“As I noted in 2001, the lack of distinct, accessible, reliable database-style records leads to hallucinations.”
Our reading, the sentence that was checked:
The lack of distinct, accessible, reliable database-style records leads to hallucinations, as Marcus noted in 2001.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Why I don’t share Sam Altman’s confidence that AGI is basically a solved problem, paragraph 18.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-sam-altman-thinks-that-agi-is-basically-0a0fa4-a1
We recorded 10 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 2025-01-06. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Causal claim
- 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
- As I noted in 2001
- the lack of distinct, accessible, reliable database-style records
- leads to hallucinations
Our interpretation
- The ordinary reading is that inadequate access to separately represented, dependable facts contributes to hallucinations. Marcus does not say it is their only cause or that adding a database eliminates them.
- Retrieval from an external factual collection is a practical, imperfect test of what Marcus calls database-style records.
- The 2001 attribution is a separate factual commitment. It requires an identifiable contemporary passage expressing the relevant causal idea, although it need not use today's terminology.
- This is a general causal claim, so no observation reporting period or forecast deadline is imposed.
Counts as supported if: Comparable-system experiments show that supplying dependable, accessible factual records reduces hallucinations, and a dated 2001 publication establishes that Marcus articulated the relevant causal argument.
Counts as contradicted if: Strong comparative evidence rejects the proposed contributing mechanism within the claimed scope, or documentary evidence establishes that the claimed historical attribution is false. Continued hallucinations after adding retrieval alone would not contradict a contributing-cause claim.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
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
Retrieval research supports the narrower proposition that accessible external knowledge can improve factual generation. It does not isolate Marcus's proposed mechanism completely. His 2001 book contains a related argument about separate records for individuals, but the passages located do not verify the full historical attribution. The combined claim therefore remains unresolved. Retrieval failures are a limitation, not a refutation of a contributing-cause claim.
Limitations
- The historical search was not an exhaustive examination of every 2001 publication.
- Failure to find the attributed passage does not establish that it does not exist.
- The 2001 publication day comes from indexed publisher metadata; the readable scan identifies the publication year.
- The ICML availability date uses the final day of the proceedings' stated conference interval.
Evidence (3 items)
Supports the claim: Lewis et al.; arXiv, accepted at NeurIPS 2020,
factual language
Why it matters here: The comparison supports a contributing role for accessible external knowledge in reducing factual errors. It tests a retrieval-and-training system, not an isolated intervention on all three properties Marcus names; the availability date is that of the displayed revised version.
- Publisher
- Lewis et al.; arXiv, accepted at NeurIPS 2020
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Context only: MIT Press; scan hosted by University of Diyala,
a particular record
Why it matters here: Marcus discusses systems that represent individuals separately and proposes a database-like approach in the surrounding passage. This establishes a related historical argument, not independent evidence of its correctness or verification of the complete retrospective claim about hallucinations.
- Publisher
- MIT Press; scan hosted by University of Diyala
- Kind of source
- Secondary summary
- Describes events of
- Public from
Context only: Proceedings of Machine Learning Research,
exacerbate hallucination
Why it matters here: The study identifies circumstances where retrieval worsens hallucination and tests a method for directing attention to useful information. This limits a simple database-access explanation but does not disprove the modest contributing-cause reading.
- Publisher
- Proceedings of Machine Learning Research
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
What would settle it
An identifiable 2001 passage expressing the causal argument, together with comparable-system experiments that establish how dependable, separately accessible records affect hallucination across relevant tasks.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Why I don’t share Sam Altman’s confidence that AGI is basically a solved problem
As I noted in 2001, the lack of distinct, accessible, reliable database-style records leads to hallucinations.
- Where
- paragraph 18
- Published
- 2025-01-06
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-sam-altman-thinks-that-agi-is-basically-0a0fa4-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-sam-altman-thinks-that-agi-is-basically-0a0fa4-a1.