“We will eventually get to something better, but continuing to put all our eggs in the Henrietta’s LLM basket is absurd to the point of being delusional.”
Our reading, the sentence that was checked:
AI will eventually get to something better than LLMs.
Assessment: Unresolved No deadline stated on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Why DO large language models hallucinate?, paragraph 28.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-why-do-large-language-models-hallucinate-f1b4a6-2-a1
We recorded 9 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 2025-05-05. 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
- We will eventually get to something better
- continuing to put all our eggs in the Henrietta’s LLM basket
Our interpretation
- In context, 'better' means materially better factual reliability and checking of answers against records or sources.
- A qualifying approach moves beyond reliance on an LLM alone. A hybrid using language models alongside databases or symbolic verification can qualify.
- I require reproducible comparative evidence across varied factual tasks, with independent corroboration, to establish a successor's reliable advantage. These evidence requirements are mine, not thresholds Marcus supplied.
- Neither widespread adoption nor complete elimination of hallucinations is required by his words.
- The criticism of investment choices is evaluative rhetoric and is not part of the forecast being scored.
Counts as supported if: A post-publication non-LLM or hybrid approach demonstrates a reproducible, independently corroborated improvement in factual accuracy and source verification over the LLM approach discussed, across varied factual tasks and without obtaining its advantage merely by declining more questions.
Counts as contradicted if: Decisive evidence rules out the predicted improvement. Failure to produce a qualifying successor by any finite evaluation date does not contradict 'eventually'.
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
Marcus predicts eventual improvement without a deadline or a defined success threshold. The records show promising hybrid developments: graph-enhanced retrieval improves results in one study, and AWS offers formal checks against domain policies. These are material evidence in the forecast's direction. They do not establish the independently corroborated, broader reliability advantage required by this interpretation. Separate fact-validation research also finds inconsistent benefits from retrieval and consensus. No decisive event meets the early-resolution rule, and the passage of time cannot falsify 'eventually'.
Limitations
- The result depends on how broadly 'something better' is read.
- The graph study covers one benchmark and one model and was funded by Neo4j.
- AWS's announcement is provider evidence, not independent validation.
- Context items establish progress and limitations; none settles the forecast.
Evidence (3 items)
Context only: National Innovation Centre for Data, Newcastle University,
broader validation.
Why it matters here: The researchers report improved truthfulness and answer coverage from graph-enhanced retrieval, making this a plausible instance of the predicted direction. Their single-model, single-benchmark scope does not establish the broader corroborated advantage required by this contract.
- Publisher
- National Innovation Centre for Data, Newcastle University
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: Amazon Web Services,
Automated Reasoning checks are used to assess AI-generated responses.
Why it matters here: AWS announced general availability of checks that apply formal logic to generated responses against encoded domain policies. This establishes a concrete hybrid capability, but the announcement does not provide an independent comparison of overall factual reliability across varied tasks.
- Publisher
- Amazon Web Services
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: arXiv; Farzad Shami, Stefano Marchesin and Gianmaria Silvello,
multi-model consensus
Why it matters here: This evaluation reports inconsistent gains from retrieval and model consensus in knowledge-graph fact validation. It limits generalization from particular hybrid successes, but cannot refute an eventual successor.
- Publisher
- arXiv; Farzad Shami, Stefano Marchesin and Gianmaria Silvello
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A dated non-LLM or hybrid system with reproducible, independently corroborated gains in factual accuracy and source verification across varied tasks at comparable answer coverage would satisfy the support condition.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Why DO large language models hallucinate?
We will eventually get to something better, but continuing to put all our eggs in the Henrietta’s LLM basket is absurd to the point of being delusional.
- Where
- paragraph 28
- Published
- 2025-05-05
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-why-do-large-language-models-hallucinate-f1b4a6-2-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-why-do-large-language-models-hallucinate-f1b4a6-2-a1.