“When GPT sounds plausible, it is because every paraphrased bit that it pastes together is grounded in something that actual humans said, and there is often some vague (but often irrelevant) relationship between..”
Our reading, the sentence that was checked:
GPT sounds plausible because each piece it combines comes from something humans actually said, and the pieces are often loosely related.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, How come GPT can seem so brilliant one minute and so breathtakingly dumb the next?, paragraph 45.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-how-come-gpt-can-seem-so-brilliant-57cc2b-a1
We recorded 9 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 2022-12-01. 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
- When GPT sounds plausible
- because every paraphrased bit
- is grounded in something that actual humans said
- often some vague (but often irrelevant) relationship
Our interpretation
- Grounded means derived from human utterances, not necessarily factually correct.
- The claim asserts more than training on human text: every constituent of plausible output has the stated provenance, and its recombination explains plausibility.
- Paraphrasing allows changed wording; absence of an exact textual match is therefore not sufficient contradiction.
- The scope is the GPT generation discussed in December 2022.
Counts as supported if: Provenance analyses and interventions establish that the relevant constituents derive from human utterances and that their learned relationships account for output plausibility.
Counts as contradicted if: A well-defined plausible constituent is shown not to derive from human utterances under a defensible paraphrase criterion, or controlled interventions disprove the proposed explanation of plausibility.
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
Learning from human text is consistent with the claim, but it does not establish that every plausible fragment derives from a particular human utterance. The provenance study found novelty in GPT-2, yet novelty alone cannot refute a claim that allows paraphrase and recombination. The retrieved evidence does not settle the universal provenance claim or its causal explanation of plausibility.
Limitations
- The meanings of bit, paraphrase and grounded materially affect the test.
- The directly relevant provenance study examines an earlier GPT model.
- Neither exact copying nor novel wording alone settles the stated mechanism.
Evidence (2 items)
Context only: McCoy and colleagues / arXiv,
GPT-2 generates some novel words.
Why it matters here: The provenance study finds novelty in an earlier GPT model. Novel wording complicates a literal copying account, but it does not refute derivation through paraphrase, analogy or recombination of human text.
- Publisher
- McCoy and colleagues / arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: OpenAI researchers / arXiv,
using supervised learning.
Why it matters here: The section describes training GPT-3 on human demonstrations. This establishes a route for human language to shape outputs, but does not trace every output constituent or isolate why a particular answer sounds plausible.
- Where
- Section 3.1, Step 1
- Publisher
- OpenAI researchers / arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A provenance study of the relevant GPT generation with defined constituent and paraphrase criteria, supported by interventions testing whether identified source material causes the output's plausibility.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, How come GPT can seem so brilliant one minute and so breathtakingly dumb the next?
When GPT sounds plausible, it is because every paraphrased bit that it pastes together is grounded in something that actual humans said, and there is often some vague (but often irrelevant) relationship between..
- Where
- paragraph 45
- Published
- 2022-12-01
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-how-come-gpt-can-seem-so-brilliant-57cc2b-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-how-come-gpt-can-seem-so-brilliant-57cc2b-a1.