“And no matter what investors and CEOs might tell you, hallucinations will remain inevitable.”
Our reading, the sentence that was checked:
Hallucinations in large language models will remain inevitable.
Assessment: Unresolved No deadline stated on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Hello, Multimodal Hallucinations, paragraph 24.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-hello-multimodal-hallucinations-773cbd-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 2023-10-21. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Forecast
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No deadline stated
- Deadline state on 2026-10-29
- No deadline stated (the release’s reference date)
From the speaker
- hallucinations
- will remain inevitable
- no matter what investors and CEOs might tell you
Our interpretation
- The ordinary reading is that useful, general-purpose LLMs will retain some risk of producing false or unsupported assertions.
- No deadline is stated.
- Inevitable does not mean every answer hallucinates or that error rates cannot fall.
- A system that always abstains or operates only in a narrowly verified domain does not by itself settle the general-purpose claim.
- Near-zero benchmark error is relevant progress but is not equivalent to demonstrated elimination.
Counts as supported if: A validated impossibility result establishes unavoidable hallucination under assumptions that genuinely cover useful general-purpose LLMs and their permitted mitigations.
Counts as contradicted if: A useful general-purpose LLM system demonstrates a substantiated, independently checked guarantee against hallucinated assertions across its broad intended scope.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Did the forecast come true?
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 deadline stated
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
No decisive event found meets the early-resolution rule. Continuing hallucinations are consistent with the forecast, but do not prove inevitability. Formal lower bounds depend on specified assumptions, and improvements do not establish elimination. With no stated deadline, the forecast remains open.
Limitations
- A mathematical reading and a practical-risk reading could yield different assessments.
- Company-reported improvements are not independent guarantees.
- The search is not an exhaustive inventory of every system or theorem.
Evidence (3 items)
Context only: OpenAI,
but they still occur.
Why it matters here: The disclosure records continued hallucinations alongside improvement. This is consistent with persistence through that date but cannot establish that hallucinations will remain unavoidable indefinitely.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: Xu, Jain, and Kankanhalli via arXiv,
impossible to eliminate hallucination
Why it matters here: The revised paper establishes limitations involving computable models and ground-truth functions. Its universal formal task scope does not by itself establish unavoidable false assertions for every useful system that can verify answers or abstain.
- Publisher
- Xu, Jain, and Kankanhalli via arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Kalai and Vempala via arXiv,
may mitigate these latter types of hallucinations.
Why it matters here: The lower bound depends on calibration and particular kinds of facts. The authors distinguish those cases from repeated or systematic facts, so the result does not establish an unconditional guarantee of future hallucination.
- Publisher
- Kalai and Vempala via arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
An independently validated general-purpose system guarantee against hallucination, or a validated impossibility theorem demonstrably covering the forecast's ordinary scope, including useful abstention and external verification.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Hello, Multimodal Hallucinations
And no matter what investors and CEOs might tell you, hallucinations will remain inevitable.
- Where
- paragraph 24
- Published
- 2023-10-21
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-hello-multimodal-hallucinations-773cbd-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-hello-multimodal-hallucinations-773cbd-a1.