“And no other form of deep learning has yet solved the issues of reasoning, hallucinations, stupid errors, and factuality, either.”
Our reading, the sentence that was checked:
No form of deep learning has yet solved reasoning, hallucinations, stupid errors, or factuality.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Things are so desperate at OpenAI that Sam Altman is starting to sound like Gary Marcus, paragraph 8.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-things-are-so-desperate-at-openai-670510-a1
We recorded 2 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 2025-08-18. 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
- no other form of deep learning
- has yet solved
- reasoning, hallucinations, stupid errors, and factuality
Our interpretation
- Read as a claim that the named general reliability problems remained unresolved, including in approaches other than LLMs.
- Solved means reliable general performance across varied tasks. It does not require literal infallibility, and a narrow task-specific success does not settle a general problem.
- Use August 1–18, 2025 as the latest reporting window for the model landscape discussed in the post.
- Evidence that particular systems fail supports only those cases; it cannot automatically establish the claim about every deep-learning approach.
Counts as supported if: Broad evaluations or a sufficiently comprehensive review establish that leading LLM and alternative deep-learning approaches still lack general, reliable solutions to the named problems.
Counts as contradicted if: A documented deep-learning approach demonstrates a robust general solution to one of the named problems, with independent evaluation beyond a narrow or specially constrained setting.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
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
The primary evaluations document genuine limitations in both GPT-5 and V-JEPA 2. They are consistent with Marcus's concern, but they concern selected systems and earlier testing periods. The search found no qualifying demonstration of a general solution, yet that absence alone cannot establish that every form of deep learning remained unable to solve the named problems.
Limitations
- This is a broad negative claim spanning different technical approaches.
- The available examples support narrower claims about particular systems.
- General reliability has no threshold supplied by the speaker.
Evidence (2 items)
Context only: METR,
The model fails in 5/6 runs on this task.
Why it matters here: METR describes GPT-5 failing a software task after neglecting tests or their results. This is a concrete reliability failure, but July testing of one model does not establish the full August claim.
- Publisher
- METR
- Kind of source
- Independent measurement
- Describes events of
- Public from
Context only: arXiv,
V-JEPA 2 and the EK100 benchmark have several limitations.
Why it matters here: The authors document incorrect action predictions and limited generalization in a video-based approach. This broadens the evidence beyond LLMs, but does not cover all approaches or establish their August state.
- Publisher
- arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A contemporaneous, comprehensive evaluation of the leading relevant approaches, or an independently verified general solution that constitutes a counterexample.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Things are so desperate at OpenAI that Sam Altman is starting to sound like Gary Marcus
And no other form of deep learning has yet solved the issues of reasoning, hallucinations, stupid errors, and factuality, either.
- Where
- paragraph 8
- Published
- 2025-08-18
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-things-are-so-desperate-at-openai-670510-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-things-are-so-desperate-at-openai-670510-a1.