“Its software lies because it doesn’t actually know anything and generates answers probabilistically”
Our reading, the sentence that was checked:
AI software lies because it doesn't actually know anything and generates answers probabilistically.
Assessment: Mixed Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, The AI Hater's Manifesto, paragraph 215.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-the-ai-haters-manifesto-97790a-a1
We recorded 50 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 2026-08-25. 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
- Its software lies
- because it doesn’t actually know anything
- generates answers probabilistically
Our interpretation
- Read software as LLM-based software in this context.
- Read lies colloquially as confidently producing false factual statements, without assuming conscious intent to deceive.
- Test the proposed connection between statistical generation, factual information available to a model, and false answers.
- Use an operational meaning of knowledge: factual information a model can retrieve or correctly recognize. The claim supplies no test for conscious or human-like knowledge.
- Do not weaken does not know anything into the narrower assertion that models sometimes lack relevant information.
Counts as supported if: Mechanistic analysis or controlled experiments show that statistical generation under inadequate factual information produces false assertions. Full support also requires evidence for the asserted absence of factual knowledge.
Counts as contradicted if: Controlled evidence shows models can access relevant factual information, including recognizing claims as false that they nevertheless generated. This contradicts absence of knowledge as a general explanation, even if statistical generation contributes to errors.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
Outcome: Mixed
Part of the claim holds and part does not, and the record says which part is which.
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 the narrower mechanism: statistical training can produce false assertions when facts cannot be distinguished from plausible errors. But the claim that the software knows nothing overstates that explanation. Controlled tests show models can recognize many false statements they previously generated. The statistical contribution is supported; the blanket absence-of-knowledge explanation is contradicted under the stated operational reading. Neither result establishes intentional lying.
Limitations
- The result depends on reading knowledge as accessible factual information.
- A philosophical reading of knows would not be settled by these experiments.
- The cited studies concern particular models and theoretical assumptions, not every AI system.
Evidence (2 items)
Supports the claim: arXiv; Adam Tauman Kalai, Ofir Nachum, Santosh S. Vempala, Edwin Zhang,
natural statistical pressures
Why it matters here: The analysis supports a statistical mechanism connecting inability to distinguish facts from falsehoods with hallucination. Its conditional result does not establish that models possess no factual information.
- Publisher
- arXiv; Adam Tauman Kalai, Ofir Nachum, Santosh S. Vempala, Edwin Zhang
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Goes against the claim: arXiv; Muru Zhang, Ofir Press, William Merrill, Alisa Liu, Noah A. Smith,
their own mistakes
Why it matters here: In controlled tasks, models could recognize many false claims they had produced. This is counterevidence to the claim that false answers generally arise because models have no accessible factual knowledge.
- Publisher
- arXiv; Muru Zhang, Ofir Press, William Merrill, Alisa Liu, Noah A. Smith
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Where's Your Ed At, The AI Hater's Manifesto
Its software lies because it doesn’t actually know anything and generates answers probabilistically
- Where
- paragraph 215
- Published
- 2026-08-25
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-the-ai-haters-manifesto-97790a-a1: Current. Outcome: Mixed. 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.