“New concerns are emerging almost literally day,.”
Our reading, the sentence that was checked:
New concerns about LLM harms are emerging almost daily.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, The long shadow of GPT, paragraph 17.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-the-long-shadow-of-gpt-a35687-a2
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 2023-03-10. 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
- New concerns
- are emerging
- almost literally day
- just in the last few days
Our interpretation
- Read the apparent typo as 'almost daily,' consistent with the surrounding examples.
- The plain reading is that distinct concerns about LLM harms were becoming publicly apparent on most days. A new concrete harmful application can count; an entirely new technical mechanism is not required.
- Use March 3–10, 2023, the week leading into publication, to operationalize the ongoing observation and 'last few days.' Marcus supplied no exact reporting window.
- Count distinct concerns, not repeated coverage of the same concern. Publication dates establish when concerns were publicly reported, not necessarily when the underlying behavior first occurred.
Counts as supported if: A dated chronology shows distinct concerns or newly documented harmful applications appearing on most days during the target period.
Counts as contradicted if: A sufficiently comprehensive chronology shows only occasional new concerns, with the apparent daily flow consisting mainly of repetitions. Failure to find reports in an incomplete search is insufficient.
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
VICE documented harmful sexual content on March 6, 2023, and the James Martin Center documented fabricated nuclear-incident stories on March 7, 2023. [e1, e2] The indirect-prompt-injection research covered on March 9 had already been published on February 23, so that coverage does not establish a new concern emerging during March 3–10. [e3, e4] These records establish particular reports, but do not establish either distinct concerns emerging on most days of March 3–10 or a materially slower pace. [e1–e4]
Limitations
- The packet lacks a sufficiently comprehensive, deduplicated chronology.
- Publication dates do not establish when a concern first emerged.
- The March 3–10 window is an evaluator interpretation; the statement specifies neither an exact period nor a numerical meaning for 'almost daily'.
Evidence (4 items)
Supports the claim: VICE,
acts of bestiality.
Why it matters here: A dated concern; frequency unproven.
- Publisher
- VICE
- Kind of source
- Demonstration
- Describes events of
- Public from
Supports the claim: James Martin Center for Nonproliferation Studies,
AI-assisted disinformation.
Why it matters here: This original report discusses tests involving fabricated nuclear incidents and technical information. Its publication establishes a concern being articulated that day, but not its first discovery or an almost-daily rate.
- Publisher
- James Martin Center for Nonproliferation Studies
- Kind of source
- Demonstration
- Describes events of
- Public from
Context only: arXiv,
indirectly perform such PI attacks.
Why it matters here: The original research underlying Marcus's indirect-prompt-injection example was already public in February. This limits treating its March circulation as a newly discovered concern, but does not disprove the broader cadence claim.
- Publisher
- arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Tech Xplore / Science X,
"More than you've asked for."
Why it matters here: This March report covers the February research in e3 and explicitly describes it as reported the previous month. It illustrates why counting publication dates alone would overstate the rate of distinct new concerns.
- Publisher
- Tech Xplore / Science X
- Kind of source
- News report
- Describes events of
- Public from
What would settle it
A sufficiently comprehensive dated chronology for March 3–10, 2023, checked against earlier reports and deduplicated by concern, showing distinct concerns or newly documented harmful applications on most days, or showing only occasional new concerns.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, The long shadow of GPT
New concerns are emerging almost literally day,.
- Where
- paragraph 17
- Published
- 2023-03-10
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-the-long-shadow-of-gpt-a35687-a2: 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.