{
  "id": 706,
  "url": "https://arxiv.org/abs/2606.23112v1",
  "title": "Self-Evolution for Multi-Turn Tool-Calling Agents via Divergence-Point Preference Learning",
  "summary": "Multi-turn tool-using agents must coordinate long-horizon tool sequences while tracking dialogue state and policy constraints. Existing approaches often separate inference-time orchestration from parameter-level learning, leaving tool selection weakly structured and preference updates vulnerable to train--deployment prompt mismatch. For within-benchmark self-improvement, ToolGraph combines schema-derived topology, transition weights estimated from successful rollouts, and history-aware controls ",
  "authors": "Jiaqiang Tang",
  "category": "research",
  "topics": "regulation,privacy-surveillance,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-06-22T09:56:44.000Z",
  "fetched_at": "2026-07-14T14:14:46.032Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/706",
  "original_url": "https://arxiv.org/abs/2606.23112v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}