“I would be very surprised if that fundamental lack of reliability did not also hold true for GPT-4. Maybe it will take three paragraphs in a prompt rather than one to break it, but a big Transformer alone does not a a theory of the world create.”
Our reading, the sentence that was checked:
GPT-4 will very likely still have a fundamental lack of reliability, though it may take longer prompts to break.
Assessment: Supported Settled before its deadline on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, What does it mean when an AI fails? A Reply to SlateStarCodex’s riff on Gary Marcus, paragraph 50.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-what-does-it-mean-when-an-ai-fails-7fce77-a1
We recorded 14 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 2022-06-07. 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
- Settled before its deadline
- Deadline state on 2026-10-29
- Settled before its deadline (the release’s reference date)
From the speaker
- I would be very surprised
- fundamental lack of reliability
- also hold true for GPT-4
- Maybe it will take three paragraphs in a prompt rather than one
Our interpretation
- The ordinary reading is that GPT-4 would retain recurring factual and reasoning failures serious enough to prevent unqualified reliance on its answers.
- The statement allows improvement over earlier models and does not predict a particular error rate.
- The three-paragraph remark is illustrative, not a firm prompt-length threshold.
- The target is the forthcoming GPT-4 generation. The prediction can be resolved by evidence about that model after release; it does not require waiting for every future revision.
- This assessment addresses reliability, not the separate philosophical assertion about whether Transformers create theories of the world.
Counts as supported if: Dated evidence about released GPT-4 documents recurring hallucinations and reasoning errors that materially limit reliable use.
Counts as contradicted if: Broad, reproducible testing establishes robust factual and reasoning reliability across the relevant tasks, including adversarial variations, such that the predicted recurring limitations are absent.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Did the forecast come true?
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
- Settled before its deadline
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
OpenAI's launch-era technical report explicitly states that GPT-4 still hallucinates facts and makes reasoning errors. These are the recurring reliability problems the forecast anticipated. Substantial improvements and impressive capabilities do not contradict this forecast, which expressly allowed fewer failures. The evidence does not settle whether GPT-4 possesses a causal world model.
Limitations
- Support applies to recurring factual and reasoning unreliability, not a claim that GPT-4 has no useful capabilities.
- No numerical reliability threshold or causal explanation is established.
- The decisive disclosure is from the developer.
Evidence (2 items)
Supports the claim: OpenAI / arXiv,
makes reasoning errors
Why it matters here: OpenAI's launch-era report explicitly acknowledges both predicted failure classes and advises safeguards where reliability matters. It also reports substantial factuality improvement, which is compatible with Marcus's allowance for fewer errors.
- Publisher
- OpenAI / arXiv
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: Microsoft Research researchers / arXiv,
a true paradigm shift
Why it matters here: The researchers interpret their experiments as substantial progress, cautioning against equating unreliability with an absence of capability. Their study concerns an early development model and also acknowledges limitations, so it does not contradict recurring failures in released GPT-4.
- Publisher
- Microsoft Research researchers / arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, What does it mean when an AI fails? A Reply to SlateStarCodex’s riff on Gary Marcus
I would be very surprised if that fundamental lack of reliability did not also hold true for GPT-4. Maybe it will take three paragraphs in a prompt rather than one to break it, but a big Transformer alone does not a a theory of the world create.
- Where
- paragraph 50
- Published
- 2022-06-07
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-what-does-it-mean-when-an-ai-fails-7fce77-a1: 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-what-does-it-mean-when-an-ai-fails-7fce77-a1.