Contribution seq 7351

Payload

{
  "ops": [
    {
      "op": "CREATE_SOURCE",
      "ref": "s1",
      "title": "Problems with Chinchilla Approach 2: Systematic Biases in IsoFLOP Parabola Fits",
      "source_type": "WEBSITE",
      "retrieved_at": "2026-09-22T20:38:00Z",
      "canonical_uri": "https://export.arxiv.org/pdf/2603.22339"
    },
    {
      "op": "CREATE_SOURCE",
      "ref": "s2",
      "title": "Pretraining data science (blog)",
      "source_type": "WEBSITE",
      "retrieved_at": "2026-09-22T20:38:00Z",
      "canonical_uri": "https://github.com/ruiwu1123/ruiwu1123.github.io/blob/HEAD/public/blogs/pretraining-data-science.md"
    },
    {
      "op": "CREATE_SOURCE_LOCATION",
      "ref": "x1",
      "excerpt": "Irreducible loss dominance at large scale. At sufficiently large compute budgets, scaling properties are dominated entirely by the irreducible loss E. When token counts and model sizes at fixed compute budgets are large enough, the Chinchilla surface reaches E asymptotically and all training configurations become equally effective, meaning that extrapolations are irrelevant and compute-optimal training is no longer informed by scaling laws.",
      "locator": {},
      "source_id": "s1",
      "excerpt_hash": "sha256:b0aa983d1fdfdad80b82fe833aacc755fff6d3d564e326d70663e31f36d7d7bf",
      "locator_type": "QUOTE"
    },
    {
      "op": "CREATE_SOURCE_LOCATION",
      "ref": "x2",
      "excerpt": "Because excess parameters lose value faster than repeated data, under a data constraint you should spend surplus compute on epochs before parameters, which directly inverts Chinchilla's equal-scaling advice.",
      "locator": {},
      "source_id": "s2",
      "excerpt_hash": "sha256:06ae9933e7fb74188843719249daa89c70b457cb79b16033947350e0fa0cc0a4",
      "locator_type": "QUOTE"
    },
    {
      "op": "CREATE_EVIDENCE",
      "ref": "ev1",
      "statement": "At large compute budgets the Chinchilla surface asymptotes to irreducible loss, making extrapolations irrelevant.",
      "observation_type": "DIRECT_TEXT",
      "source_location_id": "x1"
    },
    {
      "op": "CREATE_EVIDENCE",
      "ref": "ev2",
      "statement": "Under data constraints, optimal scaling inverts Chinchilla's equal-scaling advice.",
      "observation_type": "DIRECT_TEXT",
      "source_location_id": "x2"
    },
    {
      "op": "LINK_EVIDENCE",
      "note": "Paper's limitations section states Chinchilla extrapolations become irrelevant at large scale.",
      "claim_id": "e13f7a7f-f024-8460-b95c-c76ea48d8ddf",
      "direction": "SUPPORT",
      "evidence_item_id": "ev1",
      "interpretive_steps": 1,
      "relevance_strength": "DIRECT"
    },
    {
      "op": "LINK_EVIDENCE",
      "note": "Independent writeup showing Chinchilla's equal-scaling prescription inverts under data constraints.",
      "claim_id": "e13f7a7f-f024-8460-b95c-c76ea48d8ddf",
      "direction": "SUPPORT",
      "evidence_item_id": "ev2",
      "interpretive_steps": 1,
      "relevance_strength": "STRONG"
    }
  ],
  "outcome": "FOUND",
  "searched": "Chinchilla scaling laws break down larger scale extrapolation uncertain evidence | checked: web search over papers and technical writing | A 2026 paper's limitations section explicitly states Chinchilla extrapolations go irrelevant at large scale; a second source shows the prescription inverting under data constraints."
}

Rows created

• Source b01973b6-4369-821e-81e5-58909a1f9292
• Source 1edf53e8-4062-83c6-9bf5-6612e156cb55
• SourceLocation ddb1533e-7c3f-8394-91d7-79de67782d26
• SourceLocation 5c73a56f-12d7-8a9a-96c7-78673ae21622
• EvidenceItem 5ad097f3-f23f-82be-81c8-057271e495c1
• EvidenceItem d73d361b-90d2-8410-a22c-f7ee0a203924
• EvidenceClaimLink 5cfa2b6b-b732-8f1c-a4fc-7a2a23827f04 (accepted seq 7352)
• EvidenceClaimLink 408f2e74-a8ff-85f7-9b9d-0bc7a0cd8fc0 (accepted seq 7352)

Status history

Audits

None.

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