“We have not one, but many, serious, unsolved problems at the core of generative AI — ranging from their tendency to confabulate (hallucinate) false information, to their inability to reliably interface with external tools like Wolfram Alpha, to the instability from month to month (which makes them poor candidates for engineering use in larger systems).”
Our reading, the sentence that was checked:
Generative AI has many serious unsolved core problems, including hallucination, unreliable tool use, and month-to-month instability.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, What if Generative AI turned out to be a Dud?, paragraph 18.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-what-if-generative-ai-turned-out-30855d-a2
We recorded 23 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 2023-08-13. 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
- many, serious, unsolved problems
- confabulate (hallucinate) false information
- inability to reliably interface with external tools like Wolfram Alpha
- instability from month to month
Our interpretation
- The plain reading is that important reliability problems persisted in contemporary general-purpose language models, not that every model failed every task.
- The present-tense observation concerns systems available before publication. With no common reporting cycle, the observation window is calendar 2023 through August 13; this period convention is an evaluator choice.
- Reliability requires more than successful demonstrations. Repeated factual errors, failed tool interactions, and material changes between model versions qualify.
- The engineering parenthesis is treated as an implication of those problems, not a claim that no useful integration was possible.
Counts as supported if: Contemporary documentation and evaluations establish persistent factual errors, material tool-use failures, and behavior changes between monthly versions of leading systems.
Counts as contradicted if: Representative contemporary evaluations establish that the listed problems had been reliably resolved. Successful individual demonstrations alone do not establish this.
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
On March 14, 2023, OpenAI acknowledged that GPT-4 still hallucinated facts and made reasoning errors [e1]. The April 14 API-Bank paper reported a 40% success rate for GPT-3.5-Turbo on tasks requiring API retrieval, alongside other tool-use failures [e2]. The study revised August 1 found substantial task-dependent changes between March and June 2023 versions of GPT-3.5 and GPT-4, including improvements and regressions [e3]. These findings support the observation that the listed reliability problems remained unresolved in the systems studied [e1, e2, e3]. OpenAI's July 20 documentation offered stable pinned versions, limiting the engineering implications of changes across releases [e4].
Limitations
- API-Bank does not directly evaluate Wolfram Alpha or every model available in August 2023.
- The March–June comparison does not establish changes within an unchanged, pinned model or deterioration on every task.
- The evidence does not establish that every generative model or engineering integration was unreliable or unsuitable.
Evidence (4 items)
Supports the claim: OpenAI,
it still is not fully reliable
Why it matters here: OpenAI explicitly acknowledged continuing factual errors in its released model. The same section reports improvements, so persistence should not be confused with lack of progress.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: Minghao Li and colleagues, arXiv,
a 40% success rate
Why it matters here: The benchmark documents substantial failures in retrieving and using external APIs. This supports the general tool-reliability concern, but is not a Wolfram Alpha evaluation or a test of every August model.
- Publisher
- Minghao Li and colleagues, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: Lingjiao Chen, Matei Zaharia and James Zou, arXiv,
can vary greatly over time
Why it matters here: The study found task-dependent changes between March and June versions, including improvements and regressions. It establishes version-dependent behavior, not universal deterioration or drift within an unchanged model.
- Publisher
- Lingjiao Chen, Matei Zaharia and James Zou, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: OpenAI,
pin the model version
Why it matters here: OpenAI offered stable pinned versions while acknowledging regressions across upgrades. This limits the implication that every engineering integration necessarily faced uncontrolled monthly changes.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, What if Generative AI turned out to be a Dud?
We have not one, but many, serious, unsolved problems at the core of generative AI — ranging from their tendency to confabulate (hallucinate) false information, to their inability to reliably interface with external tools like Wolfram Alpha, to the instability from month to month (which makes them poor candidates for engineering use in larger systems).
- Where
- paragraph 18
- Published
- 2023-08-13
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-what-if-generative-ai-turned-out-30855d-a2: 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-what-if-generative-ai-turned-out-30855d-a2.