“ChatGPT has proven itself to be both unreliable and untruthful. It makes boneheaded arithmetical errors, invents fake biographical details, bungles word problems, defines non-existent scientific phenomenon, stumble over arithmetic conversion, and on and on.”
Our reading, the sentence that was checked:
ChatGPT is unreliable and untruthful: it makes arithmetic errors, fabricates biographical details, bungles word problems, and invents scientific phenomena.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Scientists, please don’t let your chatbots grow up to be co-authors, paragraph 18.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-scientists-please-dont-let-your-chatbots-fae914-a2
We recorded 11 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 2023-01-14. 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
- unreliable and untruthful
- arithmetical errors
- invents fake biographical details
- bungles word problems
- defines non-existent scientific phenomenon
- stumble over arithmetic conversion
Our interpretation
- Assess documented behavior during the six-week history discussed in the post, through January 14, 2023.
- Untruthful means producing false factual statements, without implying conscious intent to deceive.
- The ordinary reading is that ChatGPT demonstrably makes these kinds of errors, not that every answer is wrong or that a particular error rate has been measured.
- Treat the list as examples supporting the general reliability observation. Identify any example not independently recovered rather than silently treating it as verified.
Counts as supported if: Independent contemporary records demonstrate arithmetic or word-problem mistakes and fabricated biographical and scientific content presented as factual.
Counts as contradicted if: The cited demonstrations are shown to be inauthentic, outside the target period, or factually correct, and the alleged pattern of errors and fabrications is not substantiated. Some correct answers alone do not contradict an assertion that errors occur.
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
December 21, 2022 transcripts document arithmetic errors, and December 30 interactions document incorrect answers to verbal counting problems [e1, e2]. A December 5 expert account records ChatGPT explaining a nonexistent physical phenomenon and supplying nonexistent research references [e3]. An account describing December 2022 interactions records false biographical details about Homer Rodeheaver; the retrieved page includes later revisions [e4]. These records support the observation that early ChatGPT produced errors and false factual content [e1, e2, e3, e4].
Limitations
- Selected demonstrations do not establish an overall error rate or intentional deception [e1, e2, e3, e4].
- Some recorded responses were correct [e1, e2].
- The specific arithmetic-conversion example is not independently verified by these items [e1, e2, e3, e4].
- The biographical account is an updated page, with availability conservatively dated December 17, 2025, rather than a frozen December 2022 snapshot [e4].
Evidence (4 items)
Supports the claim: Vipul Naik, LessWrong; GreaterWrong mirror,
can be negative.
Why it matters here: The dated transcript records a false elementary arithmetic assertion and inconsistent calculations. It also contains correct answers, so it establishes fallibility rather than universal failure.
- Publisher
- Vipul Naik, LessWrong; GreaterWrong mirror
- Kind of source
- Demonstration
- Describes events of
- Public from
Supports the claim: Ernest Davis, arXiv,
Barbara was the fourth child to arrive to the birthday party.
Why it matters here: The immediately preceding prompt places Barbara third and Harry fourth. The table explicitly dates this interaction to December 30, establishing a contemporary failure on a simple verbal counting problem.
- Publisher
- Ernest Davis, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: Teresa Kubacka; Thread Reader archive,
a physical phenomenon that doesn’t exist.
Why it matters here: Kubacka reports asking about an invented electromagnon phenomenon and receiving a plausible explanation supported by nonexistent research. She describes checking the phenomenon and references using her subject expertise.
- Publisher
- Teresa Kubacka; Thread Reader archive
- Kind of source
- Demonstration
- Describes events of
- Public from
Supports the claim: Douglas Yeo, The Last Trombone,
born in Indiana
Why it matters here: The December 2022 account describes false biographical statements checked by Rodeheaver researchers, including his birthplace. The availability date conservatively uses the page's latest displayed revision because its text includes later additions.
- Publisher
- Douglas Yeo, The Last Trombone
- Kind of source
- Demonstration
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Scientists, please don’t let your chatbots grow up to be co-authors
ChatGPT has proven itself to be both unreliable and untruthful. It makes boneheaded arithmetical errors, invents fake biographical details, bungles word problems, defines non-existent scientific phenomenon, stumble over arithmetic conversion, and on and on.
- Where
- paragraph 18
- Published
- 2023-01-14
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-scientists-please-dont-let-your-chatbots-fae914-a2: 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-scientists-please-dont-let-your-chatbots-fae914-a2.