“A recent paper on Putnam math problems showed a similar failure on distribution shift, showing 30% decrement by o1 on math problems with minor variations eg on variable names.”
Our reading, the sentence that was checked:
A recent paper showed o1's performance drops about 30% on Putnam math problems with minor variations like changed variable names.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Why I don’t share Sam Altman’s confidence that AGI is basically a solved problem, paragraph 10.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-sam-altman-thinks-that-agi-is-basically-3a4d5f-a2
We recorded 16 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 2025-01-06. 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
- minor variations such as variable names (qualifier, stated by the speaker)
From the speaker
- A recent paper on Putnam math problems
- 30% decrement
- by o1
- minor variations eg on variable names
Our interpretation
- This is primarily a claim about what a particular paper reported, not an independent replication claim.
- Thirty percent is read as an approximate relative accuracy reduction on matched original and altered questions, not thirty percentage points.
- Variable-name changes are offered as an example; the wording does not require the entire decline to result from renaming alone.
- O1 could mean the model family or the specific model bearing that name. Results for o1-preview must be identified explicitly rather than silently substituted.
- The target is the recent pre-publication study discussed in the thread, bounded by the 2024 research context and January 6, 2025.
Counts as supported if: A securely identified pre-post version reports an approximately thirty percent relative decline for the intended o1 model on matched Putnam problems with the described variations.
Counts as contradicted if: The identified version reports a materially different result, or establishes that the claimed model and perturbation comparison were not tested.
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 January 16, 2025 reproduction quotes an abstract reporting around a 30% accuracy reduction for o1-preview on variations compared with corresponding original Putnam problems [e1]. It describes alterations to variables and constants [e1]. This supports the numerical report's plausibility, but does not establish what a manuscript available by January 6, 2025 reported, or establish the result for the specifically named o1 model [e1]. The claim remains unresolved.
Limitations
- The evidence does not securely identify and date the manuscript version available before the claim.
- The reproduced abstract names o1-preview, leaving the intended meaning of o1 unresolved.
- The reproduction does not provide both matched accuracy figures needed to check the relative decline.
- The evidence does not isolate the effect of variable-name changes.
Evidence (1 item)
Context only: Michael Tsai,
significantly lower accuracy
Why it matters here: The reproduced abstract attributes an approximately thirty percent relative decline to o1-preview on corresponding original and variation problems. This corroborates the numerical report, but the dated reproduction falls outside the target period and does not establish results for a different o1 version.
- Publisher
- Michael Tsai
- Kind of source
- Secondary summary
- Describes events of
- Public from
What would settle it
A verifiably dated manuscript available by January 6, 2025, identifying the tested model, matched problem subset, perturbations, and original and variation accuracies. Assessment would also require resolving whether the claim's o1 means the model family or the specifically named model.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Why I don’t share Sam Altman’s confidence that AGI is basically a solved problem
A recent paper on Putnam math problems showed a similar failure on distribution shift, showing 30% decrement by o1 on math problems with minor variations eg on variable names.
- Where
- paragraph 10
- Published
- 2025-01-06
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-sam-altman-thinks-that-agi-is-basically-3a4d5f-a2: 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-sam-altman-thinks-that-agi-is-basically-3a4d5f-a2.