“(Note that test-time compute systems are not across the board improvements like GPT-4 was relative to GPT-3 or GPT-3 was relative to GPT-2, but rather improvements in certain domains such as coding and math.)”
Our reading, the sentence that was checked:
Test-time compute systems improve only certain domains like coding and math, not across the board as GPT-4 did over GPT-3.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Five ways in which the last 3 months — and especially the DeepSeek era — have vindicated “Deep learning is hitting a wall", paragraph 6.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-five-ways-in-which-the-last-3-months-b7401a-a1
We recorded 8 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 2025-02-09. 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
- test-time compute systems
- not across the board improvements
- like GPT-4 was relative to GPT-3
- improvements in certain domains such as coding and math
Our interpretation
- The plain reading is that contemporary reasoning systems offered selective improvements rather than a similarly broad advance in general usefulness. Coding and math are examples, not an exhaustive list.
- Across the board means broad improvement across ordinary task categories, not improvement on every individual benchmark.
- The target period is the three months preceding publication, following the post's stated context. Assessments concern model versions available then.
- A direct comparison with earlier GPT transitions requires reasonably comparable task coverage and baselines.
Counts as supported if: Cross-domain evaluations show that contemporary reasoning models' improvements were concentrated in selected domains, with substantial ordinary task categories showing little improvement or regression, and materially narrower gains than the earlier GPT transitions.
Counts as contradicted if: Comparable evaluations show broad improvements across technical and nontechnical tasks, similar in breadth to the earlier GPT transitions.
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 supports uneven gains: DeepSeek documented several areas where R1 remained worse than V3. But the records also show improvements outside math and coding. Marcus said 'such as,' so those additional gains do not automatically refute him. The stronger comparison with the breadth of earlier GPT advances remains unsettled without a common evaluation and a clearer breadth threshold.
Limitations
- The decisive model evaluations are developer-reported.
- Treating across-the-board improvement as literally zero regressions would make the claim easier to support than its ordinary reading warrants.
Evidence (2 items)
Supports the claim: DeepSeek-AI / arXiv,
fall short of DeepSeek-V3
Why it matters here: These limitations support the selective-improvement reading. However, the same report also describes gains in factual questions, document analysis, writing, and open-domain answering, so it does not establish that improvements were narrowly confined.
- Publisher
- DeepSeek-AI / arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: OpenAI,
hallucinate less frequently
Why it matters here: The factuality evaluations provide material evidence of benefits beyond math and coding. They do not by themselves establish an advance comparable in breadth to the earlier GPT transitions.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
A comparable, version-specific evaluation across technical and nontechnical domains for both the reasoning-model transition and the earlier GPT transitions.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Five ways in which the last 3 months — and especially the DeepSeek era — have vindicated “Deep learning is hitting a wall"
(Note that test-time compute systems are not across the board improvements like GPT-4 was relative to GPT-3 or GPT-3 was relative to GPT-2, but rather improvements in certain domains such as coding and math.)
- Where
- paragraph 6
- Published
- 2025-02-09
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-five-ways-in-which-the-last-3-months-b7401a-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-five-ways-in-which-the-last-3-months-b7401a-a1.