“And the reason why OpenAI hasn’t fixed the hallucination problem isn’t because it doesn’t want to, but because it can’t. They’re an inevitable side-effect of LLMs as a whole.”
Our reading, the sentence that was checked:
Hallucinations are an inevitable side-effect of LLMs, so OpenAI cannot fix them.
Assessment: Unresolved Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, There Is No AI Revolution, paragraph 52.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-wheres-the-money-dcf3e3-a1
We recorded 42 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 2025-02-24. 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
- isn’t because it doesn’t want to, but because it can’t
- an inevitable side-effect of LLMs as a whole
Our interpretation
- The plain reading is that hallucinations are an unavoidable property of useful general-purpose LLM operation, making a fix technically impossible.
- A theorem about answering every possible question must be distinguished from a theorem requiring false assertions when uncertainty and abstention are allowed.
- A finite error-free benchmark or a system that always refuses to answer is insufficient to establish a general solution.
- The asserted architectural explanation is assessed causally, without adding a forecast deadline.
Counts as supported if: A theoretical result establishes unavoidable false factual assertions under assumptions covering useful general-purpose LLM operation and the ordinary meaning of hallucination.
Counts as contradicted if: A reproducible general solution, or a theoretical result covering the same practical scope, establishes that useful LLM operation need not produce false factual assertions.
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
There is a formal impossibility result, but its definition counts disagreement with a required correct output. That does not directly establish that useful systems must assert falsehoods when they may acknowledge uncertainty. Other research identifies incentives that encourage guessing and proposes mitigation. Neither record settles the unqualified claim that OpenAI cannot fix hallucinations.
Limitations
- The result depends on what counts as fixing hallucinations.
- A theoretical limit on universal correctness is not a measurement of unavoidable practical falsehood rates.
- The research does not establish OpenAI's internal reasons or intentions.
Evidence (2 items)
Context only: arXiv,
impossible to eliminate hallucination
Why it matters here: This supports inevitability within the authors' formal framework. Their definition concerns disagreement with a prescribed correct output, so applying it to useful systems that can acknowledge uncertainty requires an additional argument.
- Publisher
- arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Nature,
an incentive problem
Why it matters here: The paper identifies incentives favoring guesses over uncertainty and proposes evaluation changes. This challenges a blanket architectural explanation but does not demonstrate complete elimination in useful general-purpose operation.
- Publisher
- Nature
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
What would settle it
A general result explicitly covering useful factual responses with abstention and verification, or a reproducible solution with guarantees matching that scope.
Where it was said (1)
Said once in the sources this register covers.
: Where's Your Ed At, There Is No AI Revolution
And the reason why OpenAI hasn’t fixed the hallucination problem isn’t because it doesn’t want to, but because it can’t. They’re an inevitable side-effect of LLMs as a whole.
- Where
- paragraph 52
- Published
- 2025-02-24
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-wheres-the-money-dcf3e3-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.