“See also my essay on Google’s Palm-SayCan, for what might happen if this tech were embedded in robots, which would be even more risky.”
Our reading, the sentence that was checked:
Embedding LLM technology in robots would be even more risky than using it in home assistants.
Assessment: Unresolved No deadline stated on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, How come smart assistants have virtually no ability to converse, despite all the spectacular progress with large language models?, paragraph 24.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-how-come-smart-assistants-have-virtually-34cc58-a1
We recorded 10 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 2022-11-22. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Conditional 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)
Conditions
- if LLM tech were embedded in robots (antecedent, stated by the speaker)
From the speaker
- if this tech were embedded in robots
- would be even more risky
- Google’s Palm-SayCan
Our interpretation
- The comparison is with the LLM-powered conversational home assistant described immediately before this sentence, including its ability to control home devices.
- The plain reading is that robotic embodiment increases the likelihood or severity of harmful failures. Marcus does not specify a numerical increase or a particular safety measure.
- Embedding includes an LLM directing robot actions through other software; it does not require the model to run onboard.
- A research demonstration satisfies the antecedent because Marcus does not require commercial deployment.
- A demonstration of robot hazards alone does not establish the comparative claim.
Counts as supported if: Credible comparative evidence shows greater likelihood or severity of harmful failures from LLM-controlled robots than from comparable LLM-powered home assistants, with differences in tasks, exposure and safeguards accounted for.
Counts as contradicted if: Credible comparative evidence establishes equal or lower risk for LLM-controlled robots across representative uses and relevant harms, with enough precision to exclude a meaningful increase. A single safe demonstration or statistically inconclusive comparison is insufficient.
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
The condition has happened: language-model technology has directed robots. Robot evaluations document unsafe actions, making Marcus's concern plausible. However, those findings do not establish that robots are riskier than the conversational home assistants he describes, which also control devices. Safeguard research further shows that system design materially changes unsafe-action rates. No located evidence meets the comparative early-resolution rule. With no stated deadline, the forecast remains open.
Limitations
- The evidence supports the existence of robotic hazards, not the claimed comparison.
- Risk and the relevant assistant baseline require evaluator interpretation.
- Research demonstrations do not directly measure harm in ordinary household deployment.
Evidence (4 items)
Supports the claim: Google,
directly output robotic actions.
Why it matters here: Google describes a language-and-vision model directing robotic actions and reports physical robot trials. This satisfies the antecedent under the stated interpretation, without establishing greater risk.
- Publisher
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: arXiv,
a deployed commercial robotic system.
Why it matters here: The researchers report attacks eliciting harmful physical actions from LLM-controlled robots. This substantiates a hazard, but supplies no matched estimate of home-assistant risk; the exact experiment dates are not given on the abstract page.
- Publisher
- arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Robocurve,
Fable refused 20 of 100
Why it matters here: The evaluation reports refusals and harmful-task completions across five physical-robot tasks. It has no home-assistant control group and explicitly limits generalization beyond its scenes; individual experiment dates are not stated.
- Publisher
- Robocurve
- Kind of source
- Independent measurement
- Describes events of
- Public from
Context only: arXiv,
unsafe plans
Why it matters here: The authors report substantial safety improvements in simulated and physical robot experiments. This cautions against treating unguarded robot failures as an inherent comparative disadvantage, but does not compare robots with home assistants; exact experiment dates are unstated.
- Publisher
- arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A transparent comparative evaluation or deployment study measuring harmful-failure likelihood and severity for LLM-controlled robots and comparable home assistants, accounting for tasks, exposure and safeguards.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, How come smart assistants have virtually no ability to converse, despite all the spectacular progress with large language models?
See also my essay on Google’s Palm-SayCan, for what might happen if this tech were embedded in robots, which would be even more risky.
- Where
- paragraph 24
- Published
- 2022-11-22
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-how-come-smart-assistants-have-virtually-34cc58-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-how-come-smart-assistants-have-virtually-34cc58-a1.