“Ultimately though, what LLMs are great at seeming is plausible – while being much less good at getting their facts straight.”
Our reading, the sentence that was checked:
LLMs are good at seeming plausible but much less good at getting facts right.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Three years on, ChatGPT still isn't what it was cracked up to be – and it probably never will be, paragraph 72.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-three-years-on-chatgpt-still-isnt-4e84e5-a1
We recorded 31 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 2025-11-29. 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
- LLMs
- great at seeming ... plausible
- much less good at getting their facts straight
Our interpretation
- The plain reading is that LLM answers often sound substantially more credible than their factual correctness warrants.
- Plausibility means credibility to human readers, not the model's expressed confidence.
- The claim describes the current state. With no reporting cycle specified, the target period is November 2025 through the post date.
- The comparison requires evidence connecting perceived credibility with verified correctness. Separate demonstrations of fluency and factual errors are insufficient to establish the size or breadth of the gap.
Counts as supported if: Evaluations of contemporary LLMs show a substantial, recurring gap between human assessments of answer credibility and independently verified factual correctness across a reasonable range of factual tasks.
Counts as contradicted if: Comparable evaluations show that perceived credibility generally tracks factual correctness, with no substantial recurring gap.
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 plausibility experiment shows that generated arguments can make incorrect choices seem more plausible. OpenAI also reports improved factual reliability. Neither measures the broad comparison Marcus makes for contemporary LLMs. These findings are relevant, but they do not settle the claim.
Limitations
- The located plausibility experiment concerns prompted rationales rather than ordinary factual answers.
- The factuality improvement is company-reported and predates the target period.
- Unresolved does not mean the claim is false.
Evidence (2 items)
Context only: arXiv; Shramay Palta and colleagues,
PRO rationales appear sufficiently persuasive
Why it matters here: Human readers found arguments for incorrect answer choices persuasive in this experiment. Its deliberately generated rationales and two commonsense datasets do not establish a general plausibility-versus-factuality gap for November's LLMs.
- Publisher
- arXiv; Shramay Palta and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: OpenAI,
less likely to hallucinate
Why it matters here: OpenAI reports improved factual reliability at GPT-5's release. This limits extrapolation from older models, but improvement alone neither establishes nor disproves Marcus's comparison with perceived plausibility.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
A dated evaluation of November 2025 model versions pairing blinded human credibility judgments with independent factual checks across varied factual questions.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Three years on, ChatGPT still isn't what it was cracked up to be – and it probably never will be
Ultimately though, what LLMs are great at seeming is plausible – while being much less good at getting their facts straight.
- Where
- paragraph 72
- Published
- 2025-11-29
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-three-years-on-chatgpt-still-isnt-4e84e5-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-three-years-on-chatgpt-still-isnt-4e84e5-a1.