{
  "schema_version": "1.1",
  "id": "archive:https://arxiv.org/abs/2611.26590v1",
  "slug": "2611-26590v1-1w2fjt7",
  "url": "https://feed7.dev/p/2611-26590v1-1w2fjt7",
  "title": "Probing the \"Conscious\" Mind: Selective Access and Global Workspace in Vision-Language Models",
  "why_included": "A study of vision-language models finds that concepts can be present in internal representations before the model can reliably select them for a task. Its analysis tool compares concept accessibility across image and text pathways.",
  "summary": "The paper introduces **MC-Lens**, which maps vision-encoder and language-decoder states into a shared semantic space and supports interventions on accessed concepts. Experiments find that concept information is readable internally, while reliable selective access develops later; accessed concepts also affect downstream reasoning.",
  "practical_implication": "For builders using vision-language models, output quality may depend on whether the model can select relevant visual concepts for a particular task, not only whether those concepts are represented. MC-Lens offers a research method for probing that distinction.",
  "agent_context": "The paper introduces **MC-Lens**, which maps vision-encoder and language-decoder states into a shared semantic space and supports interventions on accessed concepts. Experiments find that concept information is readable internally, while reliable selective access develops later; accessed concepts also affect downstream reasoning.\n\nFor builders using vision-language models, output quality may depend on whether the model can select relevant visual concepts for a particular task, not only whether those concepts are represented. MC-Lens offers a research method for probing that distinction.\n\nThe findings describe controlled experiments across model families and sizes. The material does not establish that these measurements predict performance in deployed applications.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2611.26590v1",
    "published_at": "2026-11-19T18:46:50.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "model",
  "domains": [
    "image",
    "research"
  ],
  "topics": [
    "reasoning"
  ],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "The findings describe controlled experiments across model families and sizes. The material does not establish that these measurements predict performance in deployed applications."
  ],
  "connected_context": {
    "meaning": "MC-Lens separates whether a vision-language model contains readable information about a concept from whether it can selectively access that concept for a task. The interventions also connect accessed concepts to downstream reasoning. This offers a way to probe a possible source of task-specific failure, but the controlled findings do not establish deployed performance.",
    "corpus_size": 1135,
    "generated_at": "2026-11-21T11:06:57.237Z",
    "connections": [
      {
        "title": "MIRROR: Learning from the Other View for Multi-Modal Reasoning",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2607.21552v1",
        "feed7_url": "https://feed7.dev/p/2607-21552v1-1v5rc1p",
        "reason": "MC-Lens probes selective access to visual concepts; MIRROR addresses modality inconsistency by training across text, diagram, and combined views, offering a complementary way to investigate multimodal reasoning failures."
      },
      {
        "title": "Towards interpreting any residual stream direction by scaling activation inversion",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2611.08465v1",
        "feed7_url": "https://feed7.dev/p/2611-08465v1-1gcivnw",
        "reason": "Both make internal model states more inspectable: MC-Lens probes concept access and interventions, while MAEMs generate and verify text that activates a chosen residual direction. Neither alone establishes deployed behavior."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-11-19T18:46:50.000Z",
  "modified_at": "2026-11-19T18:46:50.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2611-26590v1-1w2fjt7",
    "json": "https://feed7.dev/p/2611-26590v1-1w2fjt7.json",
    "markdown": "https://feed7.dev/p/2611-26590v1-1w2fjt7.md"
  }
}