“But they don’t operate like people. They don’t, for example, ever fact check (as humans sometimes, when well motivated, do). They mimic the kinds of things of people say in various contexts. And that’s essentially all they do.”
Our reading, the sentence that was checked:
LLMs do not operate like people and never fact check; they essentially only mimic what people say in contexts.
Assessment: Contradicted Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Why DO large language models hallucinate?, paragraph 15.
- 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-65b7e1-a1
We recorded 16 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)
From the speaker
- they don’t operate like people
- They don’t, for example, ever fact check
- They mimic the kinds of things of people say in various contexts
- that’s essentially all they do
Our interpretation
- The testable categorical commitment is that LLMs never fact-check.
- Fact-checking means comparing factual assertions with evidence and using the comparison to accept, reject or revise them. It need not be infallible or conscious.
- Include LLM-based systems with retrieval and prompting because the surrounding discussion concerns Google AI Overviews and checking against Wikipedia.
- The word 'ever' includes demonstrations before publication. The target therefore has no specified beginning and ends on 2025-05-05.
- A counterexample to never fact-checking does not establish that LLMs share human mental mechanisms.
Counts as supported if: Evidence establishes that the systems in scope never perform factual verification, including when prompted or given external evidence.
Counts as contradicted if: At least one documented pre-publication LLM system compares assertions with evidence and uses the comparison to revise or validate its output.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Contradicted
The dated evidence goes against 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
RARR was documented before the post. Its language-model components compare generated claims with retrieved evidence and revise unsupported content, with evaluated improvements. That contradicts 'ever fact check' on the ordinary functional reading appropriate to the AI application discussed. It does not prove that models think like people or automatically verify every answer.
Limitations
- The counterexample uses an engineered workflow with retrieval and prompting.
- The outcome depends on including LLM-based applications and treating fact-checking functionally.
- The broader philosophical statements about mimicry and human cognition are not independently resolved.
Evidence (1 item)
Goes against the claim: Association for Computational Linguistics; author manuscript hosted by arXiv,
RARR significantly improves attribution while preserving most of the original text.
Why it matters here: RARR uses PaLM to compare generated assertions with retrieved evidence and edits the passage when disagreement is detected. Its evaluated research-and-revision procedure is a counterexample to the categorical claim under the functional reading.
- Publisher
- Association for Computational Linguistics; author manuscript hosted by arXiv
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Why DO large language models hallucinate?
But they don’t operate like people. They don’t, for example, ever fact check (as humans sometimes, when well motivated, do). They mimic the kinds of things of people say in various contexts. And that’s essentially all they do.
- Where
- paragraph 15
- Published
- 2025-05-05
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-why-do-large-language-models-hallucinate-65b7e1-a1: Current. Outcome: Contradicted. 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-65b7e1-a1.