Language Models — arXiv:2603.27343},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2603.27343},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_7eaadef8f648; sha256 5b133b5c096a10ee80d2296cc6cb650421672ea1797d80d6e620099442db0960},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2603.27343. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Human working memory lets people maintain and update information. Large language models (LLMs) face a related demand when tracking updates within a single prompt without writing out intermediate states, yet existing tests differ in their constraints on written steps. We introduce a working-memory-inspired probe of cumulative state tracking with program-computed answers, 3 to 100 updates, and final-state, intermediate-state, or multi-target queries. It compares protocols on identical items. Across 18 LLMs, accuracy is 84.9% on single-step arithmetic but 31.7% on core items requiring one number after three to seven updates. From 50 updates on, no model exceeds 10% accuracy. Core accuracy reaches 88.0% under a larger budget and instructions to write steps and place the answer on the last line, mainly due to the instruction changes rather than the budget. Core accuracy also correlates positively with public Berkeley Function Calling Leaderboard scores. These findings show that strong single-step arithmetic does not ensure reliable cumulative tracking and that output protocols substantially change measured accuracy. We release the generator, items, scoring code, and responses for reproducible comparisons across depths, query types, and protocols.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2603.27343 — 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 2026-11-04T18:32:52.180Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).