“There were problems with experimental design, misleading graphs that left out competing work, and more.”
Our reading, the sentence that was checked:
OpenAI's December 2024 o3 demo had experimental design problems and misleading graphs that omitted competing work.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, OpenAI’s dirty December o3 demo doesn’t readily replicate, paragraph 1.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-openais-dirty-december-o3-demo-doesnt-460523-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 2025-04-23. 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
- problems with experimental design
- misleading graphs that left out competing work
- December 20th
Our interpretation
- The claim concerns the December 20, 2024 presentation.
- Experimental problems means methodological shortcomings that materially weaken the conclusions presented. Using designated training data does not alone establish a defect.
- Misleading omissions means excluding relevant competing results in a way that materially distorts the comparison. A graph explicitly restricted to OpenAI models need not represent the entire field.
- Both the methodological criticism and the misleading-graph criticism require evidence.
Counts as supported if: The presentation and contemporary research establish material methodological shortcomings and omitted competing results that distort the comparisons.
Counts as contradicted if: The experimental methods adequately support the conclusions, and the graphs accurately represent their stated comparison scope without materially misleading omissions.
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 records establish competing research and a missing comparison isolating ARC training's contribution. They do not establish the full compound allegation. Training on a designated training set is not by itself a design defect, and the cited MIT result used a different evaluation split. The accessible text also does not establish the complete graphs and their accompanying qualifications.
Limitations
- This is an evidence gap, not a finding that the presentation was sound.
- The meaning of misleading requires judging the actual presentation in context.
Evidence (2 items)
Context only: ARC Prize,
ARC-untrained model
Why it matters here: The evaluator disclosed training on public ARC training tasks and lacked a comparison isolating that training's contribution. This limits attribution of the improvement but does not itself establish an invalid benchmark experiment.
- Publisher
- ARC Prize
- Kind of source
- Independent measurement
- Describes events of
- Public from
Context only: Akyürek et al., arXiv,
61.9%
Why it matters here: Substantial competing work existed before the demo. Its public-validation score cannot directly substitute for a semi-private evaluation result or establish that a particular graph was misleading.
- Where
- Abstract
- Publisher
- Akyürek et al., arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A readable, authenticated transcript and slide record, together with matched competing results and documentation showing how any methodological differences affected the conclusions.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, OpenAI’s dirty December o3 demo doesn’t readily replicate
There were problems with experimental design, misleading graphs that left out competing work, and more.
- Where
- paragraph 1
- Published
- 2025-04-23
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-openais-dirty-december-o3-demo-doesnt-460523-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-openais-dirty-december-o3-demo-doesnt-460523-a1.