“It couldn’t reason about the physical world (figure 4).”
Our reading, the sentence that was checked:
ChatGPT could not reliably reason about the physical world, though it could sometimes produce correct outputs.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Large Language Models like ChatGPT say The Darnedest Things, paragraph 3.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-large-language-models-like-chatgpt-24e93f-a1
We recorded 10 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-29. The source may have been edited after 2023-01-09. 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
- not reliably; it could sometimes produce correct outputs (qualifier, stated by the speaker)
- ChatGPT as released (November 2022) (scope, stated by the speaker)
From the speaker
- reason about the physical world
- It could sometimes produce outputs that were correct
- but not reliably
- when released
Our interpretation
- The surrounding qualification makes this a claim about unreliable performance, not total inability.
- The plain reading is that early ChatGPT repeatedly made substantive mistakes on ordinary physical reasoning questions, even though some answers were correct.
- The target is the release-era service from its November 30, 2022 launch through the post, not subsequent models.
Counts as supported if: Independent, documented tests of release-era ChatGPT show recurring errors or inconsistent answers on straightforward physical reasoning questions.
Counts as contradicted if: Broad, repeatable testing of the same release-era system establishes dependable physical reasoning and shows that the alleged failures were misreported or depended on materially misleading prompts. Isolated successes are insufficient.
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
Independent release-era testing documented repeated errors and contradictions on a basic motion question. That supports the qualified claim that ChatGPT could not be relied upon for physical reasoning. A separate evaluation found good physical commonsense performance on small samples. Those successes limit the breadth of the conclusion but do not contradict the explicit allowance for correct answers.
Limitations
- Support concerns demonstrated unreliability, not total inability or a measured failure rate.
- The strongest failure evidence is a narrow case study published after the post.
- This assessment does not extend to later ChatGPT versions.
Evidence (2 items)
Supports the claim: IOP Publishing, Physics Education,
often full of contradictions.
Why it matters here: Repeated questions about a thrown object's acceleration produced incorrect and inconsistent explanations alongside correct statements. December timing follows the manuscript's December receipt and the authors' retrospective description of experiments a few weeks after launch; the exact test day is unspecified.
- Publisher
- IOP Publishing, Physics Education
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Context only: arXiv; Yejin Bang and colleagues,
performs quite well
Why it matters here: The study identifies the December 15 version and reports good results on small commonsense samples, including physical questions. This limits any claim of general incapacity, but successful examples are compatible with Marcus's explicit qualification; the date identifies the tested version, not a documented testing day.
- Publisher
- arXiv; Yejin Bang and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Large Language Models like ChatGPT say The Darnedest Things
It couldn’t reason about the physical world (figure 4).
- Where
- paragraph 3
- Published
- 2023-01-09
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-large-language-models-like-chatgpt-24e93f-a1: 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-large-language-models-like-chatgpt-24e93f-a1.