“This is not just a problem with ChatGPT, by the way. The problem has been around for years.”
Our reading, the sentence that was checked:
LLMs' failure at chess world models goes beyond ChatGPT and has lasted for years.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Generative AI’s crippling and widespread failure to induce robust models of the world, paragraph 50.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-generative-ais-crippling-and-widespread-2e3047-a3
We recorded 21 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 2025-06-28. 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)
From the speaker
- not just a problem with ChatGPT
- The problem has been around for years
- I first took note of the problem in 2023
Our interpretation
- In context, the observable problem is unreliable chess play, including illegal moves and board-tracking failures.
- Persistence means documented recurrence across years, not failure by every model in every game.
- Beyond ChatGPT requires another identifiable model or system. Evidence from a different provider is especially clear.
- The context explicitly anchors the historical comparison in 2023; use January 1, 2023 through June 28, 2025.
- This assessment does not treat persistent illegal moves as proof that internal world models are absent.
Counts as supported if: Dated primary tests document chess-rule or board-tracking failures in 2023 and again before publication in 2025, including a system beyond ChatGPT.
Counts as contradicted if: Comparable historical records establish that the alleged failures did not recur across those years or were confined to ChatGPT.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Supported
The dated evidence matches what the claim committed to, under the criteria frozen before grading.
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
Acher's 2023 tests found illegal moves in 72 of 73 games with text-davinci-003 [e1]. His January 2025 report documented repeated illegal moves by DeepSeek-R1, a model from another provider [e2]. These results support the historical claim that chess-rule failures persisted across years and extended beyond ChatGPT before June 28, 2025 [e1, e2]. They establish observable failures, not the proposed explanation that LLMs lack world models [e1, e2].
Limitations
- The tests do not establish the absence of internal world models or a shared cause of the failures [e1, e2].
- The reports establish recurrence, not continuous failure or prevalence across all LLMs; both come from the same investigator [e1, e2].
Evidence (2 items)
Supports the claim: Mathieu Acher,
72 games contain illegal moves.
Why it matters here: The dated evaluation documents recurring illegal moves in text-davinci-003 during 2023. This establishes an early instance of the observable problem beyond the ChatGPT interface.
- Publisher
- Mathieu Acher
- Kind of source
- Independent measurement
- Describes events of
- Public from
Supports the claim: Mathieu Acher,
DeepSeek-R1 made again an illegal move: 8. Bxb6!
Why it matters here: The report documents a concrete illegal move by DeepSeek-R1, alongside other game examples. It establishes recurrence in 2025 in a model from another provider.
- Publisher
- Mathieu Acher
- Kind of source
- Independent measurement
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Generative AI’s crippling and widespread failure to induce robust models of the world
This is not just a problem with ChatGPT, by the way. The problem has been around for years.
- Where
- paragraph 50
- Published
- 2025-06-28
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-generative-ais-crippling-and-widespread-2e3047-a3: Current. Outcome: Supported. 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-generative-ais-crippling-and-widespread-2e3047-a3.
HeldBy this site
The final-text check failed: proposition_faithful.
About assessment marcus-generative-ais-crippling-and-widespread-2e3047-a2, which was not published.