“Recent “reasoning” models borrow even more from classic symbolic approaches, such as techniques like search and conditionally iterating through multiple solutions, and aggregating the results, techniques that traditional networks had long eschewed, in favor of a single feed-forward pass through a neural net.”
Our reading, the sentence that was checked:
Recent reasoning models borrow symbolic techniques such as search, iteration over multiple solutions and aggregation.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, How o3 and Grok 4 Accidentally Vindicated Neurosymbolic AI, paragraph 84.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-how-o3-and-grok-4-accidentally-vindicated-aadc40-a2
We recorded 36 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 2025-07-13. 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
- Recent “reasoning” models
- borrow even more from classic symbolic approaches
- search
- conditionally iterating through multiple solutions
- aggregating the results
- traditional networks had long eschewed
- a single feed-forward pass through a neural net
Our interpretation
- Read 'borrow' operationally: reasoning systems use exploration of alternatives, conditional retries and aggregation resembling techniques used in classical symbolic systems.
- This reading does not require documented intellectual influence or assume that every operation is implemented by an explicitly programmed symbolic search algorithm.
- Use the recent reasoning-model generation disclosed during 2024 and 2025 through July 13, 2025.
- Treat the comparison with traditional networks as a contrast with a simple feed-forward baseline, not a universal claim about every earlier neural architecture.
Counts as supported if: Primary technical descriptions establish alternative-solution exploration, conditional revision and aggregation in recent reasoning systems.
Counts as contradicted if: Technical records establish that the attributed operations are absent from the relevant systems. Evidence of learned rather than programmed operations contradicts only the stronger interpretation that explicit symbolic machinery is required.
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
OpenAI's September 12, 2024 disclosure describes o1 learning to change approach when one fails, plus AIME evaluations using consensus among 64 samples and reranking of 1,000 samples [e1]. xAI's July 9, 2025 disclosure describes Grok 4 Heavy considering multiple hypotheses in parallel [e2]. These accounts support 'borrow' in the functional sense of exploring alternatives, conditionally revising approaches and aggregating results [e1, e2]. DeepSeek's January 22, 2025 report describes reflection and alternative exploration as learned through reinforcement learning, rather than explicitly programmed [e3].
Limitations
- Developer descriptions do not establish historical influence from symbolic AI or an explicitly programmed symbolic search implementation [e1, e2, e3].
- OpenAI's aggregation results concern evaluation configurations and do not establish default chat behavior [e1].
- These disclosures do not substantiate a universal historical claim that earlier neural systems used only a single feed-forward pass [e1, e2, e3].
Evidence (3 items)
Supports the claim: OpenAI,
try a different approach
Why it matters here: OpenAI describes conditional strategy changes in o1 and reports AIME evaluations using consensus among 64 samples and reranking 1,000 samples. These document revision and aggregation, without establishing that aggregation was the default chat configuration.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: xAI,
multiple hypotheses at once
Why it matters here: The launch disclosure confirms parallel exploration of alternatives in Grok 4 Heavy before Marcus's post. It does not disclose a complete symbolic search implementation.
- Publisher
- xAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: arXiv; DeepSeek-AI,
not explicitly programmed
Why it matters here: The report describes reflection and exploration of alternatives as learned behaviors. This limits any inference that resembling classical search proves the model contains an explicitly programmed symbolic search procedure.
- Publisher
- arXiv; DeepSeek-AI
- 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, How o3 and Grok 4 Accidentally Vindicated Neurosymbolic AI
Recent “reasoning” models borrow even more from classic symbolic approaches, such as techniques like search and conditionally iterating through multiple solutions, and aggregating the results, techniques that traditional networks had long eschewed, in favor of a single feed-forward pass through a neural net.
- Where
- paragraph 84
- Published
- 2025-07-13
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-how-o3-and-grok-4-accidentally-vindicated-aadc40-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-how-o3-and-grok-4-accidentally-vindicated-aadc40-a2.