“if the definition of reasoning is to obtain valid conclusions from known facts, GPT-4 frequently falls short there, too.”
Our reading, the sentence that was checked:
GPT-4 frequently fails to reason, if reasoning means reaching valid conclusions from known facts.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, This Week in AI Doublespeak, paragraph 15.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-this-week-in-ai-doublespeak-d79f2e-a1
We recorded 3 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 2023-03-18. 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)
Conditions
- reasoning defined as obtaining valid conclusions from known facts (qualifier, stated by the speaker)
From the speaker
- reasoning is to obtain valid conclusions from known facts
- GPT-4 frequently falls short
Our interpretation
- The definition of reasoning fixes what is being assessed; it is not a forecast antecedent.
- The ordinary reading is recurring failures to draw valid conclusions from supplied or established premises, rather than merely failures to recall facts.
- Frequently requires evidence of recurring errors across a defined collection of relevant tasks. An isolated failure establishes possibility but not frequency.
- The target is the original GPT-4 during its March 14–18, 2023 launch period, the latest public model period before the post.
Counts as supported if: Evaluations attributable to the launch-period model show recurring invalid conclusions on a defined set of tasks whose necessary premises are supplied or established.
Counts as contradicted if: Broad evaluations attributable to that model show consistently valid conclusions on relevant tasks, with failures sufficiently exceptional to contradict the ordinary meaning of frequently.
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
The evidence points toward recurring reasoning failures, but the retrieved records do not securely establish their frequency for the model state described on March 18. The independent logical-reasoning study is relevant, yet its testing dates and GPT-4 snapshot are insufficiently specified. This is an evidence gap, not evidence that GPT-4 reasoned reliably.
Limitations
- Later evaluations were not rejected merely because they were published after the post; the problem is their uncertain connection to the target model period.
- Strong performance on some benchmarks can coexist with frequent failures on others.
- This assessment does not treat Marcus as claiming that GPT-4 never reasons correctly.
Evidence (1 item)
Context only: Hanmeng Liu and colleagues, arXiv,
Logical reasoning remains challenging
Why it matters here: The study supports recurring reasoning difficulties in early GPT-4. Its exact GPT-4 testing dates and model snapshot are not established, so it cannot serve as decisive evidence for March 14–18.
- Publisher
- Hanmeng Liu and colleagues, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
Dated launch-period test records, or a verified unchanged launch snapshot, evaluated on a defined collection of supplied-premise reasoning tasks with answer-level error analysis.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, This Week in AI Doublespeak
if the definition of reasoning is to obtain valid conclusions from known facts, GPT-4 frequently falls short there, too.
- Where
- paragraph 15
- Published
- 2023-03-18
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-this-week-in-ai-doublespeak-d79f2e-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-this-week-in-ai-doublespeak-d79f2e-a1.