“I think that the recurring core technical problems that we have seen (as discussed below) with LLMs aren’t going way; instead they inherent to the technology.”
Our reading, the sentence that was checked:
The recurring core technical problems of LLMs are inherent to the technology and will not go away.
Assessment: Unresolved No deadline stated on 2026-10-29 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 6.
- 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-c20947-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
- 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
- recurring core technical problems
- with LLMs
- aren’t going way
- inherent to the technology
Our interpretation
- The statement combines an indefinite forecast with an explanation based on the technology itself.
- The post identifies unreliable reasoning, hallucinations, unreliable external-tool use, controllability and alignment among the recurring problems.
- The ordinary reading concerns persistent practical failure patterns. It does not require mathematically perfect answers to every possible question.
- Reducing error rates is relevant progress but does not by itself show that the recurring problems have gone away.
Counts as supported if: Evidence establishes that the named practical failure patterns necessarily persist across the relevant LLM technology, rather than merely appearing in current implementations.
Counts as contradicted if: Independent, reproducible evidence shows that an in-scope LLM system has reliably overcome the named recurring problems across varied, unfamiliar tasks and realistic deployment conditions.
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
Research supports persistence under some formal assumptions, while other research identifies changeable incentives that contribute to hallucination. Neither establishes the whole claim about practical reasoning, tool use and alignment. Continuing errors cannot prove that these problems will never go away, and improvements alone do not refute Marcus's forecast.
Limitations
- The formal and practical meanings of hallucination differ.
- The forecast bundles several problems without thresholds or a deadline.
- Evidence about one failure class cannot settle the entire bundle.
Evidence (3 items)
Context only: arXiv,
impossible to eliminate hallucination
Why it matters here: The paper supplies theoretical support under a definition involving computable models and ground-truth functions. That does not establish that all the practical failure patterns Marcus names are unavoidable at useful human-level reliability.
- Publisher
- arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Nature,
incentive problem
Why it matters here: The authors identify training and evaluation incentives that reward guessing and propose changes. This challenges an unqualified claim of architectural inevitability, but does not demonstrate elimination of hallucinations or the other named problems.
- Publisher
- Nature
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Context only: OpenAI,
but they still occur.
Why it matters here: The company reports both improvement and continuing hallucinations before the forecast. This establishes background persistence, not permanent necessity or a subsequent resolution.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
Broad, independently replicated evidence that the named practical failures have been overcome within LLM technology, or a validated necessity result covering those failures under realistic conditions.
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
I think that the recurring core technical problems that we have seen (as discussed below) with LLMs aren’t going way; instead they inherent to the technology.
- Where
- paragraph 6
- Published
- 2025-11-29
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-three-years-on-chatgpt-still-isnt-c20947-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-c20947-a1.