“High-reliability is not their forte. and nor are outliers. They won’t take humans out of the loop.”
Our reading, the sentence that was checked:
LLMs (transformers) are not highly reliable and do not handle outliers well, and they won't take humans out of the loop in the pursuit of Level 5 self-driving.
Assessment: Unresolved No deadline stated on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Rethinking “driverless cars”, paragraph 19.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-rethinking-driverless-cars-c1c477-a2
We recorded 13 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 2023-11-05. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- 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
- in the context of transformers not getting us to Level 5 self-driving (scope, stated by the speaker)
From the speaker
- High-reliability is not their forte
- nor are outliers
- They won’t take humans out of the loop
Our interpretation
- The preceding sentence identifies LLMs and transformers and expressly discusses Level 5 driving. I assess the forecast in that context.
- The Level 5 context defines the target; it is not an antecedent that must first occur before the forecast can be assessed.
- The plain reading is that these models will remain unable to provide sufficiently reliable driving, including unusual situations, without human assistance in the driving task.
- Removing humans from the loop means removing required onboard or remote human driving, fallback, and case-specific guidance needed to resolve driving situations. It does not mean eliminating human training, maintenance or emergency services.
- The reliability and outlier remarks provide qualitative reasons for the forecast. They supply no separate numerical threshold, so later benchmark results cannot by themselves settle those remarks as historical observations.
Counts as supported if: A validated impossibility result showing that the relevant model class cannot provide Level 5 driving without required human driving assistance could support the categorical forecast. Individual failures or continued assistance do not establish permanent inability.
Counts as contradicted if: Independent documentation establishes Level 5 operation materially delivered by transformer-based language models, with reliability and unusual-situation validation and no required onboard or remote human driving assistance.
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
On February 17, 2026, Waymo said its remote assistants provide advice that its driving system can accept or reject, rather than remotely driving vehicles [e5]. On August 4, 2026, NVIDIA reported that Alpamayo 2 Super ranked first among nearly 40 models on the LingoQA driving-reasoning benchmark [e6]. Neither disclosure independently validates transformer-based language models delivering Level 5 driving without required human assistance, and neither establishes that such operation is impossible [e5, e6]. The evidence therefore leaves the forecast unresolved [e5, e6].
Limitations
- Both items are company disclosures, not independent validation of qualifying Level 5 operation.
- Benchmark rankings do not establish absolute safety or reliability across unusual driving situations.
- The evidence does not establish that required human assistance is permanently unavoidable.
- The qualitative reliability and outlier remarks supply no numerical threshold for a separate verdict.
Evidence (2 items)
Context only: Waymo,
Waymo’s service does not rely on remote drivers.
Why it matters here: Waymo distinguishes remote driving from advice that its driving system can accept or reject. This challenges treating all remote assistance as human driving, but does not establish an LLM-based Level 5 system without human assistance.
- Publisher
- Waymo
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: NVIDIA,
nearly 40 models evaluated.
Why it matters here: NVIDIA reports progress in driving-specific reasoning, relevant to the claim's pessimistic premises. A relative benchmark ranking does not establish absolute safety, comprehensive outlier handling, or human-free Level 5 operation.
- Publisher
- NVIDIA
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
Independent documentation of Level 5 operation materially delivered by transformer-based language models, with validated reliability in ordinary and unusual situations and no required onboard or remote human driving, fallback or case-specific guidance, would contradict the forecast. A validated impossibility result for the relevant model class could support it; continued shortcomings alone cannot.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Rethinking “driverless cars”
High-reliability is not their forte. and nor are outliers. They won’t take humans out of the loop.
- Where
- paragraph 19
- Published
- 2023-11-05
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-rethinking-driverless-cars-c1c477-a2: 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-rethinking-driverless-cars-c1c477-a2.