“Large language models are not databases. They are glommers-together-of-bits-that don’t always belong together.”
Our reading, the sentence that was checked:
LLMs are not databases; they combine bits of text that don't always belong together.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Is ChatGPT Really a “Code Red” for Google Search?, paragraph 47.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-is-chatgpt-really-a-code-red-for-e753d6-a2
We recorded 13 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 2022-12-30. 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 are not databases
- glommers-together-of-bits-that don’t always belong together
Our interpretation
- Read 'databases' using the post's explanation: systems that store, retrieve and update explicit records.
- Read the metaphor as saying that models generate text from learned relationships and sometimes produce incorrect combinations or associations. It need not mean literal copying and pasting of stored passages.
- Assess the late-2022 model generation discussed in the post, using November–December as the latest deployment and reporting period before publication.
- The claim does not supply a failure-rate threshold; documented instances suffice for 'don’t always'.
Counts as supported if: Technical descriptions distinguish predictive text generation from conventional record retrieval, and documented results show invented or incorrectly associated factual content.
Counts as contradicted if: Technical records establish that the relevant models operate as conventional record databases, or establish that the alleged factual combinations were accurate and provide no support for the stated failure mode.
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 BLOOM paper published November 9, 2022 describes autoregressive language modeling as iteratively predicting the probability of the next token [e1]. This supports the contextual distinction between predictive text generation and retrieval of explicit database records [e1]. The Galactica paper published November 16, 2022 reports that paper-title citation identifiers were more prone to hallucination errors at lower model scales [e2]. That finding supports the limited claim that generated text can contain incorrect associations [e2].
Limitations
- The cited passages do not establish a general error rate across models or tasks.
- The evidence supports the metaphor as a description of possible incorrect associations, not as a literal copying-and-pasting mechanism.
- The assessment uses 'database' to mean an explicit record system; it does not establish that models cannot encode knowledge or use external databases.
Evidence (2 items)
Supports the claim: BigScience collaboration / arXiv,
the next token
Why it matters here: The technical description identifies generation as successive probabilistic predictions. This supports the distinction from retrieving explicit records; the database comparison is the evaluator's inference from the described mechanism.
- Publisher
- BigScience collaboration / arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: Galactica research team / arXiv,
hallucination error
Why it matters here: The authors report hallucinated citation titles in their experiments, supporting the claim of incorrect generated associations. Section 2 also describes models as implicit knowledge bases, which limits any broader reading that models cannot store knowledge.
- Publisher
- Galactica research team / 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, Is ChatGPT Really a “Code Red” for Google Search?
Large language models are not databases. They are glommers-together-of-bits-that don’t always belong together.
- Where
- paragraph 47
- Published
- 2022-12-30
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-is-chatgpt-really-a-code-red-for-e753d6-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-is-chatgpt-really-a-code-red-for-e753d6-a2.