The team behind Chatbot Arena found that GPT-4 as a judge agreed with human preferences over 80% of the time, the same level as humans with each other, but also that it gave consistent verdicts when answer order was swapped in only 65% of cases; it gave its own answers a 10% higher win rate, though the authors say their limited data cannot determine whether this is a self-enhancement bias.

SocioLens project,

Claim LD06-SL-C19 (v1.0)

Provenance and status

Claim type: Estimate
Publication status: Eligible for publication, with qualifiers: source position
Verification state: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.
Does not license: Judge reliability for current models or for tasks outside chat
Limitations: 2023 models; the authors run the platform whose method the paper supports
Source (as cited): Zheng, Chiang, Sheng et al., “Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena,” NeurIPS 2023 Datasets and Benchmarks Track, arXiv:2306.05685
Source type: Peer-reviewed conference paper (recorded as “Peer-reviewed conference paper”)
Institutional position: Academic team that runs Chatbot Arena (LMSYS)
Maturity of the evidence: Rapidly maturing evidence
Recorded date: 15 October 2026
Currentness / correction: This is the current version. No correction or concern is recorded on its sources.

Sources this version cites

Zheng, Chiang, Sheng et al., “Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena,” NeurIPS 2023 Datasets and Benchmarks Track, arXiv:2306.05685 (Peer-reviewed conference paper)
Read the original

Version

This is the current version, v1.0. Earlier versions stay available, and their records never change.

Versions: v1.0 · current

Raw record: every recorded field, with its coded values

All 25 canonical Claim fields, verbatim, with the coded values derived from them. The plain-language labels above are read from these.

Claim type (coded): ESTIMATE (as printed: “ESTIMATE”)
Publication status (coded): ELIGIBLE (SOURCE_POSITION) (as printed: “ELIGIBLE WITH SOURCE-POSITION QUALIFIER”)
Verification state (coded): CONTENT_VERIFIED_PRIMARY
Source type (coded): PEER_REVIEWED_CONFERENCE_PAPER
Maturity (coded): RAPIDLY_MATURING

#	Canonical field	Canonical value (verbatim)	Normalized (Claim Value Vocabulary v1.0.2)
1	CLAIM ID	LD06-SL-C19	—
2	PUBLIC-FACING CLAIM	The team behind Chatbot Arena found that GPT-4 as a judge agreed with human preferences over 80% of the time, the same level as humans with each other, but also that it gave consistent verdicts when answer order was swapped in only 65% of cases; it gave its own answers a 10% higher win rate, though the authors say their limited data cannot determine whether this is a self-enhancement bias.	—
3	CLAIM TYPE	ESTIMATE	base_types: ESTIMATE
4	RESEARCH STRAND	B — Leaderboards, judges and disclosure	—
5	MATURITY	RAPIDLY MATURING	states: RAPIDLY_MATURING
6	SOURCE	Zheng, Chiang, Sheng et al., “Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena,” NeurIPS 2023 Datasets and Benchmarks Track, arXiv:2306.05685	source_ids: SRC-SL-ZHENG-ET-AL-MT-BENCH-NEURIPS-2023
7	SOURCE TYPE	Peer-reviewed conference paper	classes: PEER_REVIEWED_CONFERENCE_PAPER
8	INSTITUTIONAL POSITION	Academic team that runs Chatbot Arena (LMSYS)	—
9	SCOPE	Open-ended chat questions (MT-Bench and Arena data)	—
10	DESIGN	Comparison of model judges with expert and crowd votes; bias experiments	—
11	SAMPLE / DATASET	MT-Bench questions and Chatbot Arena votes	—
12	MODEL / TOOL / VERSION	GPT-4, GPT-3.5, Claude-v1 and others as judges	—
13	COMPARATOR	Human preferences	—
14	MAIN RESULT	Over 80% agreement with humans; position consistency 65.0% for GPT-4, 46.2% for GPT-3.5, 23.8% for Claude-v1; own-answer win-rate gains of 10% (GPT-4) and 25% (Claude-v1), which the authors say cannot establish a self-enhancement bias given limited data and small differences; under a verbosity attack GPT-4 failed in 8.7% of cases against 91.3% for the other judges	—
15	INFERENTIAL TARGET	Whether a model judge is a valid stand-in for human preference	—
16	DOES NOT LICENSE	Judge reliability for current models or for tasks outside chat	—
17	VERIFICATION STATE	CONTENT VERIFIED	state: CONTENT_VERIFIED_PRIMARY · verified_by: UNKNOWN
18	LIMITATIONS	2023 models; the authors run the platform whose method the paper supports	—
19	CONTRADICTORY / SUPPORTING SOURCES	LD06-SL-C18	—
20	GEOGRAPHY / LANGUAGE	Not checked / English	—
21	DISTRIBUTIONAL LEDGER	Not analysed in this claim	—
22	R01 / FIELD-MAP LINKS	LD-06, D4b	—
23	WHAT WOULD CHANGE THIS CLAIM?	Repeat bias measurements on current judges	—
24	PUBLICATION STATUS	ELIGIBLE WITH SOURCE-POSITION QUALIFIER	state: ELIGIBLE · qualifiers: SOURCE_POSITION
25	RECORDED DATE	2026-10-15	—

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

The sources as recorded, with their record IDs and coded source types:

Pinned Sources (field 6, version-specific)

“Zheng, Chiang, Sheng et al., “Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena,” NeurIPS 2023 Datasets and Benchmarks Track, arXiv:2306.05685” → Zheng et al. (2023), “Judging LLM-as-a-Judge with…” (SRC-SL-ZHENG-ET-AL-MT-BENCH-NEURIPS-2023--v1.0) — Zheng, Chiang, Sheng et al., “Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena,” NeurIPS 2023 Datasets and Benchmarks Track, arXiv:2306.05685; class PEER_REVIEWED_CONFERENCE_PAPER (explicit public provenance pin)

Cited by Living Debate versions

The evaluation apparatus: Can benchmarks, leaderboards and audits carry the weight placed on them? (v0.1) LD-06

Cite this page

SocioLens project. (2026, October 15). The team behind Chatbot Arena found that GPT-4 as a judge agreed with human preferences over 80% of the time, the same level as humans with each other, but also that it gave consistent verdicts when answer order was swapped in only 65% of cases; it gave its own answers a 10% higher win rate, though the authors say their limited data cannot determine whether this is a self-enhancement bias [Claim LD06-SL-C19, version v1.0]. SocioLens. https://sociolens-web.pages.dev/claims/ld06-sl-c19/v1.0/

@misc{sociolens_ld06_sl_c19_v1_0,
  author = {{SocioLens project}},
  title = {The team behind Chatbot Arena found that GPT-4 as a judge agreed with human preferences over 80\% of the time, the same level as humans with each other, but also that it gave consistent verdicts when answer order was swapped in only 65\% of cases; it gave its own answers a 10\% higher win rate, though the authors say their limited data cannot determine whether this is a self-enhancement bias.},
  year = {2026},
  date = {2026-10-15},
  howpublished = {\url{https://sociolens-web.pages.dev/claims/ld06-sl-c19/v1.0/}},
  note = {Claim LD06-SL-C19, version v1.0}
}

The link points to this version, whose record never changes. For a finding, cite the original study as well: SocioLens summarizes research and does not replace it.

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 03db4016214d8e6f18fa9114686d4b7ffe139c39429e762cdab19242122d6690.

Repository-authored inputs (not Drive exports)

SocioLens claim — LD06-SL-C19 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 19 — LD06-SL-C19” SHA-256 e7be603bf130cb2a7a73a5e986f367c218399035a988d6a0b15e86bb3425fde9; fixture fixtures/authored/claims/LD06-SL-C19--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-C19 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-C19 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-ZHENG-ET-AL-MT-BENCH-NEURIPS-2023--v1.0 — basis: field 6 of ENTRY 19 — LD06-SL-C19 cites "Zheng, Chiang, Sheng et al., “Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena,” NeurIPS 2023 Datasets and Benchmarks Track, arXiv:2306.05685"; the only field-6 segment of this ClaimVersion's repository-authored ledger block.
SOURCE_PROVENANCE_PIN = SRC-SL-ZHENG-ET-AL-MT-BENCH-NEURIPS-2023--v1.0 — basis: explicit version-pinned public provenance of this ClaimVersion (Amendment v1.2 §6): the pinned field-6 Source record "Zheng, Chiang, Sheng et al., “Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena,” NeurIPS 2023 Datasets and Benchmarks Track, arXiv:2306.05685".