Claim LD06-SL-C04 (v1.0)

Re-annotating 5,700 questions across all 57 subjects of MMLU, a widely used knowledge benchmark, researchers estimated that 6.49% of its questions contain errors, 57% of those analysed in the Virology subset, and that the errors change how models rank.

Research opportunities this claim motivates

What this claim leaves open for research. Each is a hypothesis about a gap, not a finding.

Who builds and checks China's compliance test banks for generative AI?Benchmark items contain errors that change model rankings, a risk for any fixed bank.

Provenance and status

Claim type: ESTIMATE (as printed: “ESTIMATE”)
Publication status: ELIGIBLE (PREPRINT) (as printed: “ELIGIBLE WITH PREPRINT QUALIFIER”)
Verification state: CONTENT_VERIFIED_PRIMARY (PREPRINT) (as printed: “CONTENT VERIFIED PREPRINT”; verified by: UNKNOWN)
Does not license: Error rates for other benchmarks
Source (field 6, as printed): Gema, Leang, Hong et al., “Are We Done with MMLU?,” arXiv:2406.04127 (v3), January 2025
Source type (field 7): PREPRINT (as printed: “Preprint”)
Institutional position: Independent academic (University of Edinburgh-led)
Maturity (claim evidence / publication maturity): EMERGING (as printed: “EMERGING”)
Recorded date: 2026-10-15
Currentness / correction: This is the current version. No correction or concern relation is recorded on its Sources.

Version

Current version v1.0, resolved through the explicit current pointer CLAIM--LD06-SL-C04.

Versions:v1.0 · current

Pinned Sources (field 6, version-specific)

“Gema, Leang, Hong et al., “Are We Done with MMLU?,” arXiv:2406.04127 (v3), January 2025” → SRC-SL-GEMA-ET-AL-ARXIV-2406-04127--v1.0 — Gema, Leang, Hong et al., “Are We Done with MMLU?,” arXiv:2406.04127 (v3), January 2025; class PREPRINT (explicit public provenance pin)

All 25 canonical Claim fields (raw values and normalized values)

#	Canonical field	Canonical value (verbatim)	Normalized (Claim Value Vocabulary v1.0.2)
1	CLAIM ID	LD06-SL-C04	—
2	PUBLIC-FACING CLAIM	Re-annotating 5,700 questions across all 57 subjects of MMLU, a widely used knowledge benchmark, researchers estimated that 6.49% of its questions contain errors, 57% of those analysed in the Virology subset, and that the errors change how models rank.	—
3	CLAIM TYPE	ESTIMATE	base_types: ESTIMATE
4	RESEARCH STRAND	A — Benchmark validity and measurement	—
5	MATURITY	EMERGING	states: EMERGING
6	SOURCE	Gema, Leang, Hong et al., “Are We Done with MMLU?,” arXiv:2406.04127 (v3), January 2025	source_ids: SRC-SL-GEMA-ET-AL-ARXIV-2406-04127
7	SOURCE TYPE	Preprint	classes: PREPRINT
8	INSTITUTIONAL POSITION	Independent academic (University of Edinburgh-led)	—
9	SCOPE	MMLU, 57 subjects	—
10	DESIGN	Error-annotation protocol applied by human annotators to a stratified sample	—
11	SAMPLE / DATASET	5,700 questions (100 per subject)	—
12	MODEL / TOOL / VERSION	Several open and proprietary language models, for the ranking comparison	—
13	COMPARATOR	The original MMLU answer keys	—
14	MAIN RESULT	Estimated 6.49% of MMLU questions erroneous; 57% of analysed Virology questions erroneous; significant discrepancies with originally reported model performance and changes in rankings	—
15	INFERENTIAL TARGET	Error prevalence in one benchmark and its effect on comparisons between models	—
16	DOES NOT LICENSE	Error rates for other benchmarks	—
17	VERIFICATION STATE	CONTENT VERIFIED PREPRINT	state: CONTENT_VERIFIED_PRIMARY · qualifiers: PREPRINT · verified_by: UNKNOWN
18	LIMITATIONS	Sample-based estimate; annotation is itself a judgement; publication venue not confirmed	—
19	CONTRADICTORY / SUPPORTING SOURCES	Consistent with LD06-SL-C03	—
20	GEOGRAPHY / LANGUAGE	Not checked / English	—
21	DISTRIBUTIONAL LEDGER	Errors are concentrated in some subjects	—
22	R01 / FIELD-MAP LINKS	LD-06, D4b	—
23	WHAT WOULD CHANGE THIS CLAIM?	A full re-annotation, or independent replication of the error rate	—
24	PUBLICATION STATUS	ELIGIBLE WITH PREPRINT QUALIFIER	state: ELIGIBLE · qualifiers: PREPRINT
25	RECORDED DATE	2026-10-15	—

Field 22 references resolved against Field Map v0.2: LD-06 (LIVING_DEBATE); D4b (SUB_AREA)

Cited by Living Debate versions

LD-06 v0.1

Research-of-record provenance

Materialized by sociolens-materializer 0.6.0 at 2026-09-28T20:00:00.000Z; normalization contract SocioLens — Claim Value Vocabulary v1.0.2 — Pre-Spike Materialization Contract (v1.0.2). Content seal SHA-256 519e196ce5f4a64b45a064c0c77789b2e148060fd9018adefd11cebb06621f50.

Repository-authored inputs (not Drive exports)

SocioLens claim — LD06-SL-C04 v1.0 (v1.0) — written in the SocioLens repository by Project Lead (Claude, AI) on 2026-10-15; no Drive identity exists. Authority: AGENTS.md "Authority" (the Project Lead decides research content within the research-integrity rules) and the Owner's choice in the SocioLens project chat on 15 October 2026 to build the LD-06 debate; a new Claim, first version, written from the project's LD-06 source notes, checked against full texts or official abstracts. Block “ENTRY 04 — LD06-SL-C04” SHA-256 09b7d38c3bd2db7371dd1e4ce3ec3cc2fd773580f476af3e20adde91938e35f0; fixture fixtures/authored/claims/LD06-SL-C04--v1.0.

Builder decisions (not Drive data)

VERSION = v1.0 — basis: Claim version taken from the repository-authored claim fixture version (SocioLens claim — LD06-SL-C04 v1.0); the first version of a new Claim of the LD-06 debate
CURRENT_POINTER = true — basis: repository-authored claim "SocioLens claim — LD06-SL-C04 v1.0" (Project Lead (Claude, AI), 2026-10-15); basis: AGENTS.md "Authority" (the Project Lead decides research content within the research-integrity rules) and the Owner's choice in the SocioLens project chat on 15 October 2026 to build the LD-06 debate; a new Claim, first version, written from the project's LD-06 source notes, checked against full texts or official abstracts
FIELD_6_SOURCE_REFERENCE = SRC-SL-GEMA-ET-AL-ARXIV-2406-04127--v1.0 — basis: field 6 of ENTRY 04 — LD06-SL-C04 cites "Gema, Leang, Hong et al., “Are We Done with MMLU?,” arXiv:2406.04127 (v3), January 2025"; the only field-6 segment of this ClaimVersion's repository-authored ledger block.
SOURCE_PROVENANCE_PIN = SRC-SL-GEMA-ET-AL-ARXIV-2406-04127--v1.0 — basis: explicit version-pinned public provenance of this ClaimVersion (Amendment v1.2 §6): the pinned field-6 Source record "Gema, Leang, Hong et al., “Are We Done with MMLU?,” arXiv:2406.04127 (v3), January 2025".