“even if their models are all extremely similar, as they’re all trained on the same quickly-dwindling amount of training data.”
Our reading, the sentence that was checked:
Big tech's AI models are all extremely similar because they are trained on the same quickly-dwindling pool of training data.
Assessment: Unresolved Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, The Shareholder Supremacy, paragraph 89.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-tss-1bdafd-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 2024-07-01. 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
- their models are all extremely similar
- as they’re all trained on the same
- quickly-dwindling amount of training data
- Microsoft, Meta, Google and Amazon
Our interpretation
- Assess contemporary general-purpose models associated with the named companies, including relevant partnerships, rather than every specialized model they ever produced.
- The ordinary reading is substantial similarity in useful capabilities caused by substantial overlap in training sources and rapid exhaustion of unused suitable data.
- Read same as largely shared sources, not necessarily byte-identical training corpora.
- Read dwindling as a shrinking supply of additional usable material relative to training demand; data are not physically consumed by training.
- Use the first half of 2024 as the contemporary reference period. Similar benchmark scores alone would not establish the proposed cause.
Counts as supported if: Comparable evaluations establish substantial model similarity, corpus evidence establishes substantial overlap and rapid exhaustion of additional usable data, and causal evidence connects that overlap or constraint to the similarity.
Counts as contradicted if: Comparable evidence materially refutes a central component, or controlled evidence shows that the proposed shared-data mechanism does not explain the claimed similarity.
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
The sources establish public-data use, a plausible future public-text constraint, and research on model convergence. They do not establish the complete causal claim. Training disclosures do not measure overlap across the named companies, multimodal data complicate the shared-pool premise, and representation convergence is not the same as extreme similarity in useful capabilities. No evidence located isolates data overlap or depletion as the cause.
Limitations
- Incomplete training disclosures prevent strong conclusions in either direction.
- Company-authored capability reports are not independent comparisons.
- The data-exhaustion study is a conditional projection, not a demonstrated explanation of 2024 model behavior.
Evidence (4 items)
Context only: Meta,
publicly available sources
Why it matters here: Public-source training is consistent with possible overlap among developers. This disclosure does not quantify overlap with other companies or show that overlap caused similar capabilities.
- Publisher
- Meta
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: Epoch AI,
If trends continue
Why it matters here: The study makes prospective data constraints plausible under continued scaling. It does not demonstrate that a shrinking data pool was already causing the named companies' models to resemble each other in 2024.
- Where
- Opening summary
- Publisher
- Epoch AI
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Google Gemini Team, via arXiv,
image, audio
Why it matters here: Gemini's documented training extends beyond public text, so a public-text exhaustion estimate does not describe its entire data supply. The report also documents long-context capability differences, but neither fact alone settles an undefined claim of extreme overall similarity.
- Publisher
- Google Gemini Team, via arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Minyoung Huh and coauthors, via arXiv,
are converging.
Why it matters here: This research provides evidence relevant to model similarity, including convergence across modalities. Its hypothesis concerns shared statistical structure, not depletion of a common training corpus, and internal representation alignment is not equivalent to interchangeable capabilities.
- Publisher
- Minyoung Huh and coauthors, via arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
Comparable historical corpus inventories and capability evaluations, plus controlled experiments varying corpus overlap or data constraints while accounting for model size, training compute, architecture and post-training.
Where it was said (1)
Said once in the sources this register covers.
: Where's Your Ed At, The Shareholder Supremacy
even if their models are all extremely similar, as they’re all trained on the same quickly-dwindling amount of training data.
- Where
- paragraph 89
- Published
- 2024-07-01
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-tss-1bdafd-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.