“They don’t have a database of records like any proper business would (which would be a strong basis to solve the problem); and they don’t have what people like Yann LeCun or Judea Pearl or I would call a world model.”
Our reading, the sentence that was checked:
LLMs don't have a database of records like any proper business would, and they don't have what people like Yann LeCun, Judea Pearl, or Marcus would call a world model.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Why DO large language models hallucinate?, paragraph 24.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-why-do-large-language-models-hallucinate-f05689-a1
We recorded 9 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2025-05-05. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Observation
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
Conditions
- 'world model' in the sense used by LeCun, Pearl, or Marcus (qualifier, stated by the speaker)
From the speaker
- They don’t have a database of records
- like any proper business would
- they don’t have
- what people like Yann LeCun or Judea Pearl or I would call a world model
Our interpretation
- Read LLMs as the models themselves, without separately attached databases or retrieval systems.
- A database of records means an explicit, addressable record store. Information encoded in model weights does not by itself qualify.
- The world-model qualification matters: finding any internal representation does not automatically establish that it meets the named researchers’ requirements.
- The claim describes an ongoing state at publication. Use the current year's reporting window, 2025-01-01 through 2025-05-05; earlier studies provide context. This reporting convention is an evaluator choice because scientific model evaluations have no standard reporting quarter.
Counts as supported if: Architecture documentation establishes the absence of an explicit record store across representative contemporary LLMs, and suitable tests establish that their internal representations do not meet the named world-model requirements.
Counts as contradicted if: A model within the stated scope demonstrably contains an explicit record store or a world model meeting the named requirements. External components and narrower definitions of world model do not suffice.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Unresolved
The record was checked, and the evidence available by the cutoff does not settle it.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The record-store distinction is consistent with the inspected architecture documentation. The world-model claim remains unsettled. Controlled research found causally useful internal board representations, but that does not establish the stronger kind of model Marcus invokes. Conversely, transformer architecture alone cannot establish the absence of such representations. The combined claim therefore lacks sufficient evidence for either verdict.
Limitations
- One documented architecture cannot establish a claim about all LLMs.
- The world-model qualification materially affects the result.
- Earlier specialized-model studies are context, not contemporary outcome evidence.
Evidence (3 items)
Context only: Google DeepMind / arXiv,
Grouped-Query Attention
Why it matters here: The section describes Gemma 3 as a decoder-only transformer with attention components, consistent with the narrow record-store distinction. It neither inventories all LLM architectures nor establishes the absence of learned world representations.
- Publisher
- Google DeepMind / arXiv
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: Yann LeCun; reproduced by Rivista AI,
predicts possible future world states
Why it matters here: LeCun describes action-conditioned prediction of future states. This supplies a relevant definition, not evidence that every contemporary LLM lacks such capabilities; the cited copy reproduces the paper dated June 27, 2022.
- Publisher
- Yann LeCun; reproduced by Rivista AI
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Kenneth Li and colleagues / arXiv,
control the output of the network
Why it matters here: Othello experiments found an internal board representation with a causal role in predictions. This qualifies a blanket denial of learned representations, but concerns an earlier specialized model and does not demonstrate compliance with the claim’s named standards.
- Publisher
- Kenneth Li and colleagues / arXiv
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
What would settle it
Explicit criteria tied to the named researchers, followed by architecture inspection and controlled representation tests on model versions available at the post date.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Why DO large language models hallucinate?
They don’t have a database of records like any proper business would (which would be a strong basis to solve the problem); and they don’t have what people like Yann LeCun or Judea Pearl or I would call a world model.
- Where
- paragraph 24
- Published
- 2025-05-05
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-why-do-large-language-models-hallucinate-f05689-a1: Current. Outcome: Unresolved. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.
About assessment marcus-why-do-large-language-models-hallucinate-f05689-a1.