“In “Deep learning is hitting a wall”, I warned that problems with reasoning and hallucination wouldn’t go away simply as a function of scaling, and they haven’t.”
Our reading, the sentence that was checked:
Problems with reasoning and hallucination in deep learning systems have not gone away as a function of scaling.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, GPT 4.5 is no GPT-5, paragraph 7.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-gpt-45-is-no-gpt-5-b2ee0b-a1
We recorded 14 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 2025-02-26. 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
- problems with reasoning and hallucination
- wouldn’t go away simply as a function of scaling
- they haven’t
Our interpretation
- Assess the present-tense factual assertion that reasoning and hallucination problems remained in leading systems by publication.
- Use January 1 through February 26, 2025 as the latest early-2025 reporting window for the ongoing state described.
- Persistence of errors is distinct from absence of improvement.
- This observation does not establish that further scaling could never solve these problems or identify their causal mechanism.
Counts as supported if: Evaluations describing leading models in the target period document both hallucinations and reasoning failures.
Counts as contradicted if: Credible broad evaluations show that leading scaled systems had reliably eliminated both classes of error by publication.
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
Prepublication evaluations document continuing hallucinations and failures on difficult knowledge and reasoning questions. That supports the observation that these problems had not gone away. The sources also show improvements, and they do not establish that scaling was ineffective or could never remove the remaining problems.
Limitations
- Humanity's Last Exam combines knowledge and reasoning; not every wrong answer identifies a reasoning failure.
- The evidence establishes persistence, not a causal explanation or permanent ceiling.
- The o3-mini system-card file is a February revision of a January launch report.
Evidence (3 items)
Supports the claim: OpenAI,
on par or better
Why it matters here: The January launch system card, in its February file version, reports a 14.8% PersonQA hallucination rate for o3-mini. Its comparison models also have nonzero rates, despite improvements.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: Scale AI and Center for AI Safety,
expert closed-ended questions
Why it matters here: The benchmark report describes continuing failures on expert knowledge and reasoning questions across frontier models, including GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Pro, and o1. It also acknowledges improvements over earlier models.
- Publisher
- Scale AI and Center for AI Safety
- Kind of source
- Benchmark result
- Describes events of
- Public from
Context only: OpenAI,
performs better
Why it matters here: OpenAI reported improved research-mathematics performance. This qualifies any suggestion of stagnation, while remaining consistent with the narrower claim that reasoning problems persisted.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, GPT 4.5 is no GPT-5
In “Deep learning is hitting a wall”, I warned that problems with reasoning and hallucination wouldn’t go away simply as a function of scaling, and they haven’t.
- Where
- paragraph 7
- Published
- 2025-02-26
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-gpt-45-is-no-gpt-5-b2ee0b-a1: 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.