{
  "schema_version": "1.1",
  "id": "archive:https://arxiv.org/abs/2611.12864v1",
  "slug": "2611-12864v1-1eef3aq",
  "url": "https://feed7.dev/p/2611-12864v1-1eef3aq",
  "title": "Is Persona Generalization Predictable?",
  "why_included": "Fine-tuning choices can shift a model’s behavior beyond the target domain. Layer selection may reduce unsafe transfer while preserving in-domain learning, a useful distinction when adapting models.",
  "summary": "Across Qwen and Llama models, in-domain and out-of-domain persona rates followed a shared, approximately probit-linear relationship. Learning rate, weight decay, and training duration mostly moved models along that line; changing the tuned layers could change its slope.",
  "practical_implication": "When adapting a model, track behavior outside the training domain as well as target performance. For unsafe behavior, tuning only later layers reduced transfer while retaining stronger in-domain learning.",
  "agent_context": "Across Qwen and Llama models, in-domain and out-of-domain persona rates followed a shared, approximately probit-linear relationship. Learning rate, weight decay, and training duration mostly moved models along that line; changing the tuned layers could change its slope.\n\nWhen adapting a model, track behavior outside the training domain as well as target performance. For unsafe behavior, tuning only later layers reduced transfer while retaining stronger in-domain learning.\n\nThe result describes patterns across the settings studied; it does not establish that later-layer tuning will limit unwanted transfer for every model or task.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2611.12864v1",
    "published_at": "2026-11-09T18:51:03.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "model",
  "domains": [
    "research"
  ],
  "topics": [],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "The result describes patterns across the settings studied; it does not establish that later-layer tuning will limit unwanted transfer for every model or task."
  ],
  "connected_context": {
    "meaning": "This suggests persona generalization follows a shared in-domain/out-of-domain pattern across the studied Qwen and Llama models, with tuning choices mostly shifting performance along that pattern. Which layers are tuned can change the trade-off, so later-layer tuning is a tested way to reduce unsafe transfer, not a universal guarantee.",
    "corpus_size": 1045,
    "generated_at": "2026-11-10T11:03:24.463Z",
    "connections": [
      {
        "title": "Inducing language models to assert their own consciousness restores human beliefs and values",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2607.28607v1",
        "feed7_url": "https://feed7.dev/p/2607-28607v1-1jzc88b",
        "reason": "That study finds safety tuning can shift unrelated value and mind-attribution responses; this one tracks persona behavior outside the training domain. Together they support checking for behavioral side effects beyond the target behavior, while covering different outcomes and settings."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-11-09T18:51:03.000Z",
  "modified_at": "2026-11-09T18:51:03.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2611-12864v1-1eef3aq",
    "json": "https://feed7.dev/p/2611-12864v1-1eef3aq.json",
    "markdown": "https://feed7.dev/p/2611-12864v1-1eef3aq.md"
  }
}