These findings motivate a state-action controller that starts from the base model action and selectively edits it using structural readouts, without requiring a complete predicted belief state as an intermediate representation.
Abstract
Task-oriented dialogue requires maintaining and updating information across turns, yet language models expose no explicit belief-state object. We study how conversational state is represented, updated, and used inside eight instruction-tuned language models from four families on MultiWOZ and SGD. Structure and values separate: which domains, slots, and requests are active is linearly readable just before the model acts, whereas exact values are far more readable where the user stated them. After a user changes a value, both values remain accessible at their mentions, and causal interventions show that both continue to influence the model's action. In natural closed-loop interaction, query failures separate into cases of weak structural support, incorrect value resolution, and failure to deploy otherwise-supported constraints, with targeted interventions producing systematically different repair behavior across these cases. These findings motivate a state-action controller that starts from the base model action and selectively edits it using structural readouts, without requiring a complete predicted belief state as an intermediate representation. On held-out MultiWOZ interaction across five models, it raises the base model exact-query accuracy from .318 to .621 and task success from .272 to .371 at negligible added cost. Overall, reliable interaction requires not only retaining conversational information, but resolving which available constraints currently apply and ensuring that they govern action.
The results suggest that broad SFT brings most of the model's capability improvement; turn-local supervision can be effective when failure detection is precise, with observed transfer concentrated primarily within-family.
When a large language model handles a multi-turn task and a user proposes a change but ultimately rejects it, the model should continue as if nothing changed. We find a surprising failure: merely mentioning a rejected change can derail task execution, even when the user's final intent remains unchanged. To systematical...
Junle Chen, Wei Chen, Zheng-Jun Huang et al.· 0 citations
The results suggest that authority computations may be task-dependent, rather than reusable across tasks, and suggest that authority computations may be task-dependent, rather than reusable across tasks.
Benjamin Shih, John Winnicki, Arianna Cao· 0 citations
Findings show that historical evidence use should be evaluated under state variation, rather than inferred from outcome-only correctness, and that open-weight models rely heavily on context continuity and are reluctant to spontaneously switch to tool-mediated retrieval when provenance fails.
Yu Liu, Wen-Xiao Zhang, Cheng Hu et al.· 0 citations
Research for interactive LLMs from retaining more context toward preserving uncertainty is redirected from retaining more context toward preserving uncertainty, to motivate uncertainty-preserving state management.
Jian-Zhe Lin, Xiao-Lin Li, Fei Wang et al.· arXiv.org· 0 citations
It is demonstrated that valence-related activation patterns leave hidden traces that predictably govern later choices, even when all visible tokens are identical across conditions, and whether these traces are accompanied by any subjective experience relevant to model welfare remains unclear.