“You do not know what you’re going to get each time, and hallucinations multiply with the complexity of the thing you’re asking for, or whether a task contains multiple steps (which is a fatal blow to the idea of “agents.””
Our reading, the sentence that was checked:
LLM hallucinations multiply with task complexity and multi-step tasks, which is fatal to the idea of agents.
Assessment: Unresolved Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, The Case Against Generative AI, paragraph 97.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-the-case-against-generative-ai-b42d05-a1
We recorded 46 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 2025-09-29. 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)
Conditions
- complex or multi-step tasks (qualifier, stated by the speaker)
From the speaker
- hallucinations multiply with the complexity
- a task contains multiple steps
- a fatal blow to the idea of “agents.”
Our interpretation
- The surrounding text expressly includes reasoning, instruction-following and task-execution errors within 'hallucinations'.
- 'Multiply' means errors become more prevalent as complexity or step count increases; no numerical multiplier is specified.
- The ordinary reading of 'fatal blow' is that error accumulation prevents agents from completing useful autonomous work reliably. Occasional failures alone do not establish this.
- Assess the causal argument about systems available at publication, rather than an unstated prediction about all future systems.
Counts as supported if: Controlled comparisons show increasing complexity or step count causes more errors, and evidence establishes that these errors prevent reliable useful autonomous work despite available methods of checking and recovery.
Counts as contradicted if: Controlled comparisons undermine the proposed relationship, or repeated evaluations demonstrate reliable useful autonomous work despite complexity and multiple steps, sufficient to refute the claimed fatal consequence.
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
Controlled research supports increased failures at higher complexity in some tasks. Follow-up experiments show that tools and test design materially affect those results. This supports the narrower concern about errors but does not settle the claimed fatal consequence for agents. Calling the whole claim supported would overstate the evidence; calling it contradicted would overlook the documented failures.
Limitations
- Puzzle performance is not a direct measure of practical autonomous work.
- The cited studies are preprints.
- The speaker supplies no operational meaning for 'fatal'.
Evidence (3 items)
Supports the claim: Shojaee et al., arXiv,
complete accuracy collapse
Why it matters here: Supports the complexity-related error claim in the tested puzzles. It does not establish that errors are fatal to practical agents.
- Publisher
- Shojaee et al., arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Song, Yue and Zhang, arXiv,
proper tool use
Why it matters here: Tool-assisted results caution against treating unassisted puzzle failures as an intrinsic limit on agents. Relative improvement does not itself establish reliable autonomous work.
- Where
- Abstract, results
- Publisher
- Song, Yue and Zhang, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Dellibarda Varela et al., arXiv,
over 100 agent pairs
Why it matters here: Some apparent complexity failures depended on unsolvable test configurations. The agent pairs here are puzzle participants, not deployed AI agents.
- Publisher
- Dellibarda Varela et al., arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
Repeated controlled evaluations of representative autonomous tasks, with explicit reliability requirements and error-recovery methods, showing whether increasing complexity prevents agents from meeting those requirements.
Where it was said (1)
Said once in the sources this register covers.
: Where's Your Ed At, The Case Against Generative AI
You do not know what you’re going to get each time, and hallucinations multiply with the complexity of the thing you’re asking for, or whether a task contains multiple steps (which is a fatal blow to the idea of “agents.”
- Where
- paragraph 97
- Published
- 2025-09-29
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-the-case-against-generative-ai-b42d05-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 wyea-the-case-against-generative-ai-b42d05-a1.