“Reasoning flubs will continue to haunt generative AI.”
Our reading, the sentence that was checked:
Through 2025 reasoning errors will continue to plague generative AI.
Assessment: Supported Deadline passed on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, 25 AI Predictions for 2025, from Marcus on AI, paragraph 31.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-25-ai-predictions-for-2025-from-marcus-22c0aa-a3
We recorded 38 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 2025-01-01. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Forecast
- Deadline, as the speaker wrote it
- “end of this year”
- Deadline state when assessed
- Deadline passed
- Deadline state on 2026-10-29
- Deadline passed (the release’s reference date)
From the speaker
- Reasoning flubs
- will continue to haunt generative AI
- where we will be at the end of this year
Our interpretation
- Plain reading: through the end of 2025, leading generative AI models, including 'reasoning' models, still make well-documented reasoning errors that researchers and benchmarks treat as a real limitation.
- Strong results on some hard tasks do not by themselves contradict the claim if documented reasoning failures remain common elsewhere.
Counts as supported if: Credible 2025 research or benchmarks show leading models failing at reasoning tasks in systematic or notable ways, such as collapse on harder problems or low scores on reasoning benchmarks humans can solve.
Counts as contradicted if: Credible evidence shows that by late 2025 leading models reasoned reliably across standard reasoning tests, with reasoning errors rare and no longer a notable limitation.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Did the forecast come true?
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
- Deadline passed
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The forecast period ended on 31 December 2025. Apple researchers first posted a paper on 7 June 2025, and it was accepted at NeurIPS 2025. It reported that large reasoning models face a complete accuracy collapse beyond certain puzzle complexities (e1). On 5 December 2025, the ARC Prize Foundation reported that the top verified commercial model, Opus 4.5 (Thinking, 64k), scored 37.6% on ARC-AGI-2. It also said the Grand Prize remained unclaimed (e2). The same post noted material progress on ARC-AGI-2 during 2025 (e2). In July 2025, an advanced Gemini Deep Think model and an OpenAI model each scored 35 of 42 at the International Mathematical Olympiad, which is gold-medal standard. Each missed only the hardest problem (e3). That shows strong results on some hard tasks. It does not show that reasoning errors became rare. Under the contract, documented reasoning failures in leading models during 2025 count as support, so the forecast holds on a plain reading.
Limitations
- 'Flubs' and 'haunt' set no threshold, so the verdict rests on a plain reading.
- The evidence does not test whether the Apple paper's methods and conclusions hold up. The packet notes that critiques of the paper were not reviewed.
- The evidence covers controlled puzzles and one benchmark. It does not measure how often reasoning errors occur in everyday use.
- The math olympiad result comes from a secondary summary (e3). Google's own announcement could not be read.
Evidence (3 items)
Supports the claim: Apple researchers (arXiv; NeurIPS 2025),
Through extensive experiments, we show that LRMs face a complete accuracy collapse beyond certain complexities.
Why it matters here: A 2025 study found that frontier reasoning models' accuracy collapses entirely beyond certain puzzle complexities. That is a documented, systematic reasoning failure.
- Where
- Abstract
- Publisher
- Apple researchers (arXiv; NeurIPS 2025)
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Supports the claim: ARC Prize Foundation,
As of today, the top verified commercial model, Opus 4.5 (Thinking, 64k), scores 37.6% for $2.20/task.
Why it matters here: At year end, the best verified commercial model still failed most ARC-AGI-2 reasoning tasks, and the Grand Prize stayed unclaimed. The same post also notes material progress.
- Publisher
- ARC Prize Foundation
- Kind of source
- Benchmark result
- Describes events of
- Public from
Goes against the claim: Simon Willison's Weblog,
Gemini achieved the exact same score as OpenAI, 35/42, and were able to solve the same set of questions - 1 through 5, failing only to answer 6
Why it matters here: Gold-medal-level results at the 2025 Math Olympiad show strong reasoning on some hard tasks. This cuts against the claim's tone, though it does not show that reasoning errors became rare in general.
- Where
- paragraph 3
- Publisher
- Simon Willison's Weblog
- Kind of source
- Secondary summary
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, 25 AI Predictions for 2025, from Marcus on AI
Reasoning flubs will continue to haunt generative AI.
- Where
- paragraph 31
- Published
- 2025-01-01
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-25-ai-predictions-for-2025-from-marcus-22c0aa-a3: 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-25-ai-predictions-for-2025-from-marcus-22c0aa-a3.