“Google makes a big deal about how the system is to do some degree interpretable, but there is no obvious way to integrate large language models with the kind of formal formal verification methods that are now routine in the design of microprocessors, USB drivers and large aircraft.”
Our reading, the sentence that was checked:
There is no obvious way to integrate large language models with the kind of formal verification methods that are now routine in designing microprocessors, USB drivers and large aircraft.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, What’s wrong with Google’s new robot project, paragraph 34.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-whats-wrong-with-googles-new-robot-965b92-a1
We recorded 20 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 2022-08-18. 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)
Conditions
- the kind of formal verification methods now routine in microprocessors, USB drivers and large aircraft (scope, stated by the speaker)
From the speaker
- no obvious way
- integrate large language models
- formal formal verification methods
- routine in the design of microprocessors, USB drivers and large aircraft
Our interpretation
- Read the central claim as the absence of an established method for verifying relevant LLM-controlled behavior to the assurance standard invoked by the engineering comparison.
- An LLM producing a formally checked mathematical artifact is not automatically verification of the LLM's behavior.
- A certificate for a restricted Transformer property is relevant counterevidence, but does not automatically establish comparable assurance for a household robot.
- Use the 2022 technical record through the post as the latest public reporting window; earlier work supplies context.
- The cited engineering fields specify the comparison, not an antecedent that triggers a forecast.
Counts as supported if: A sufficiently comprehensive contemporaneous technical record establishes that available methods could not provide the relevant assurance for large language models at practical scale.
Counts as contradicted if: A method established by August 18, 2022 integrates large language models with formal verification and provides the relevant behavioral guarantees at a scope comparable to the cited engineering uses.
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
Formal connections already existed: researchers certified restricted Transformer properties and used LLMs to produce formal mathematical specifications. Neither establishes assurance for an LLM-controlled household robot comparable to the engineering examples. Conversely, finding only limited methods does not prove that no suitable route was established. The evidence therefore does not settle the contextual claim.
Limitations
- A blanket claim that no formal integration existed would be too strong.
- The engineering comparison and the meaning of 'obvious' remain underspecified.
- Later advances cannot by themselves contradict an observation about August 2022.
Evidence (2 items)
Context only: arXiv; Zhouxing Shi and colleagues,
the first robustness verification algorithm for Transformers.
Why it matters here: This establishes an earlier connection between Transformer networks and formal verification. Its restricted robustness guarantees do not establish industrial assurance for generative LLM-controlled robots.
- Where
- Abstract
- Publisher
- arXiv; Zhouxing Shi and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Yuhuai Wu and colleagues,
formal specifications and proofs.
Why it matters here: The paper reports LLM-generated formal specifications used with theorem-proving work. It challenges a blanket reading that no integration existed, but does not verify the general behavior of the language model or a physical robot.
- Publisher
- arXiv; Yuhuai Wu and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A contemporaneous method with explicit, practically demonstrated behavioral guarantees at the relevant scale, or a comprehensive historical technical assessment establishing the limits of all relevant approaches.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, What’s wrong with Google’s new robot project
Google makes a big deal about how the system is to do some degree interpretable, but there is no obvious way to integrate large language models with the kind of formal formal verification methods that are now routine in the design of microprocessors, USB drivers and large aircraft.
- Where
- paragraph 34
- Published
- 2022-08-18
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-whats-wrong-with-googles-new-robot-965b92-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-whats-wrong-with-googles-new-robot-965b92-a1.