“As I've discussed before, today's models are rife with hallucinations that make integrating them actively dangerous for many businesses, and the only way to have fewer hallucinations is to give the models more compute power to process and analyze data, a dwindling resource that AI is actively helping to destroy.”
Our reading, the sentence that was checked:
Today's models are rife with hallucinations that make integrating them actively dangerous for many businesses, and the only way to reduce hallucinations is to give models more compute power to process and analyze data, which is a dwindling resource that AI is actively helping to destroy.
Assessment: Mixed Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, Bubble Trouble, paragraph 35.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-bubble-trouble-bac8b1-a1
We recorded 22 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 2024-04-04. 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
- for many businesses (scope, stated by the speaker)
From the speaker
- today's models are rife with hallucinations
- actively dangerous for many businesses
- the only way to have fewer hallucinations
- give the models more compute power
- a dwindling resource
- AI is actively helping to destroy
Our interpretation
- Assess methods and model risks documented before April 4, 2024.
- The ordinary reading of more compute power is increased model-processing capacity or computational scale, not every calculation involved in developing an alternative method.
- The exclusive claim fails if a method improves factuality through changes to an existing model that can be incorporated without increasing its serving computation.
- Business danger means a material risk from incorrect outputs in consequential workflows, not that every integration causes harm.
- Treat the hallucination-risk, mitigation and resource-depletion clauses separately.
Counts as supported if: Evidence establishes substantial hallucination risks in business workflows, necessity of increased processing capacity for reducing them, and an identifiable resource decline caused by AI.
Counts as contradicted if: A demonstrated alternative reduces hallucinations without increasing the deployed model's processing capacity, or evidence directly refutes another asserted causal link.
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
The risk warning has support: contemporaneous research found frequent errors in consequential legal tasks. The exclusive remedy does not hold under the ordinary processing-capacity reading. A 2023 study improved truthfulness by changing model activations and described incorporating the intervention into existing parameters. This does not show that preparation is computation-free or that hallucinations are solved. The claim that AI was destroying a dwindling resource remains unestablished.
Limitations
- The counterexample concerns deployed processing capacity, not total development computation.
- Benchmark truthfulness is narrower than reliability across all business uses.
- Support for business risk is task-specific.
- The resource-depletion clause remains unresolved.
Evidence (2 items)
Supports the claim: arXiv; Matthew Dahl and colleagues,
unsupervised integration
Why it matters here: The January version reports frequent hallucinations on verifiable legal questions, supporting a material risk in consequential business workflows. It supports the risk clause within legal tasks, not a measured count of affected businesses or the compute-depletion claim.
- Publisher
- arXiv; Matthew Dahl and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Goes against the claim: arXiv; Kenneth Li and colleagues,
offline editing
Why it matters here: The paper reports improved Alpaca truthfulness on TruthfulQA through activation intervention and explains how to incorporate it into existing parameters. This counters the necessity of greater deployed processing capacity, although learning the intervention requires offline computation and the result is benchmark-specific.
- Publisher
- arXiv; Kenneth Li and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
Matched total-compute experiments would resolve the broader compute reading. A defined resource and measured causal depletion trend are needed for the final clause.
Where it was said (1)
Said once in the sources this register covers.
: Where's Your Ed At, Bubble Trouble
As I've discussed before, today's models are rife with hallucinations that make integrating them actively dangerous for many businesses, and the only way to have fewer hallucinations is to give the models more compute power to process and analyze data, a dwindling resource that AI is actively helping to destroy.
- Where
- paragraph 35
- Published
- 2024-04-04
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-bubble-trouble-bac8b1-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.