Metraipolis › archive › dc_bd75358954ec

What if LLMs Ate Their Words: Causal History Effects in Multi-Turn Interaction — arXiv:2609.05882

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.05882
License (as archived): CC0-1.0
Archived: 2026-10-16T12:34:46.227Z
SHA-256 of the archived text: 1f9d26cc74dbe5944997c5785480f3f1c3630ed76ec3a6a288cc8b32fc5388c2
Stable URL of this page: https://metraipolis.com/documento/dc_bd75358954ec

How to cite

What if LLMs Ate Their Words: Causal History Effects in Multi-Turn Interaction — arXiv:2609.05882 — arXiv (en), https://arxiv.org/abs/2609.05882 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_bd75358954ec (Metraipolis archive dc_bd75358954ec, sha256 1f9d26cc74dbe5944997c5785480f3f1c3630ed76ec3a6a288cc8b32fc5388c2)

BibTeX

@misc{dcbd75358954ec,
  title = {What if LLMs Ate Their Words: Causal History Effects in Multi-Turn Interaction — arXiv:2609.05882},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.05882},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_bd75358954ec; sha256 1f9d26cc74dbe5944997c5785480f3f1c3630ed76ec3a6a288cc8b32fc5388c2},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.05882. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Multi-turn interaction creates a feedback process in which an LLM's previous responses become context for later behavior. Prior work shows substantial multi-turn degradation and that assistant-generated history can affect later behavior. However, it remains unclear how these effects manifest across models, tasks, turns, and inside a model. We study these gaps across six task families and five models. Degradation from fully specified single-turn input (FULL) to progressively revealed multi-turn interaction (SHARDED) is clearly task- and model-dependent, and stronger one-shot performance does not imply greater interaction robustness. We then retrospectively analyze completed SHARDED conversations by replaying the user messages already observed in each trajectory while editing only assistant-generated history. Replacing prior assistant responses with neutral content (termed neutralization) changes downstream min-max normalized performance by +.027 across 2,973 trajectories. On a prespecified length-controlled subset, short and length-matched neutralization yield nearly identical effects (+.069 versus +.068), showing that simple context shortening is insufficient to explain the effect of history editing. Turn Surgery further intervenes on one assistant turn at a time. Among 237 selected degraded trajectories, 63.7% contain at least one beneficial intervention, while most tested positions remain unchanged; for binary tasks, 48.4% admit a fail-to-success reversal. An open-weight case study links behaviorally consequential history changes to measurable downstream state differences, but finds task-dependent rather than universal internal signatures. Overall, assistant-generated history has active but selective effects on multi-turn performance, motivating selective rather than uniform history management.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-09 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.05882 — 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 2026-10-16T12:30:24.991Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-09?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).