“Although there are some practical use cases, such as coding, it is an inherently unreliable technology.”
Our reading, the sentence that was checked:
Generative AI has some practical uses such as coding but is an inherently unreliable technology.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Turns out Generative AI was a scam, paragraph 17.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-turns-out-generative-ai-was-a-scam-1080d5-a2
We recorded 8 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 2026-02-23. 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)
Conditions
- despite practical use cases such as coding (qualifier, stated by the speaker)
From the speaker
- some practical use cases
- such as coding
- an inherently unreliable technology
Our interpretation
- The ordinary reading is that useful applications coexist with recurrent, material errors that make unchecked reliance unsafe. It does not mean every output is wrong.
- I use 2025, the latest complete calendar year before the post, as the reporting period for this ongoing state.
- I assess systems then available. A claim that no future generative system could become reliable would require different evidence.
Counts as supported if: Evidence of practical uses, including coding, alongside evaluations showing recurrent material errors across commonly used systems.
Counts as contradicted if: Representative evaluations show dependable performance across ordinary uses without material verification needs, or credible evidence defeats the assertion that any practical coding uses exist.
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
The 2025 evidence supports reported coding utility alongside recurrent material errors. In Stack Overflow's 2025 survey, approximately 70% of agent users reported time savings on development tasks; these were perceived benefits, not measured gains [e1]. In news responses generated in late May and early June 2025, the EBU/BBC evaluation found significant issues in 45% of responses across four assistants [e2]. Its comparable BBC subset improved from 51% to 37%, showing progress alongside persistent problems [e2]. METR's experiment using February–June 2025 tools found that experienced developers took 19% longer on familiar repositories despite believing AI saved time, limiting broader claims of coding productivity gains [e3].
Limitations
- Coding utility is supported here by self-reports; the experiment does not establish gains across coding tasks [e1, e3].
- The multi-assistant reliability evaluation concerns news responses, not every ordinary application [e2].
- These 2025 findings do not establish that unreliability is unavoidable in all applications or future systems [e1, e2, e3].
Evidence (3 items)
Supports the claim: Stack Overflow,
productivity.
Why it matters here: Developer respondents report useful applications in their work, supporting the modest assertion that practical coding uses exist. Self-selection and subjective productivity estimates limit the strength of this evidence.
- Publisher
- Stack Overflow
- Kind of source
- Independent measurement
- Describes events of
- Public from
Supports the claim: European Broadcasting Union and BBC,
(31%).
Why it matters here: Responses generated in late May and early June across four assistants and multiple languages contained significant problems in 45% of cases. The report also records improvement in the comparable BBC subset, so persistent problems do not imply no progress.
- Publisher
- European Broadcasting Union and BBC
- Kind of source
- Independent measurement
- Describes events of
- Public from
Context only: METR researchers, arXiv,
slowed developers down.
Why it matters here: In this experiment, experienced developers working on familiar repositories became slower despite believing AI helped. This limits reliance on perceived productivity and broad claims of coding gains, but does not negate the existence of some useful applications.
- Where
- Abstract
- Publisher
- METR researchers, arXiv
- 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, Turns out Generative AI was a scam
Although there are some practical use cases, such as coding, it is an inherently unreliable technology.
- Where
- paragraph 17
- Published
- 2026-02-23
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-turns-out-generative-ai-was-a-scam-1080d5-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-turns-out-generative-ai-was-a-scam-1080d5-a2.