“Our new systems work better, but they are still superficial, fine for lab demos but often at a loss in the real world.”
Our reading, the sentence that was checked:
Current AI systems work better than older ones but are still superficial: fine for lab demos, often at a loss in the real world.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Happy Groundhog Day, The AI Edition, paragraph 21.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-happy-groundhog-day-the-ai-edition-33b9a0-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 2023-02-03. 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
- Our new systems work better
- still superficial
- fine for lab demos
- often at a loss in the real world
Our interpretation
- The plain reading is improved apparent capability alongside recurring material failures in ordinary deployment. Superficial is read as unreliable performance, not a separately testable claim about consciousness or understanding.
- The surrounding discussion primarily concerns contemporary generative AI, but does not identify a fixed model population.
- Often means recurring failures, not universal failure or failure on most tasks.
- Use January 2023, the latest complete month before the post, for the ongoing deployment claim. Earlier evaluations provide context for the conceded improvement.
- A full assessment requires enough evidence to connect demonstrated capability with deployment performance beyond one selected product.
Counts as supported if: Contemporary deployment records and comparative evaluations show improved capabilities but recurring material failures across ordinary uses of the systems under discussion.
Counts as contradicted if: Broad, comparable deployment evidence shows dependable performance on those ordinary uses, with failures sufficiently unusual to conflict with the claim of recurring practical unreliability.
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
CNET supplies a concrete example of recurring deployment errors, while the earlier model evaluation documents improvement. These findings fit parts of Marcus's description. They do not establish how broadly the deployment problem held or a general gap between demonstrations and ordinary use. Treating one deployment as sufficient would make the verdict depend on an unstated narrowing of his claim.
Limitations
- Documented failures are not a representative failure rate.
- Useful performance would not alone contradict a claim that failures occur often.
- The assessment concerns systems around the post date, not later models.
Evidence (2 items)
Supports the claim: CNET,
we’ve corrected these stories
Why it matters here: CNET's January audit found multiple erroneous AI-assisted articles and prompted a publication pause. This supports recurring practical failures in one deployment, but the account also implicates editorial processes and does not establish performance across AI systems.
- Publisher
- CNET
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: OpenAI,
more reliable and easier to control
Why it matters here: The authors report comparative improvements in instruction following and factual reliability. This bears on Marcus's conceded progress, but is an earlier evaluation rather than evidence of January deployment failure rates.
- Publisher
- OpenAI
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
Contemporaneous deployment audits across several relevant systems and tasks, paired with comparable demonstrations or evaluations and explicit records of successes and material failures.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Happy Groundhog Day, The AI Edition
Our new systems work better, but they are still superficial, fine for lab demos but often at a loss in the real world.
- Where
- paragraph 21
- Published
- 2023-02-03
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-happy-groundhog-day-the-ai-edition-33b9a0-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-happy-groundhog-day-the-ai-edition-33b9a0-a1.