“As more internet content is created, either partially or entirely through generative AI, the models themselves will find themselves increasingly inbred, training themselves on content written by their own models which are, on some level, permanently locked in 2023, before the advent of a tool that is specifically intended to replace content created by human beings.”
Our reading, the sentence that was checked:
As more internet content is made with generative AI, models will be trained more and more on AI-generated content and become increasingly inbred.
Assessment: Unresolved No deadline stated on 2026-10-29 Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, Are We Watching The Internet Die?, paragraph 13.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-are-we-watching-the-internet-die-a4f7ec-a1
We recorded 29 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 2024-03-11. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Conditional forecast
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No deadline stated
- Deadline state on 2026-10-29
- No deadline stated (the release’s reference date)
Conditions
- as more internet content is created through generative AI (antecedent, stated by the speaker)
From the speaker
- As more internet content is created
- partially or entirely through generative AI
- models themselves will find themselves increasingly inbred
- training themselves on content written by their own models
- on some level, permanently locked in 2023
Our interpretation
- The antecedent is growth in AI-generated internet content.
- The predicted outcome is increasing recursive dependence on earlier model outputs in subsequent model training, with resulting homogenization or loss of fresh human information.
- The qualified reference to being locked in 2023 is read as a claim about information ancestry, not an exact technical cutoff shared by every model.
- The claim concerns the trajectory of models trained from the internet. Deliberately constructed laboratory feedback loops alone do not establish that trajectory.
Counts as supported if: Longitudinal training-data and model analyses show that growing AI web content increasingly enters successive models' training sets, creates recursive ancestry and produces the predicted narrowing.
Counts as contradicted if: Comparable analyses show that successive models avoid increasing recursive dependence or retain diverse fresh information despite the antecedent. A limited successful mitigation is counterevidence but does not settle an unlimited forecast.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Did the forecast come true?
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 deadline stated
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The antecedent is supported: a primary study measures growth in AI-generated web text. Its training experiments also make the risk concrete. But this does not establish the full prediction that successive deployed models increasingly feed on their own ancestry and become progressively narrowed. Other research shows that retaining earlier data can mitigate collapse. No deadline was stated, and no event found meets the early-resolution rule.
Limitations
- Web-content estimates depend on imperfect AI detection.
- The strongest recent corpus study is a preprint.
- Experimental model behavior cannot establish undisclosed production training practices.
Evidence (2 items)
Supports the claim: arXiv; University of Maryland and Pangram Labs researchers,
increasingly AI-generated.
Why it matters here: The study reports an increasing detector-labeled AI share in FineWeb-filtered Common Crawl text, reaching 31.1% in August 2026. Its 800-model experiments show possible benefits and harms, but do not track recursive ancestry across successive deployed model generations.
- Publisher
- arXiv; University of Maryland and Pangram Labs researchers
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Proceedings of Machine Learning Research,
collapse is mitigated.
Why it matters here: The paper finds different outcomes when synthetic data replace earlier data versus accumulate alongside retained data. This challenges inevitability, but experimental mitigation does not establish what happened in undisclosed production training corpora.
- Publisher
- Proceedings of Machine Learning Research
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
What would settle it
A longitudinal audit linking growing AI web content to actual training mixtures, recursive provenance and declining diversity or fresh information across successive model generations.
Where it was said (1)
Said once in the sources this register covers.
: Where's Your Ed At, Are We Watching The Internet Die?
As more internet content is created, either partially or entirely through generative AI, the models themselves will find themselves increasingly inbred, training themselves on content written by their own models which are, on some level, permanently locked in 2023, before the advent of a tool that is specifically intended to replace content created by human beings.
- Where
- paragraph 13
- Published
- 2024-03-11
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-are-we-watching-the-internet-die-a4f7ec-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-are-we-watching-the-internet-die-a4f7ec-a1.