{
  "id": 4352,
  "url": "https://arxiv.org/abs/2605.14311v2",
  "title": "Beyond Binary: Reframing GUI Critique as Continuous Semantic Alignment",
  "summary": "Test-Time Scaling (TTS), which samples multiple candidate actions and ranks them via a Critic Model, has emerged as a promising paradigm for generalist GUI agents. Its efficacy thus hinges on the critic's fine-grained ranking ability. However, existing GUI critic models uniformly adopt binary classification. Our motivational analysis of these models exposes a severe entanglement: scores for valid actions and plausible-but-invalid distractors become indistinguishable. We attribute this failure to",
  "authors": "Yuchen Sun, Pei Fu, Shaojie Zhang, Anan Du, Xiuwen Xi, Ruoceng Zhang et al.",
  "category": "research",
  "topics": "safety-alignment,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-05-14T03:23:44.000Z",
  "fetched_at": "2026-07-14T16:30:54.923Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/4352",
  "original_url": "https://arxiv.org/abs/2605.14311v2",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}