“Any one example can be fixed, but examples keep rolling in because no robust solution has been found.”
Our reading, the sentence that was checked:
LLM commonsense failure examples keep appearing because no robust solution has been found, even though individual examples can be fixed.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, AI still lacks “common” sense, 70 years later, paragraph 19.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-ai-still-lacks-common-sense-70-years-34937c-a1
We recorded 15 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 2025-01-05. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Causal claim
- 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
- Any one example can be fixed
- examples keep rolling in
- because no robust solution has been found
Our interpretation
- In context, examples refers to LLM commonsense failures, especially physical and spatial failures.
- A robust solution means dependable generalization to new problems rather than success only on previously identified examples.
- Any one example can be fixed is read as the practical possibility of correcting individual cases, not a formal theorem about every conceivable input.
- The causal proposition is that fixes fail to generalize adequately, explaining continued failures.
- The ongoing state is assigned to late 2024 through the publication date, 2025-01-05.
Counts as supported if: Before-and-after evaluations show correction of individual commonsense failures but recurring failures on new variants, with controls connecting that persistence to inadequate generalization.
Counts as contradicted if: A validated method corrects individual cases and reliably generalizes across diverse held-out commonsense tasks, or controlled evidence establishes another explanation for the continuing failures.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
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
Research documents continuing spatial errors, and intervention research shows that some improvements transfer beyond trained tasks. Neither establishes the claimed sequence of isolated fixes followed by new failures caused by inadequate generalization. The evidence does not justify calling the complete causal claim supported or contradicted.
Limitations
- Spatial reasoning studies cover only part of commonsense reasoning.
- Partial generalization is compatible with continued lack of broad reliability.
- No numerical definition of robust is supplied.
Evidence (2 items)
Context only: arXiv; Himanshu Gupta and colleagues,
spatial relations
Why it matters here: PolyMATH documents spatial and reasoning errors in contemporary multimodal models. It does not follow individual fixes over time or establish why new failures appear; precise experimental dates are also unspecified.
- Where
- Abstract
- Publisher
- arXiv; Himanshu Gupta and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Yihong Tang and colleagues,
spatial reasoning improvements
Why it matters here: Sparkle reports transfer from training basic spatial capabilities to more complex and different visual tasks. This cautions against assuming improvements merely patch isolated cases, but it does not demonstrate a comprehensive commonsense solution.
- Publisher
- arXiv; Yihong Tang and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
Controlled studies tracking specific commonsense corrections and performance on fresh variants in the same contemporary models, with evidence separating inadequate generalization from other causes of failure.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, AI still lacks “common” sense, 70 years later
Any one example can be fixed, but examples keep rolling in because no robust solution has been found.
- Where
- paragraph 19
- Published
- 2025-01-05
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-ai-still-lacks-common-sense-70-years-34937c-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-ai-still-lacks-common-sense-70-years-34937c-a1.