“Large language models lack mechanisms for verifying truth; we need to find new ways to integrate them with the tools of classical AI, such as databases, webs of knowledge, and reasoning.”
Our reading, the sentence that was checked:
LLMs lack mechanisms for verifying truth.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, AI's Jurassic Park moment, paragraph 24.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-ais-jurassic-park-moment-811c3a-a2
We recorded 15 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 2022-12-10. 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
- Large language models lack mechanisms for verifying truth
- integrate them with the tools of classical AI
- databases, webs of knowledge, and reasoning
Our interpretation
- In context, this concerns standalone generative language models before adding external checking tools.
- Verifying truth means checking a factual assertion against independent evidence or an explicit verification procedure. Learned factual recall and plausible generation do not alone constitute verification.
- The claim does not require a verifier to be infallible.
- The recommendation to develop integrations is not separately graded.
- For this ongoing technical state, I use the latest pre-post research year: January 1 through December 10, 2022.
Counts as supported if: Contemporary technical descriptions show that default standalone generation produces claims without an independent factual-checking procedure, with grounding or verification requiring additional mechanisms.
Counts as contradicted if: Contemporary technical descriptions demonstrate an independent factual-verification procedure within the standalone models covered by this reading.
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 the contextual reading about standalone generation, contemporary technical work supports the claim: default models generated ungrounded assertions and left verification to users. The qualification matters. LaMDA already added external tools to research and revise claims, and GopherCite supplied supporting evidence. The evidence therefore supports the architectural limitation, not a broader assertion that no LLM-based checking systems existed.
Limitations
- The result depends on distinguishing standalone models from augmented systems.
- The cited papers do not establish the architecture of every model available in 2022.
- Missing independent verification does not mean every generated statement is false or that models cannot perform useful factual tasks.
Evidence (2 items)
Supports the claim: DeepMind researchers, via arXiv,
LMs generate ungrounded claims
Why it matters here: The paper explicitly describes default generation as ungrounded and introduces additional training and evidence provision to address it. This supports the contextual reading about standalone models, rather than an assertion that no augmented system existed.
- Publisher
- DeepMind researchers, via arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Google researchers, via arXiv,
an information retrieval system
Why it matters here: LaMDA already used external tools to research and revise generated claims. This limits any claim that all contemporary LLM systems lacked checking mechanisms, but describes the augmentation distinguished in the contract.
- Publisher
- Google researchers, via 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, AI's Jurassic Park moment
Large language models lack mechanisms for verifying truth; we need to find new ways to integrate them with the tools of classical AI, such as databases, webs of knowledge, and reasoning.
- Where
- paragraph 24
- Published
- 2022-12-10
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-ais-jurassic-park-moment-811c3a-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.