“The answer is that at present we have no viable option; AlphaStar and LLMs would surely be unsafe and inadequate; and we don’t yet have anything markedly better.”
Our reading:
There is currently no viable general intelligence to install in a domestic humanoid robot; LLMs and AlphaStar would be unsafe and inadequate.
Assessment: Unresolved
What happened
Was it true for the period described?
Outcome: Unresolved
We checked, and the evidence so far doesn’t settle it.
Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.
Evidence up to . Checked .
Why
The records are consistent with caution about practical readiness, but do not settle the full assertion. SayCan demonstrated useful kitchen-task planning, and Code as Policies demonstrated LLM-generated robot control. Neither establishes safe general household operation. Gato was described as a research prototype without user deployment, but that does not establish that every October option was inadequate. Conversely, missing safety demonstrations do not prove that AlphaStar and all contemporary LLM systems would surely be unsafe. The available evidence therefore supports neither a decisive endorsement nor a contradiction under the stated household-controller reading.
Caveats
- The selected experiments predate the target snapshot and are context rather than outcome evidence.
- Task success, practical adequacy, and physical safety are different measures.
- Some AlphaStar primary material could not be retrieved in a usable form.
- A narrower claim that no safe general household controller had been demonstrated would require less evidence than the categorical wording assessed here.
Evidence (3 items)
Context only: Robotics at Google and Everyday Robots,
a robot’s performance can be improved
Why it matters: The dated update reports 84% planning success and 74% execution success; the results describe 101 tasks in two kitchen settings. This establishes useful LLM-based household-task capabilities, but neither general household safety nor failure after granting perfect motor control.
- Publisher
- Robotics at Google and Everyday Robots
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- About events on
- Published
Context only: arXiv; Jacky Liang and colleagues,
write robot policy code
Why it matters: The original preprint describes LLM-generated control policies demonstrated across robot platforms. This challenges a reading that LLMs had no useful control capabilities, but does not establish safe general operation of a domestic humanoid.
- Publisher
- arXiv; Jacky Liang and colleagues
- Kind of source
- Preprint, not peer reviewed
- About events on
- Published
Context only: DeepMind; Scott Reed and colleagues,
not currently deploying Gato to any users
Why it matters: The paper describes a generalist research agent, calls the work proof-of-concept, and reports no user deployment. This is consistent with limited practical readiness, but its May status does not establish the absence of every viable option in October.
- Publisher
- DeepMind; Scott Reed and colleagues
- Kind of source
- Preprint, not peer reviewed
- About events on
- Published
Probably right?
One AI thinks this is probably right, and the other says it’s too early to tell. The evidence isn’t conclusive, so the verdict above stays unresolved. It counts only when you pick Speculative under “How sure?”.
- Anthropic’s AI: The best LLM robot systems of 2022 were research prototypes. SayCan failed about 26% of 101 tasks in two kitchens, and Gato was proof-of-concept and not deployed. Nothing in the packet shows a usable general household controller in October 2022.
How we judged it
- Kind of claim
- Observation
- Deadline
- None given. We don’t add one.
Conditions
- as general intelligence for a domestic humanoid robot (their words)
Where we had to interpret
- The plain reading is that available AI could not provide sufficiently reliable and safe general household decision-making for a domestic humanoid robot.
- The surrounding hypothetical grants successful motor control. Failures of locomotion or manipulation hardware alone therefore cannot establish the claim.
- Viability means practical suitability across ordinary household tasks and unexpected situations. It does not require perfect performance or a particular philosophical definition of intelligence.
- The target is the publicly observable state on 2022-10-16. This technology has no regular reporting period, so I use the post-date snapshot rather than inventing a quarterly or annual window.
- Earlier experiments inform that snapshot but remain context when their measured events occurred outside the target period.
- The household application defines the scope of this observation; it is not a conditional forecast.
Right if: Evidence establishing that the available candidate controllers lacked the reliability, general household competence, and safety needed for this use, including evidence supporting the asserted inadequacy and unsafety of AlphaStar and contemporary LLM systems. Missing demonstrations alone are insufficient to establish the entire claim.
Wrong if: A documented system available at the target date that could provide safe, adequate general household decision-making with functioning motor skills would contradict the no-option assertion. Successful isolated tasks alone would not suffice.
We wrote this test after the outcome was known.
What would settle it
Dated evaluations of systems available on 2022-10-16 showing safe, reliable general household decision-making would contradict the no-option assertion. A sufficiently comprehensive historical assessment documenting capability and safety failures across the available candidates, with motor limitations separated out, could support it.
We found 14 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, Three baffling claims about AI and machine learning in four days, statistical errors in top journals, and claims from Yann LeCun that you should not believe., paragraph 86.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-29. The post may have been edited after 2022-10-16.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, Three baffling claims about AI and machine learning in four days, statistical errors in top journals, and claims from Yann LeCun that you should not believe.
The answer is that at present we have no viable option; AlphaStar and LLMs would surely be unsafe and inadequate; and we don’t yet have anything markedly better.
- Where
- paragraph 86
- Published
- 2022-10-16
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Unresolved. Evidence up to 2026-10-28.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.