“Scaling won’t solve this.”
Our reading, the sentence that was checked:
Scaling LLMs will not solve their inability to reason about truth and detect misinformation.
Assessment: Unresolved No deadline stated on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Bravo @ElonMusk for wanting Twitter to become the world’s “most accurate source of information”, paragraph 16.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-bravo-elonmusk-for-wanting-twitter-67f63a-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 2022-11-09. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Forecast
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No deadline stated
- Deadline state on 2026-10-29
- No deadline stated (the release’s reference date)
From the speaker
- Scaling won’t solve this
- No matter how much data you can collect
- they suck at detecting it
- reasoning about truth
Our interpretation
- In context, this predicts that increasing LLM scale alone will not produce reliable factual reasoning and misinformation detection.
- Scale means increased model size, training data or training compute within a comparable modeling approach.
- A solution means broadly reliable practical performance, not perfect knowledge or a philosophical demonstration of understanding.
- Improvement is weaker than solving the problem. Gains from retrieval, new supervision, tools or changed training methods do not by themselves identify the effect of scale.
- No deadline is supplied, so the forecast has an open horizon.
Counts as supported if: A compelling general limitation showing that scale alone cannot deliver the stated capability would support the forecast. A finite sequence of failures provides supporting context but cannot settle this indefinite prediction.
Counts as contradicted if: A well-controlled, independently replicated evaluation shows that increasing scale alone produces broadly reliable factual reasoning and misinformation detection on diverse, previously unseen claims.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Did the forecast come true?
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 deadline stated
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
Marcus gives no deadline. Later studies show that larger models can become more truthful and better at fact-checking, which weighs against treating scaling as useless. Other evaluations document substantial remaining errors. Neither pattern settles his stronger claim that scaling will not solve the problem. The searched evidence contains no decisive, independently replicated demonstration of broad reliability caused by scale alone. Continued shortcomings cannot establish an unlimited prediction.
Limitations
- Truthful answer generation and misinformation detection are related but distinct tasks.
- The studies test finite model families, scales and benchmarks.
- Experimental dates are generally reported only at study-year precision.
- Reliability and scaling lack numerical definitions in the original statement.
Evidence (4 items)
Context only: arXiv; Blanca Calvo Figueras and colleagues,
Larger models are more truthful.
Why it matters here: The study finds better truthfulness in larger same-family models. It also identifies post-training alignment as a possible contributor and limits generalization beyond TruthfulQA, so it does not demonstrate a scale-only solution.
- Publisher
- arXiv; Blanca Calvo Figueras and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Ihsan A. Qazi and colleagues,
higher Precision than smaller models
Why it matters here: The evaluation supplies material evidence of better fact-checking performance in larger models across multilingual claims. Differences in model families, prompting and abstention prevent attribution to scale alone or a conclusion that the problem is solved.
- Publisher
- arXiv; Ihsan A. Qazi and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Matthew R. DeVerna and colleagues,
Standard models perform poorly
Why it matters here: The study reports continuing political fact-checking weaknesses across 15 models and stronger results with curated context. These finite results are consistent with Marcus's concern but cannot establish that future scaling will never solve it.
- Where
- Abstract
- Publisher
- arXiv; Matthew R. DeVerna and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Association for Computational Linguistics; author version hosted on arXiv,
semantically unrelated to actual errors
Why it matters here: ReFACT reports persistent error-localization failures across tested scales from 1B to 70B parameters. This is evidence of a remaining limitation in scientific factuality, not a general impossibility result for all future scales.
- Where
- Abstract, version 3
- Publisher
- Association for Computational Linguistics; author version hosted on arXiv
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
What would settle it
A dated, independently replicated comparison showing that scale alone produces broadly reliable factual reasoning and misinformation detection on diverse unseen claims would contradict the forecast. Continued failures would leave its open horizon unresolved.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Bravo @ElonMusk for wanting Twitter to become the world’s “most accurate source of information”
Scaling won’t solve this.
- Where
- paragraph 16
- Published
- 2022-11-09
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-bravo-elonmusk-for-wanting-twitter-67f63a-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-bravo-elonmusk-for-wanting-twitter-67f63a-a1.