“But I am willing to wager that if someone put in that kind of effort and carefully built out the test, no current system would come close to human performance.”
Our reading, the sentence that was checked:
If a careful comprehension challenge test were built, no current AI system would come close to human performance on it.
Assessment: Unresolved No deadline stated on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, AI has (sort of) passed the Turing Test; here’s why that hardly matters, paragraph 34.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-ai-has-sort-of-passed-the-turing-cef6a3-a3
We recorded 9 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 2025-04-02. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Conditional forecast
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No deadline stated
- Deadline state on 2026-10-29
- No deadline stated (the release’s reference date)
Conditions
- if someone carefully built out the comprehension challenge test (antecedent, stated by the speaker)
- current systems (antecedent, stated by the speaker)
From the speaker
- if someone put in that kind of effort
- carefully built out the test
- no current system
- would come close to human performance
Our interpretation
- The condition is that someone builds the comprehension challenge he describes: a well-funded ($25-50M scale) test where systems watch full TV shows or films and answer questions. The questions are refreshed to avoid contamination, answers are scored carefully (perhaps by hand), and the machine is kept on the same footing as a first-time human viewer.
- 'Current system' means AI systems available around the post date, 2025-04-02.
- 'Not come close' means a large gap between the best such system and human viewers.
Counts as supported if: A test meeting the condition is built, and systems from around April 2025 score far below human viewers on it.
Counts as contradicted if: A test meeting the condition is built, and a system from around April 2025 scores at or near human level on it.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Did the forecast come true?
Outcome: Unresolved
The record was checked, and the evidence available by the cutoff does not settle it.
- Evidence as of
- Deadline state when assessed
- No deadline stated
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The claim is conditional. If someone carefully builds the comprehension challenge Marcus described, no system available around April 2025 would come close to human performance. He gave no deadline. The contract scores the claim only after a test meeting his design conditions is published with human and April-2025-era system results. The evidence packet contains no such test as of 2026-10-28. The only evidence item is LVBench, a long-video benchmark paper available 2025-10-20 (e1). It reports human accuracy of 94.4% on average. Gemini-2.5-Pro scored 67.4% and Seed1.5-VL scored 64.0%. That gap points the same way as the claim. But LVBench uses multiple-choice questions. Per e1, it is not refreshed against contamination and lacks the other design features in his condition. So it does not meet the antecedent, and the forecast stays unresolved.
Limitations
- The evidence does not show that any test meeting the stated conditions exists: full shows, refreshed questions, careful scoring and secure first-time-viewer conditions.
- LVBench results do not show how systems would perform on open-ended explanation questions.
- The contract sets no numeric threshold for 'come close', so how a future result is read will depend on interpretation.
- The search may have missed a recent or obscure implementation of the challenge.
Evidence (1 item)
Context only: ICCV 2025 (CVF Open Access),
humans achieve a very high ac- curacy of 94.4% on average, setting a strong benchmark across all categories. In contrast, the overall performance of Gemini-2.5-Pro and Seed1.5-VL was considerably lower, at 67.4% and 64.0%, respectively.
Why it matters here: A long-video benchmark shows a large human-model gap, which points the same way as the claim. But it is multiple-choice, is not refreshed against contamination, and lacks the other design features he set as the condition, so it does not trigger the condition.
- Publisher
- ICCV 2025 (CVF Open Access)
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
What would settle it
Publication of a benchmark paper or official results for a test that meets the stated design conditions. It would need to report scores for systems available around April 2025 alongside scores for human first-time viewers.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, AI has (sort of) passed the Turing Test; here’s why that hardly matters
But I am willing to wager that if someone put in that kind of effort and carefully built out the test, no current system would come close to human performance.
- Where
- paragraph 34
- Published
- 2025-04-02
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-ai-has-sort-of-passed-the-turing-cef6a3-a3: 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-has-sort-of-passed-the-turing-cef6a3-a3.