“That is not going to magically change next year, at a time when AI still can’t reliably solving river-crossing problems.”
Our reading, the sentence that was checked:
Current AI cannot reliably solve river-crossing puzzles.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Nobel Prizes and The AI Hype Hall of Fame, paragraph 16.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-nobel-prizes-and-the-ai-hype-hall-612ee6-2-a1
We recorded 7 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 2025-03-09. 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
- AI still can’t reliably
- river-crossing problems
Our interpretation
- In context, AI means contemporary general-purpose language and reasoning models, rather than every historical programmed puzzle solver.
- Reliably means consistent success across varied, solvable river-crossing puzzles, including ordinary changes to familiar constraints. One failure or one success is insufficient.
- Marcus supplies no numerical threshold, model list, prompting protocol, or tool policy; I do not invent a percentage cutoff.
- For the ongoing observation, I use the current-year reporting window before the post, January 1 through March 9, 2025. This is an evaluator-selected period, not an explicit interval in his words.
- Later evaluations are decisive only if they establish performance of the relevant model versions during that target period; otherwise they are context.
Counts as supported if: Reproducible evaluations of leading models in the target period show recurring material failures across varied, solvable river-crossing tasks under clearly described, reasonable conditions.
Counts as contradicted if: Reproducible evaluations establish consistent success by leading contemporary models across a sufficiently varied river-crossing suite, rather than isolated familiar examples.
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 search found failures, successes, and substantial sensitivity to puzzle design and prompting. However, the cited systematic studies fall outside the target period, while the illustrative examples do not establish reliability across March 2025 systems. Later failures cannot simply be backdated, and later successes cannot be treated as contemporaneous counterexamples. The available record does not settle the historical claim.
Limitations
- Unresolved does not mean the models were reliable.
- An isolated wrong answer establishes a failure, not the failure rate of the class.
- The later studies are methodological context, not decisive historical evidence.
Evidence (3 items)
Context only: arXiv; Parshin Shojaee and colleagues,
a complete accuracy collapse
Why it matters here: The paper reports failures on controlled puzzles, including river crossing. Its original record does not date the experiments within January 1–March 9, so it cannot establish the historical observation under this contract.
- Where
- Abstract
- Publisher
- arXiv; Parshin Shojaee and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Iñaki Dellibarda Varela and colleagues,
over 100 agent pairs
Why it matters here: This replication reports strong river-crossing performance with solvable configurations and revised interaction methods. It limits broad conclusions from failure studies, but does not establish performance during Marcus's target period.
- Where
- Abstract
- Publisher
- arXiv; Iñaki Dellibarda Varela and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Sean Williams and James Huckle,
kept separate during the crossing.
Why it matters here: The paper gives GPT-4 Turbo a successful response to a modified river-crossing puzzle while showing failures by other models. This cautions against treating isolated failures as universal inability, but the tests precede the target period.
- Publisher
- arXiv; Sean Williams and James Huckle
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
Dated evaluation logs, or a faithful evaluation of preserved contemporary model versions, covering varied solvable puzzles with repeated trials and disclosed prompts and tool access.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Nobel Prizes and The AI Hype Hall of Fame
That is not going to magically change next year, at a time when AI still can’t reliably solving river-crossing problems.
- Where
- paragraph 16
- Published
- 2025-03-09
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-nobel-prizes-and-the-ai-hype-hall-612ee6-2-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-nobel-prizes-and-the-ai-hype-hall-612ee6-2-a1.