“What it actually shows is some improvements, mostly modest, over the last few months for a bunch of different measures.”
Our reading, the sentence that was checked:
OpenAI's GPT-4 Turbo graph shows only mostly modest improvements over recent months.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Evidence that LLMs are reaching a point of diminishing returns - and what that might mean, paragraph 8.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-evidence-that-llms-are-reaching-a-570f6e-a1
We recorded 20 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 2024-04-13. 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
- last few months (qualifier, stated by the speaker)
From the speaker
- some improvements
- mostly modest
- over the last few months
- a bunch of different measures
Our interpretation
- Assess the recent GPT-4 Turbo comparison published by OpenAI immediately before the post.
- Read mostly modest as small movements on most displayed benchmark-score scales, allowing larger changes on a minority.
- Use absolute score changes because the graph displays benchmark scores; do not silently substitute error reduction or a general capability multiplier.
- Interpret the last few months as January through the April 13 post. The comparison itself identifies January 25 and April 9 model versions.
Counts as supported if: The contemporaneous comparison shows improvements on several measures, with most changes amounting to only a few points on the displayed 0–100 scales.
Counts as contradicted if: Most displayed measures show large score changes, or the comparison fails to show improvements across several measures.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
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
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The contemporaneous primary table supports the ordinary graphical reading. In both prompting configurations, four of six changes are within about five score points; GPQA and MATH improve more. Marcus said mostly modest, which allows those larger gains. The result supports this narrow description of the recent comparison, not a conclusion about general capability growth or future progress.
Limitations
- Modest is a qualitative judgment, and relative changes can look larger.
- The readable chart posts omit the image's numerical content; the assessment uses their linked, dated primary evaluation table.
- The table reports company evaluations without confidence intervals.
- The historical repository revision is dated April 13; the announcement was published April 12.
Evidence (2 items)
Supports the claim: OpenAI,
Evals are sensitive to prompting, and there's significant variation in the formulations used in recent publications and libraries.
Why it matters here: In the assistant configuration, April-minus-January changes are 4.5 DROP F1 points and 7.9, 8.9, 4.5, 1.3, and 1.6 percentage points on GPQA, MATH, MGSM, MMLU, and HumanEval. The ChatGPT configuration likewise has small changes on most measures, with larger GPQA and MATH gains and a slight HumanEval decline.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Arithmetic
- Subtract each January score from its April counterpart within the same prompting configuration. Assistant differences: 4.5, 7.9, 8.9, 4.5, 1.3, 1.6. ChatGPT differences: 2.0, 9.4, 8.0, 4.9, 1.7, -0.6. (inputs: Assistant January, ordered DROP/GPQA/MATH/MGSM/MMLU/HumanEval: 81.5, 41.4, 64.5, 85.1, 85.4, 86.6.; Assistant April, same order: 86.0, 49.3, 73.4, 89.6, 86.7, 88.2.; ChatGPT January, same order: 83.4, 39.7, 64.2, 83.7, 84.8, 88.2.; ChatGPT April, same order: 85.4, 49.1, 72.2, 88.6, 86.5, 87.6.; One decimal place, matching the published scores.)
- Describes events of
- Public from
Context only: OpenAI,
We’ve improved capabilities in writing, math, logical reasoning, and coding.
Why it matters here: Identifies the release and improvements being discussed. The announcement links OpenAI's simple-evals repository, which supplies the numerical comparison.
- Where
- Announcement text
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Evidence that LLMs are reaching a point of diminishing returns - and what that might mean
What it actually shows is some improvements, mostly modest, over the last few months for a bunch of different measures.
- Where
- paragraph 8
- Published
- 2024-04-13
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-evidence-that-llms-are-reaching-a-570f6e-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-evidence-that-llms-are-reaching-a-570f6e-a1.