“Hallucinations haven’t solved, despite constant assurances.”
Our reading, the sentence that was checked:
Hallucinations have not been solved despite constant assurances that they would be.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Ezra Klein’s new take on AGI – and why I think it’s probably wrong, paragraph 46.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-ezra-kleins-new-take-on-agi-and-why-0b5f36-a2
We recorded 29 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 2025-03-05. 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
- Hallucinations haven’t solved
- despite constant assurances
Our interpretation
- Read the grammatical omission as 'hallucinations have not been solved'.
- The ordinary reading is that contemporary LLM systems still generate fabricated or unsupported factual answers in practical use. It does not claim that hallucination rates never improved.
- Use February 2025, the latest completed reporting month before the post, for the ongoing model-performance observation.
- Read 'constant assurances' as repeated public assurances about preventing or overcoming hallucinations, not a literal uninterrupted stream or a promise by every provider.
Counts as supported if: Contemporary model evaluations document continuing factual hallucinations, while earlier public records establish repeated assurances about preventing or overcoming them.
Counts as contradicted if: Contemporary evidence establishes a broadly effective solution to factual hallucinations in the systems under discussion. Improvements or successful narrow safeguards alone do not contradict the claim.
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
Hallucinations persisted in February 2025: OpenAI's February 27 evaluation reported PersonQA hallucination rates of 0.19 for GPT-4.5, 0.30 for GPT-4o and 0.20 for o1 [e1]. A research paper available May 30, 2024 documents repeated assurances from legal AI providers beginning in 2023 about preventing or mitigating hallucinations [e2]. The evidence also shows qualifications: GPT-4.5 scored better than the comparison models on PersonQA, and its February 27 launch announcement expressed an expectation of fewer hallucinations, without claiming elimination [e1, e3].
Limitations
- PersonQA is a company-reported, task-specific evaluation; its rates do not establish rates across everyday use.
- The documented assurances concern named legal AI providers, not all LLM developers.
- The evidence does not measure the frequency implied by 'constant' or establish whether hallucinations can eventually be solved.
Evidence (3 items)
Supports the claim: OpenAI,
More work is needed to understand hallucinations holistically
Why it matters here: Table 4 reports a PersonQA hallucination rate of 0.19 for GPT-4.5, compared with 0.30 for GPT-4o and 0.20 for o1. These results show improvement alongside substantial remaining failures; they do not measure all everyday usage.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: Stanford University researchers,
largely prevents hallucination in legal research tasks.
Why it matters here: The paper documents assurances from multiple legal AI providers beginning in 2023. These establish repeated assurances in a specific application domain, not universal promises by all LLM developers.
- Publisher
- Stanford University researchers
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: OpenAI,
We also expect it to hallucinate less.
Why it matters here: The contemporary launch promised an expected reduction, not elimination. This qualifies any reading that industry statements uniformly guaranteed a complete solution.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Ezra Klein’s new take on AGI – and why I think it’s probably wrong
Hallucinations haven’t solved, despite constant assurances.
- Where
- paragraph 46
- Published
- 2025-03-05
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-ezra-kleins-new-take-on-agi-and-why-0b5f36-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-ezra-kleins-new-take-on-agi-and-why-0b5f36-a2.