“As a result, every single LLM-based application — ChatGPT, Google Gemini, Anthropic's Claude, Meta's LLaMA, to name but a few — is trained on an indeterminately-large portion of the internet made up of both Common Crawl (a 250-billion page repository of web content) and what appears to be as much of the rest of the internet as they're able to download, legally or otherwise.”
Our reading, the sentence that was checked:
Every single LLM-based application, including ChatGPT, Google Gemini, Anthropic's Claude and Meta's LLaMA, is trained on an indeterminately large portion of the internet made up of Common Crawl and what appears to be as much of the rest of the internet as they are able to download, legally or otherwise.
Assessment: Contradicted Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, They're Looting The Internet, paragraph 39.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-the-great-looting-of-the-internet-2adae9-a1
We recorded 17 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 2024-04-15. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Observation
- 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
- indeterminately-large portion of the internet (qualifier, stated by the speaker)
- what appears to be as much of the rest of the internet as they're able to download (qualifier, stated by the speaker)
- legally or otherwise (qualifier, stated by the speaker)
From the speaker
- every single LLM-based application
- ChatGPT, Google Gemini, Anthropic's Claude, Meta's LLaMA
- trained on an indeterminately-large portion of the internet
- made up of both Common Crawl
- what appears to be as much of the rest of the internet as they're able to download
- legally or otherwise
Our interpretation
- The claim concerns the completed training of models underlying applications available by 2024-04-15. The historical training interval has no stated beginning.
- I take 'every single' literally, including specialized LLM applications; the named products are examples, not an exclusive list.
- The testable commitment is that each underlying model used Common Crawl and substantial additional internet material.
- The hedged language does not supply a minimum corpus size or establish that every developer acted illegally.
Counts as supported if: Training records establish the stated combination of Common Crawl and additional internet material for every application within scope.
Counts as contradicted if: A documented LLM application available before the post used an underlying model trained without Common Crawl or without the stated combination of sources.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Contradicted
The dated evidence goes against 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
StarCoder supplies a documented counterexample to 'every single.' Its paper restricts pretraining to The Stack, and its repository documents code-generation and assistant applications available before the post. That recipe does not use Common Crawl. LLaMA's documentation does support the narrower observation that prominent models use Common Crawl alongside other sources. A supporting example cannot rescue a universal statement with a counterexample.
Limitations
- The result addresses the explicit universal scope, not whether web-derived training is common.
- StarCoder is a specialized code LLM.
- The result does not determine the legality of any training corpus.
Evidence (3 items)
Goes against the claim: BigCode research collaboration, arXiv,
trained on permissively licensed data from The Stack
Why it matters here: The methods restrict StarCoderBase's training set to The Stack's GitHub repositories; StarCoder adds Python fine-tuning from that corpus. This documented training recipe does not include Common Crawl.
- Publisher
- BigCode research collaboration, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Goes against the claim: BigCode,
perform code generation
Why it matters here: The repository provides an application of the model and announces a coding assistant before Zitron's post. Together with e1, it establishes an application-level counterexample.
- Publisher
- BigCode
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: Meta AI, arXiv,
diverse pre-processed CommonCrawl datasets
Why it matters here: LLaMA's documented mixture includes Common Crawl, C4, and other sources. This supports the broader observation about major models, but cannot establish the universal claim.
- Publisher
- Meta AI, arXiv
- 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, They're Looting The Internet
As a result, every single LLM-based application — ChatGPT, Google Gemini, Anthropic's Claude, Meta's LLaMA, to name but a few — is trained on an indeterminately-large portion of the internet made up of both Common Crawl (a 250-billion page repository of web content) and what appears to be as much of the rest of the internet as they're able to download, legally or otherwise.
- Where
- paragraph 39
- Published
- 2024-04-15
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-the-great-looting-of-the-internet-2adae9-a1: Current. Outcome: Contradicted. 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-great-looting-of-the-internet-2adae9-a1.