“Models like ChatGPT and Claude are deeply dependent on training data to improve their outputs, and their very existence is actively impeding the creation of the very thing they need to survive.”
Our reading, the sentence that was checked:
Models like ChatGPT and Claude depend on training data and their existence is impeding the creation of new human data they need.
Assessment: Supported Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, Bubble Trouble, paragraph 10.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-bubble-trouble-cfdfbf-a3
We recorded 23 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)
From the speaker
- Models like ChatGPT and Claude
- deeply dependent on training data to improve their outputs
- their very existence is actively impeding the creation
- the very thing they need to survive
Our interpretation
- The claim concerns human-created material useful for training language models.
- Impeding creation means reducing human contributions relative to what would have occurred without widespread LLM availability.
- The ordinary reading requires a demonstrated effect on relevant content sources, not a quantified decline in all human writing worldwide.
- Survive is treated as rhetorical language about maintaining and improving models, not a separately timed prediction of company failure.
- The causal effect must already have occurred by April 4, 2024; later retrospective studies may establish it.
Counts as supported if: Evidence establishes the importance of training for model outputs and credibly attributes reduced creation of relevant human content to LLM availability or use before the post.
Counts as contradicted if: Comparable causal evidence rules out the claimed reduction in relevant human-content creation, or shows that the alleged decline concerns only visits rather than contributions.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
Outcome: Supported
The dated evidence matches what the claim committed to, under the criteria frozen before grading.
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
OpenAI's March 15, 2023 report describes GPT-4's document-based pretraining and factuality improvements from post-training, supporting the importance of training for model outputs ([e3](https://arxiv.org/abs/2303.08774v1)). A retrospective study using less-exposed comparison platforms estimated a 25% relative reduction in Stack Overflow activity within six months of ChatGPT's release, through May 2023, and documented reduced content creation ([e1](https://discovery.ucl.ac.uk/id/eprint/10202775)). This supports a causal reduction in relevant human contributions before April 4, 2024 ([e1](https://discovery.ucl.ac.uk/id/eprint/10202775)). A second study, published May 6, 2024 and covering October 2021 through March 2023, found fewer Stack Overflow questions but no evidence of declining activity in Reddit developer communities, limiting how broadly the finding applies ([e2](https://www.nature.com/articles/s41598-024-61221-0)).
Limitations
- The causal estimate depends on the validity of the comparison platforms.
- The evidence does not establish a worldwide net decline in human content, a separate Claude effect, or reduced journalism output.
- The evidence does not establish a threat to model or company survival, or that additional human text is the only route to improvement.
Evidence (3 items)
Supports the claim: PNAS Nexus; institutional record at UCL Discovery,
reduces the production
Why it matters here: The study estimates a 25% relative reduction in Stack Overflow activity within six months of ChatGPT's release, using less-exposed comparison platforms. It examines content creation and reports no significant deterioration in post quality that would explain the loss as merely low-quality material.
- Publisher
- PNAS Nexus; institutional record at UCL Discovery
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Supports the claim: Scientific Reports,
declines in both website visits
Why it matters here: An independent analysis covering data through March 2023 also finds reduced question creation. Its contrasting finding of no decline in Reddit developer communities limits any generalization to all human-content sources.
- Where
- Abstract
- Publisher
- Scientific Reports
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Supports the claim: OpenAI via arXiv,
improved performance on measures of factuality
Why it matters here: The report describes GPT-4 as pretrained on document prediction and documents further output improvements from post-training. It supports the training-dependence component, without establishing that an ever-growing supply of new human text is the only route to improvement.
- Where
- Abstract
- Publisher
- OpenAI via arXiv
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Where's Your Ed At, Bubble Trouble
Models like ChatGPT and Claude are deeply dependent on training data to improve their outputs, and their very existence is actively impeding the creation of the very thing they need to survive.
- Where
- paragraph 10
- Published
- 2024-04-04
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-bubble-trouble-cfdfbf-a3: Current. Outcome: Supported. 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.
HeldBy this site
The final-text check failed: proposition_faithful, citations_correct.
About assessment wyea-bubble-trouble-cfdfbf-a2, which was not published.