“It doesn’t actually solve any of the core problems of truthfulness and reliability that I laid out in my infamous March 2022 essay Deep Learning is Hitting a Wall.”
Our reading, the sentence that was checked:
GPT-4 does not solve any of the core truthfulness and reliability problems identified in Marcus's March 2022 essay.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, GPT-4’s successes, and GPT-4’s failures, paragraph 4.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-gpt-4s-successes-and-gpt-4s-failures-73f773-a1
We recorded 13 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 2023-03-15. 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
- doesn’t actually solve any
- core problems of truthfulness and reliability
- March 2022 essay Deep Learning is Hitting a Wall
Our interpretation
- Solve means provide dependable performance on a problem class, rather than merely improve an average benchmark score. It does not require mathematical perfection.
- The referenced essay supplies the scope: factual trustworthiness, reasoning and common sense, and robustness on unfamiliar or outlying inputs.
- The word any makes the statement broader than the narrower claim that some hallucinations and reasoning failures remain.
- The original model's launch-to-post state is the latest public snapshot appropriate to this ongoing observation.
Counts as supported if: Evidence mapped to the referenced core problem classes shows that each remained materially unresolved in the original GPT-4.
Counts as contradicted if: Appropriate evaluations establish that the original GPT-4 dependably solved at least one of those core problem classes.
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
Continuing hallucinations and reasoning failures are directly documented. That supports the narrower claim that central reliability problems remained. It does not by itself establish the full assertion that none of the referenced core problems was solved. The broader wording remains unresolved; improved factuality also prevents treating this as a claim that GPT-4 made no progress.
Limitations
- This is an evidence gap for the full assertion, not evidence that the problems were solved.
- Reading any loosely as fundamental problems remain would support a narrower assessment.
Evidence (2 items)
Supports the claim: OpenAI,
Most importantly, it still is not fully reliable (it “hallucinates” facts and makes reasoning errors).
Why it matters here: OpenAI directly acknowledges continuing factual and reasoning failures, supporting those parts of the claim. The same section reports a 40% improvement over its latest GPT-3.5 on internal adversarial factuality evaluations, which limits any reading that no progress occurred.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: Nautilus,
Deep-learning systems are particularly problematic when it comes to “outliers” that differ substantially from the things on which they are trained.
Why it matters here: This establishes that the referenced scope includes robustness beyond familiar training examples. It is Marcus's own writing and is used only to define scope, never as outcome evidence.
- Publisher
- Nautilus
- Kind of source
- Secondary summary
- Describes events of
- Public from
What would settle it
A defensible mapping of the referenced core problems to evaluations of the launch model, with evidence of whether each problem class remained materially unresolved.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, GPT-4’s successes, and GPT-4’s failures
It doesn’t actually solve any of the core problems of truthfulness and reliability that I laid out in my infamous March 2022 essay Deep Learning is Hitting a Wall.
- Where
- paragraph 4
- Published
- 2023-03-15
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-gpt-4s-successes-and-gpt-4s-failures-73f773-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-gpt-4s-successes-and-gpt-4s-failures-73f773-a1.