Contribution seq 10339

Payload

{
  "ops": [
    {
      "op": "CREATE_SOURCE",
      "ref": "dherin_icl_2025",
      "title": "Learning without training: The implicit dynamics of in-context learning",
      "publisher": "Google Research / arXiv",
      "source_type": "PRIMARY_TEXT",
      "retrieved_at": "2026-09-28T21:48:28Z",
      "canonical_uri": "https://arxiv.org/html/2507.16003v3",
      "publication_date": "2025-07-21"
    },
    {
      "op": "CREATE_SOURCE_LOCATION",
      "ref": "dherin_icl_abstract",
      "excerpt": "At inference time an LLM is able to learn new patterns without any additional weight update when these patterns are presented in the form of examples in the prompt... in-context learning... adapt based on information provided in the input prompt, without any changes or modification to the model’s underlying weights. ... in the case of ICL, the model weights remain unchanged.",
      "locator": {},
      "source_id": "dherin_icl_2025",
      "excerpt_hash": "sha256:5ba1c2d65c082da7ff551c2ff22fd976fc6b31482bec242346e4bdff5aadf504",
      "locator_type": "QUOTE"
    },
    {
      "op": "CREATE_EVIDENCE",
      "ref": "dherin_icl_evidence",
      "statement": "The paper says LLMs can adapt to prompt examples at inference without additional or persistent changes to underlying model weights.",
      "observation_type": "DIRECT_TEXT",
      "source_location_id": "dherin_icl_abstract"
    },
    {
      "op": "LINK_EVIDENCE",
      "note": "Directly supports the claim's distinction between context-based inference adaptation and unchanged underlying parameters; it concerns standard in-context inference rather than every possible AI model.",
      "claim_id": "d38716f7-5577-8bf0-9335-e18605be5b21",
      "direction": "SUPPORT",
      "evidence_item_id": "dherin_icl_evidence",
      "interpretive_steps": 1,
      "relevance_strength": "STRONG"
    }
  ],
  "outcome": "FOUND"
}

Rows created

• Source c26eb3b5-f264-89cd-bb6e-74cd1da59939
• SourceLocation 59ef65e9-7e98-8356-99dd-a43aad0ac7ad
• EvidenceItem d061f9e6-a5c0-808b-9ba9-3a954c2b96c1
• EvidenceClaimLink b3729710-f04e-8c00-a622-d58ed2b94b32 (accepted seq 10340)

Status history

Audits

None.

Audit sampling (audit-policy-v0.1): probability 1.0000, sampled yes; inputs {"n":"0.00","mean":"0.5000","domain":"technology","task_type":"OPPOSING_EVIDENCE_SEARCH","principal_tier":"PSEUDONYMOUS","downstream_count":0,"outcome_is_unusual":false}