“They are inherent to how the system makes predictions, and to the fact that LLMs themselves don’t have the wherewithal to look things up and reason over them.”
Our reading, the sentence that was checked:
Hallucinations are inherent to how LLMs make predictions and to their inability to look things up and reason over them.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, When Will the GenAI Bubble Burst?, paragraph 9.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-when-will-the-genai-bubble-burst-457817-a1
We recorded 19 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 2024-03-31. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Causal claim
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
From the speaker
- inherent to how the system makes predictions
- LLMs themselves
- don’t have the wherewithal to look things up and reason over them
Our interpretation
- The ordinary reading is that hallucinations arise partly from the prediction mechanism and partly from inadequate factual lookup and reasoning.
- Inherent means a structural source of errors, not that every answer must be false.
- The phrase 'LLMs themselves' may exclude external retrieval tools. Evidence about an augmented system therefore does not automatically refute the narrower claim about an unaided model.
- The assessment concerns the mechanisms of the LLMs discussed in March 2024. Later research may explain those mechanisms but cannot substitute newer systems for them.
Counts as supported if: Research establishes the claimed causal contributions for the relevant models, rather than merely showing that hallucinations occur.
Counts as contradicted if: Research shows that the asserted causal mechanism does not hold, or demonstrates lookup and reasoning within the scope meant by 'LLMs themselves'. Improvements alone do not disprove a structural risk.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
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 horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
Research supports a structural connection between statistical prediction and some hallucinations. It does not establish the full causal claim without qualifications. ReAct also shows that language models could already coordinate retrieval and reasoning with an external interface. Whether that contradicts Marcus depends on what he excludes with 'LLMs themselves'. The combined claim cannot fairly receive an unqualified verdict.
Limitations
- A conditional mathematical result is not a universal result about deployed systems.
- Successful retrieval and reasoning on selected tasks do not establish general reliability.
Evidence (2 items)
Supports the claim: arXiv; Adam Tauman Kalai and Santosh S. Vempala,
calibrated predictors of facts.
Why it matters here: The result supports a structural prediction-related source of hallucinations under calibration and factual-distribution assumptions. The authors explicitly caution that it does not establish inevitable hallucination for every language model or post-training procedure.
- Publisher
- arXiv; Adam Tauman Kalai and Santosh S. Vempala
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Google Research,
verbal reasoning traces
Why it matters here: The experiments use a frozen PaLM model with a Wikipedia interface to combine retrieved information and model-generated reasoning. This challenges a broad incapacity claim, but the external interface means it does not settle the narrower wording 'LLMs themselves'.
- Publisher
- Google Research
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
A clearly specified model boundary, together with controlled evidence establishing or refuting the claimed mechanisms across representative factual tasks.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, When Will the GenAI Bubble Burst?
They are inherent to how the system makes predictions, and to the fact that LLMs themselves don’t have the wherewithal to look things up and reason over them.
- Where
- paragraph 9
- Published
- 2024-03-31
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-when-will-the-genai-bubble-burst-457817-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-when-will-the-genai-bubble-burst-457817-a1.