How can visual reasoning be preserved within discrete autoregressive sequences without external tools or pixel-level supervision?

{
  "doc": "guo-atlas-2026-fulltext.md",
  "author": null,
  "context": "Core research question addressed by ATLAS: bridging interpretability of agentic methods, efficiency of discrete tokens, and scalability of autoregressive training.",
  "category": "ai",
  "enrichment": {
    "is_stale": true
  },
  "norm_label": "How can visual reasoning be preserved within discrete autoregressive sequences without external tools or pixel-level supervision?",
  "source_url": null,
  "graphify_id": "question_visual_reasoning_boundaries",
  "source_file": "guo-atlas-2026-fulltext.md",
  "imported_from": "/Users/antonborzov/Documents/Research.nosync/papers/extract_typed_out/guo-atlas-2026-fulltext/graph.json",
  "extracted_type": "question",
  "source_location": "Introduction, §2",
  "graphify_file_type": "question"
}