“The truth, though, is that GPT often fails in problems involving theory of mind.”
Our reading, the sentence that was checked:
GPT often fails at problems involving theory of mind.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, How Not to Test GPT-3, paragraph 13.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-how-not-to-test-gpt-3-a617b1-a1
We recorded 12 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-29. The source may have been edited after 2023-02-17. 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
- GPT often fails
- problems involving theory of mind
Our interpretation
- The context concerns the GPT-3 family, including systems then described as GPT-3.5 and ChatGPT, rather than every subsequent GPT model.
- The plain reading of 'often' is recurring, substantive failures across theory-of-mind problems or ordinary variations. It does not necessarily mean failure on most questions.
- The target period is February 2023, as of the February 17 post. This is the latest contemporary evaluation period, rather than a formal reporting quarter.
Counts as supported if: Contemporary independent testing documents recurring failures across several theory-of-mind tasks or meaningful variations.
Counts as contradicted if: Comparable contemporary testing establishes reliable performance across those tasks and variations, and shows that the reported failures were mistaken or invalid.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
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
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
Contemporary independent experiments document repeated failures across small variations of two false-belief tasks. That supports the ordinary reading that failures were recurring. Kosinski's substantial successes remain material: the evidence supports uneven performance, not universal inability or failure on most tasks.
Limitations
- The result depends on reading 'often' as recurring failures rather than a specified population failure rate.
- The decisive study is a preprint using deliberately selected variations.
- This assessment concerns February 2023 systems.
Evidence (2 items)
Supports the claim: arXiv; Tomer D. Ullman,
All variations cause the LLM to fail.
Why it matters here: Ullman reports failures across four unexpected-transfer variations and four unexpected-contents variations in the contemporary GPT-3.5 system. These establish recurring failures, although the selected examples do not estimate their prevalence across all possible tasks.
- Publisher
- arXiv; Tomer D. Ullman
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Michal Kosinski,
solved 93% of ToM tasks
Why it matters here: This contemporary study reports substantial success and limits any suggestion of general inability. Its different task battery does not negate the recurring failures in Ullman's variations.
- Where
- Abstract
- Publisher
- arXiv; Michal Kosinski
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, How Not to Test GPT-3
The truth, though, is that GPT often fails in problems involving theory of mind.
- Where
- paragraph 13
- Published
- 2023-02-17
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-how-not-to-test-gpt-3-a617b1-a1: 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.