Linear probes on Llama-2-70B’s activations predicted the locations of world places with out-of-sample R² of 0.911 and the death years of historical figures with 0.835; the authors say these representations “do not constitute a dynamic causal world model in their own right.”

SocioLens project,

Claim LD01-SL-C17 (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: Evidence that models use these representations as a causal model of the world, which the authors “conjecture, but do not show”
Limitations: Read in arXiv version 3 (4 March 2024); a probe can decode information a model does not use; held-out-block tests raise errors, which the authors read as weak evidence that probes partly learn the mapping themselves
Source (as cited): Gurnee & Tegmark, “Language Models Represent Space and Time,” ICLR 2024
Source type: Peer-reviewed conference paper (recorded as “Peer-reviewed conference paper”)
Institutional position: Independent academic (MIT)
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

Gurnee & Tegmark, “Language Models Represent Space and Time,” ICLR 2024 (Peer-reviewed conference paper)
Read the original

Version

This page shows version v1.0. Its record never changes, even when newer evidence arrives; the site may add a note to the page or withhold a figure not cleared for public use.

This pinned version is also the current version.

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-C17	—
2	PUBLIC-FACING CLAIM	Linear probes on Llama-2-70B’s activations predicted the locations of world places with out-of-sample R² of 0.911 and the death years of historical figures with 0.835; the authors say these representations “do not constitute a dynamic causal world model in their own right.”	—
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	Gurnee & Tegmark, “Language Models Represent Space and Time,” ICLR 2024	source_ids: SRC-SL-GURNEE-TEGMARK-ARXIV-2310-02207
7	SOURCE TYPE	Peer-reviewed conference paper	classes: PEER_REVIEWED_CONFERENCE_PAPER
8	INSTITUTIONAL POSITION	Independent academic (MIT)	—
9	SCOPE	English-language entity datasets (places, people, works, headlines); Llama-2 and Pythia models	—
10	DESIGN	Linear and nonlinear probing of activations; neuron ablations	—
11	SAMPLE / DATASET	Six datasets of places and dated entities	—
12	MODEL / TOOL / VERSION	Llama-2 (7B, 13B, 70B) and Pythia models	—
13	COMPARATOR	Nonlinear probes; held-out blocks of data	—
14	MAIN RESULT	Out-of-sample R² for Llama-2-70B with a linear probe: world 0.911, US 0.864, New York City 0.359, historical figures 0.835, entertainment 0.885, headlines 0.746; nonlinear probes add little	—
15	INFERENTIAL TARGET	Whether language models hold structured representations that reflect the world, rather than only superficial statistics	—
16	DOES NOT LICENSE	Evidence that models use these representations as a causal model of the world, which the authors “conjecture, but do not show”	—
17	VERIFICATION STATE	CONTENT VERIFIED	state: CONTENT_VERIFIED_PRIMARY · verified_by: UNKNOWN
18	LIMITATIONS	Read in arXiv version 3 (4 March 2024); a probe can decode information a model does not use; held-out-block tests raise errors, which the authors read as weak evidence that probes partly learn the mapping themselves	—
19	CONTRADICTORY / SUPPORTING SOURCES	LD01-SL-C16	—
20	GEOGRAPHY / LANGUAGE	Mostly English	—
21	DISTRIBUTIONAL LEDGER	Not analysed in this claim	—
22	R01 / FIELD-MAP LINKS	LD-01, T4, D3	—
23	WHAT WOULD CHANGE THIS CLAIM?	Causal tests showing the models do or do not use these representations	—
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)

“Gurnee & Tegmark, “Language Models Represent Space and Time,” ICLR 2024” → Gurnee & Tegmark (2024), “Language Models Represent Space…” (SRC-SL-GURNEE-TEGMARK-ARXIV-2310-02207--v1.0) — Gurnee & Tegmark, “Language Models Represent Space and Time,” ICLR 2024; 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). Linear probes on Llama-2-70B’s activations predicted the locations of world places with out-of-sample R² of 0.911 and the death years of historical figures with 0.835; the authors say these representations “do not constitute a dynamic causal world model in their own right.” [Claim LD01-SL-C17, version v1.0]. SocioLens. https://sociolens-web.pages.dev/claims/ld01-sl-c17/v1.0/

@misc{sociolens_ld01_sl_c17_v1_0,
  author = {{SocioLens project}},
  title = {Linear probes on Llama-2-70B’s activations predicted the locations of world places with out-of-sample R² of 0.911 and the death years of historical figures with 0.835; the authors say these representations “do not constitute a dynamic causal world model in their own right.”},
  year = {2026},
  date = {2026-10-16},
  howpublished = {\url{https://sociolens-web.pages.dev/claims/ld01-sl-c17/v1.0/}},
  note = {Claim LD01-SL-C17, 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 63868f18b56e0967ce535b3d07fcaabf95fb8a5716d493ce87a7614710f6e2f3.

Repository-authored inputs (not Drive exports)

SocioLens claim — LD01-SL-C17 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 17 — LD01-SL-C17” SHA-256 593b32f886c6ddccf1df1c737d850b2df217bc6eb75246c88c8ca943ffca7d59; fixture fixtures/authored/claims/LD01-SL-C17--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-C17 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-C17 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-GURNEE-TEGMARK-ARXIV-2310-02207--v1.0 — basis: field 6 of ENTRY 17 — LD01-SL-C17 cites "Gurnee & Tegmark, “Language Models Represent Space and Time,” ICLR 2024"; the only field-6 segment of this ClaimVersion's repository-authored ledger block.
SOURCE_PROVENANCE_PIN = SRC-SL-GURNEE-TEGMARK-ARXIV-2310-02207--v1.0 — basis: explicit version-pinned public provenance of this ClaimVersion (Amendment v1.2 §6): the pinned field-6 Source record "Gurnee & Tegmark, “Language Models Represent Space and Time,” ICLR 2024".