“Instead, big tech has found itself peddling increasingly-more-expensive iterations of near-identical Large Language Models — a direct result of them all having to use the same training data, which it’s now running out of.”
Our reading, the sentence that was checked:
Big tech's LLMs are increasingly expensive, near-identical iterations because they all use the same training data, which is running out.
Assessment: Unresolved Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, Godot Isn't Making it, paragraph 23.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-godot-isnt-making-it-85796e-a1
We recorded 33 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 2024-12-03. 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
- increasingly-more-expensive iterations
- near-identical Large Language Models
- a direct result
- all having to use the same training data
- now running out of
Our interpretation
- Assess the frontier-model development cycle discussed in 2024 through publication.
- Read 'near-identical' as little meaningful difference in capabilities or useful behavior, not literally identical model weights.
- The surrounding discussion primarily concerns development and operating costs; customer prices are a plausible alternative reading.
- Read shared data as substantially overlapping usable training material. The causal claim requires evidence that limited data explains the cost and capability pattern.
- 'Running out' means a binding or worsening shortage of useful new data, not necessarily complete exhaustion.
Counts as supported if: Comparable evidence shows rising costs and little meaningful capability improvement across the relevant developers, and identifies shared, increasingly scarce training data as a substantial cause.
Counts as contradicted if: Comparable evidence shows substantial capability or cost improvements inconsistent with the claimed pattern, or establishes that the asserted shared-data bottleneck does not explain it.
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
There is evidence for rising training costs and a reported data bottleneck affecting Orion's coding performance. That does not establish that all major developers use effectively the same data or that this explains near-identical models. Anthropic's reported improvements at unchanged customer prices also complicate the claim. The broad causal conclusion remains unsettled.
Limitations
- The strongest specific support comes from reporting on anonymous sources.
- Training costs, operating costs, and customer prices are different measures.
- Company benchmarks do not independently establish real-world capability gains.
Evidence (4 items)
Supports the claim: Bloomberg, syndicated by Yahoo Finance,
coding data
Why it matters here: This supports a narrower causal link between insufficient coding data and one model's disappointing results. It relies on anonymous sources because the internal evaluations and datasets were not public.
- Publisher
- Bloomberg, syndicated by Yahoo Finance
- Kind of source
- News report
- Describes events of
- Public from
Context only: arXiv; Cottier and colleagues,
2.4x per year
Why it matters here: The study supports a historical rise in training costs. Its scope is broader than the particular 2024 LLM iterations and does not establish dataset scarcity as the cause.
- Publisher
- arXiv; Cottier and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Epoch AI,
between 2026 and 2032
Why it matters here: This projects exhaustion of usable public human-generated text under specified trends. It neither establishes that useful data was already exhausted in 2024 nor excludes shortages in particular domains.
- Where
- Opening summary
- Publisher
- Epoch AI
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Anthropic,
the same price and speed
Why it matters here: Anthropic reports coding and tool-use gains without higher customer prices. This challenges the customer-price reading, but customer prices do not measure development or operating costs.
- Publisher
- Anthropic
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
Comparable model costs, independent capability evaluations, and dataset disclosures or controlled experiments showing whether shared data scarcity caused the claimed pattern.
Where it was said (1)
Said once in the sources this register covers.
: Where's Your Ed At, Godot Isn't Making it
Instead, big tech has found itself peddling increasingly-more-expensive iterations of near-identical Large Language Models — a direct result of them all having to use the same training data, which it’s now running out of.
- Where
- paragraph 23
- Published
- 2024-12-03
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-godot-isnt-making-it-85796e-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.