A GPT-style model trained only on sequences of legal Othello moves, with no rules or board given, made illegal moves 0.01% of the time; its activations encoded the board state, and editing that internal state changed its predicted moves as the edited board implied.

SocioLens project,

Claim LD01-SL-C16 (v1.0)

Provenance and status

Claim type: Estimate
Publication status: Eligible for publication, with a qualifier
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: A claim that deployed language models hold comparable models of the world, or that they understand
Limitations: Read in arXiv version 5 (26 June 2024); a closed synthetic world, not natural language; linear probes did not recover the board well in this setting, so the evidence rests on nonlinear probes and interventions
Source (as cited): Li, Hopkins, Bau, Viégas, Pfister & Wattenberg, “Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task,” ICLR 2023
Source type: Peer-reviewed conference paper (recorded as “Peer-reviewed conference paper”)
Institutional position: Independent academic (Harvard University, MIT, Northeastern University)
Maturity of the evidence: Emerging evidence
Recorded date: 16 October 2026
Currentness / correction: This is the current version. No correction or concern is recorded on its sources.

Sources this version cites

Li, Hopkins, Bau, Viégas, Pfister & Wattenberg, “Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task,” ICLR 2023 (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 (QUALIFIED) (as printed: “ELIGIBLE WITH QUALIFIER”)
Verification state (coded): CONTENT_VERIFIED_PRIMARY
Source type (coded): PEER_REVIEWED_CONFERENCE_PAPER
Maturity (coded): EMERGING

#	Canonical field	Canonical value (verbatim)	Normalized (Claim Value Vocabulary v1.0.2)
1	CLAIM ID	LD01-SL-C16	—
2	PUBLIC-FACING CLAIM	A GPT-style model trained only on sequences of legal Othello moves, with no rules or board given, made illegal moves 0.01% of the time; its activations encoded the board state, and editing that internal state changed its predicted moves as the edited board implied.	—
3	CLAIM TYPE	ESTIMATE	base_types: ESTIMATE
4	RESEARCH STRAND	B — Evidence that could tell the accounts apart	—
5	MATURITY	EMERGING	states: EMERGING
6	SOURCE	Li, Hopkins, Bau, Viégas, Pfister & Wattenberg, “Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task,” ICLR 2023	source_ids: SRC-SL-LI-ET-AL-OTHELLO-ARXIV-2210-13382
7	SOURCE TYPE	Peer-reviewed conference paper	classes: PEER_REVIEWED_CONFERENCE_PAPER
8	INSTITUTIONAL POSITION	Independent academic (Harvard University, MIT, Northeastern University)	—
9	SCOPE	One synthetic task (the board game Othello); a model trained from scratch	—
10	DESIGN	Probing and intervention on a sequence model’s internal activations	—
11	SAMPLE / DATASET	About 20 million synthetic games and a championship-game dataset	—
12	MODEL / TOOL / VERSION	Othello-GPT, a small GPT-style model trained by the authors	—
13	COMPARATOR	An untrained model (93.29% illegal moves) and probes on it	—
14	MAIN RESULT	Illegal-move rate 0.01% when trained on synthetic games; nonlinear probes recover the board (about 1.7% error at the best layer); interventions on activations change predictions consistently with the edited board	—
15	INFERENTIAL TARGET	Whether next-token training can produce an internal model of the process generating the sequences	—
16	DOES NOT LICENSE	A claim that deployed language models hold comparable models of the world, or that they understand	—
17	VERIFICATION STATE	CONTENT VERIFIED	state: CONTENT_VERIFIED_PRIMARY · verified_by: UNKNOWN
18	LIMITATIONS	Read in arXiv version 5 (26 June 2024); a closed synthetic world, not natural language; linear probes did not recover the board well in this setting, so the evidence rests on nonlinear probes and interventions	—
19	CONTRADICTORY / SUPPORTING SOURCES	A linear follow-up and real-world representations in language models: LD01-SL-C17	—
20	GEOGRAPHY / LANGUAGE	Not applicable / English	—
21	DISTRIBUTIONAL LEDGER	Not analysed in this claim	—
22	R01 / FIELD-MAP LINKS	LD-01, T4, D3	—
23	WHAT WOULD CHANGE THIS CLAIM?	Evidence that the interventions work through features other than a board representation	—
24	PUBLICATION STATUS	ELIGIBLE WITH QUALIFIER	state: ELIGIBLE · qualifiers: QUALIFIED
25	RECORDED DATE	2026-10-16	—

Field 22 references resolved against Field Map v0.2: LD-01 (LIVING_DEBATE); T4 (TENSION); D3 (DOMAIN)

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

Pinned Sources (field 6, version-specific)

“Li, Hopkins, Bau, Viégas, Pfister & Wattenberg, “Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task,” ICLR 2023” → Li et al. (2023), “Emergent World Representations…” (SRC-SL-LI-ET-AL-OTHELLO-ARXIV-2210-13382--v1.0) — Li, Hopkins, Bau, Viégas, Pfister & Wattenberg, “Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task,” ICLR 2023; class PEER_REVIEWED_CONFERENCE_PAPER (explicit public provenance pin)

Cited by Living Debate versions

What kind of thing is generative AI, and does the answer predict anything? (v0.1) LD-01

Cite this page

SocioLens project. (2026, October 16). A GPT-style model trained only on sequences of legal Othello moves, with no rules or board given, made illegal moves 0.01% of the time; its activations encoded the board state, and editing that internal state changed its predicted moves as the edited board implied [Claim LD01-SL-C16, version v1.0]. SocioLens. https://sociolens-web.pages.dev/claims/ld01-sl-c16/v1.0/

@misc{sociolens_ld01_sl_c16_v1_0,
  author = {{SocioLens project}},
  title = {A GPT-style model trained only on sequences of legal Othello moves, with no rules or board given, made illegal moves 0.01\% of the time; its activations encoded the board state, and editing that internal state changed its predicted moves as the edited board implied.},
  year = {2026},
  date = {2026-10-16},
  howpublished = {\url{https://sociolens-web.pages.dev/claims/ld01-sl-c16/v1.0/}},
  note = {Claim LD01-SL-C16, 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 4ddffb9d1444e082cb68ea1bd237978c5e4ecf1159ba4b9f64534268e0ed2ebc.

Repository-authored inputs (not Drive exports)

SocioLens claim — LD01-SL-C16 v1.0 (v1.0) — written in the SocioLens repository by Project Lead (Claude, AI) on 2026-10-16; 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 16 October 2026 to build the LD-01 debate; a new Claim, first version, written from the project's LD-01 source notes, checked against full texts, official web pages or official abstracts. Block “ENTRY 16 — LD01-SL-C16” SHA-256 71c0330c85e0a4d9936360e3019421d8aa35a58f9e7cb6f1a443a27c7bcca2f9; fixture fixtures/authored/claims/LD01-SL-C16--v1.0.

Builder decisions (not Drive data)

VERSION = v1.0 — basis: Claim version taken from the repository-authored claim fixture version (SocioLens claim — LD01-SL-C16 v1.0); the first version of a new Claim of the LD-01 debate
CURRENT_POINTER = true — basis: repository-authored claim "SocioLens claim — LD01-SL-C16 v1.0" (Project Lead (Claude, AI), 2026-10-16); 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 16 October 2026 to build the LD-01 debate; a new Claim, first version, written from the project's LD-01 source notes, checked against full texts, official web pages or official abstracts
FIELD_6_SOURCE_REFERENCE = SRC-SL-LI-ET-AL-OTHELLO-ARXIV-2210-13382--v1.0 — basis: field 6 of ENTRY 16 — LD01-SL-C16 cites "Li, Hopkins, Bau, Viégas, Pfister & Wattenberg, “Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task,” ICLR 2023"; the only field-6 segment of this ClaimVersion's repository-authored ledger block.
SOURCE_PROVENANCE_PIN = SRC-SL-LI-ET-AL-OTHELLO-ARXIV-2210-13382--v1.0 — basis: explicit version-pinned public provenance of this ClaimVersion (Amendment v1.2 §6): the pinned field-6 Source record "Li, Hopkins, Bau, Viégas, Pfister & Wattenberg, “Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task,” ICLR 2023".