{
  "schema_version": "1.1",
  "id": "s13:https://arxiv.org/abs/2609.15972v1",
  "slug": "2609-15972v1-1dw4a6c",
  "url": "https://feed7.dev/p/2609-15972v1-1dw4a6c",
  "title": "Mind2Dialogue: Training Human-Aware Language Models by Simulating User Mental States",
  "why_included": "Mind2Dialogue trains assistants from simulated, evolving user states that are hidden at deployment. The approach could inform persistent agents, but its reported gains concern personalization rather than coding work.",
  "summary": "Mind2Dialogue uses a psychology-guided simulator to maintain personal traits while updating a shared mental state through conversation. An Oracle assistant sees that state, and **privileged distillation** trains deployed models to respond without direct access to it.",
  "practical_implication": "Builders of persistent agents can treat inferred preferences and beliefs as uncertain training signals rather than facts stored in a profile. The evaluation joins personalization with theory-of-mind tasks across Qwen, Llama, and OLMo baselines.",
  "agent_context": "Mind2Dialogue uses a psychology-guided simulator to maintain personal traits while updating a shared mental state through conversation. An Oracle assistant sees that state, and **privileged distillation** trains deployed models to respond without direct access to it.\n\nBuilders of persistent agents can treat inferred preferences and beliefs as uncertain training signals rather than facts stored in a profile. The evaluation joins personalization with theory-of-mind tasks across Qwen, Llama, and OLMo baselines.\n\nThe full corpus improved every reported personalization metric, including **26.6–40.9 percentage-point gains** in preference-following generation. The users and mental states are simulated, and the material gives no evidence for coding-agent workflows or real long-term deployments.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2609.15972v1",
    "published_at": "2026-09-14T17:55:58.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "model",
  "domains": [],
  "topics": [
    "agent-memory"
  ],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "The full corpus improved every reported personalization metric, including **26.6–40.9 percentage-point gains** in preference-following generation. The users and mental states are simulated, and the material gives no evidence for coding-agent workflows or real long-term deployments."
  ],
  "connected_context": {
    "meaning": "This adds a third memory design to the candidates: train a model from privileged, evolving representations of a user rather than inject stored context or consolidate repeated experience directly. It strengthens the case for treating user state as dynamic, but sharply narrows the evidence to simulated people and conversations; it does not establish that inferred beliefs should become durable deployment memory.",
    "corpus_size": 778,
    "generated_at": "2026-09-15T10:06:50.635Z",
    "connections": [
      {
        "title": "UniMem: Complementary Episodic-to-Parametric Memory for Boundary-Agnostic Task Streams",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2607.26017v1",
        "feed7_url": "https://feed7.dev/p/2607-26017v1-1opv1da",
        "reason": "Both move recurring experience toward model parameters, but Mind2Dialogue distills from a privileged simulated mental state while UniMem routes repeated experience between retrieval and expandable parametric memory."
      },
      {
        "title": "The Rise of Verbal Reinforcement Learning",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2609.01597v1",
        "feed7_url": "https://feed7.dev/p/2609-01597v1-1yaem2a",
        "reason": "Privileged distillation is a concrete parameter-update intervention, reinforcing the survey’s distinction between learning from feedback and merely retaining it as runtime context."
      },
      {
        "title": "Structured Memory for Edge Language Models: Persistent Context and Corpus Retrieval via O(1) SSM State Injection",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2608.02560v1",
        "feed7_url": "https://feed7.dev/p/2608-02560v1-1u8ls14",
        "reason": "PRECOG keeps memory external as reusable injected SSM states; Mind2Dialogue instead trains the deployed model to act without direct access to the privileged state."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-09-14T17:55:58.000Z",
  "modified_at": "2026-09-14T17:55:58.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2609-15972v1-1dw4a6c",
    "json": "https://feed7.dev/p/2609-15972v1-1dw4a6c.json",
    "markdown": "https://feed7.dev/p/2609-15972v1-1dw4a6c.md"
  }
}