“I do think AI will eventually work reliably.”
Our reading, the sentence that was checked:
AI will eventually work reliably.
Assessment: Unresolved No deadline stated on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, No, RAG is probably not going to rescue the current situation, paragraph 35.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-no-rag-is-probably-not-going-to-rescue-9eb5c1-a2
We recorded 15 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 2024-02-23. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Forecast
- Deadline, as the speaker wrote it
- “eventually”
- Deadline state when assessed
- No deadline stated
- Deadline state on 2026-10-29
- No deadline stated (the release’s reference date)
From the speaker
- AI
- will eventually work reliably
Our interpretation
- In context, reliable AI means systems whose factual answers and reasoning can generally be trusted across varied tasks, including tasks involving retrieved information.
- Success on a narrow task or improvement on one benchmark does not establish this broad outcome.
- Marcus supplies neither a numerical error threshold nor a deadline. No numerical threshold is imposed.
- The discussion of neurosymbolic AI is a possible route, not a condition attached to this forecast.
Counts as supported if: Public, independently replicated evaluations demonstrate broadly trustworthy factual output and reasoning across varied tasks and realistic use, with rare material errors and appropriate abstention.
Counts as contradicted if: Decisive evidence establishes that broadly reliable AI cannot ever be achieved. Failures in existing systems or the passage of time cannot establish this.
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 February 3, 2026 International AI Safety Report described improved reliability since January 2025 alongside continuing mistakes on complex tasks [e1]. The May 19, 2026 HalluWorld preprint reported strong performance on directly observed information but continuing difficulties with state tracking, causal simulation and abstention [e2]. Neither item supplies an independently replicated demonstration of broad reliability or establishes that eventual broad reliability is impossible [e1, e2]. The evidence therefore leaves the forecast unresolved [e1, e2].
Limitations
- The report describes current capabilities, and HalluWorld evaluates selected models in controlled environments; neither establishes what all future AI systems could achieve [e1, e2].
- Success on HalluWorld's directly observed information does not demonstrate reliable performance across varied realistic tasks [e2].
Evidence (2 items)
Context only: International AI Safety Report,
no combination of methods ensures the high reliability
Why it matters here: The report describes improving reliability alongside continuing failures. Its assessment of present limitations neither establishes eventual success nor rules it out.
- Publisher
- International AI Safety Report
- Kind of source
- Secondary summary
- Describes events of
- Public from
Context only: arXiv; Emmy Liu and coauthors,
In the terminal setting, models also struggle with when to abstain.
Why it matters here: HalluWorld reports strong results for directly observed information but continuing difficulties with state tracking, causal simulation and abstention. This is evidence of uneven progress, not a decisive demonstration of broad reliability.
- Where
- Abstract
- Publisher
- arXiv; Emmy Liu and coauthors
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A dated, independently replicated demonstration of broadly trustworthy factual output and reasoning across varied tasks and realistic use, with rare material errors and appropriate abstention; or decisive evidence ruling out its eventual achievement.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, No, RAG is probably not going to rescue the current situation
I do think AI will eventually work reliably.
- Where
- paragraph 35
- Published
- 2024-02-23
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-no-rag-is-probably-not-going-to-rescue-9eb5c1-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-no-rag-is-probably-not-going-to-rescue-9eb5c1-a2.