“Because model developers hit a wall of diminishing returns, and the only way to make their models do more was to make them burn more tokens to generate a more accurate response (this is a very simple way of describing reasoning, a thing that OpenAI launched in September 2024 and others followed).”
Our reading, the sentence that was checked:
Inference costs rose because model developers hit diminishing returns and could only improve models by having them burn more tokens through reasoning.
Assessment: Contradicted Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, How To Argue With An AI Booster, paragraph 245.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-how-to-argue-with-an-ai-booster-d06e78-a1
We recorded 35 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 2025-08-25. 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
- hit a wall of diminishing returns
- the only way
- burn more tokens
- generate a more accurate response
- OpenAI launched in September 2024
Our interpretation
- The surrounding passage asserts that rising inference costs resulted from developers having no remaining route to better models except longer reasoning outputs.
- Burn more tokens refers to tokens generated during inference, not tokens consumed during training.
- The exclusivity is explicit: a demonstrated improvement from another route is sufficient to contradict the full explanation, even if longer reasoning also improves some results.
Counts as supported if: Controlled scaling studies show diminishing training returns left no effective alternative improvement route, and comparable workload measurements show longer reasoning caused the asserted cost increase.
Counts as contradicted if: A credible pre-publication model comparison demonstrates capability improvements through another route, or controlled cost evidence refutes the asserted causal chain.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
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
The exclusive explanation is contradicted. Qwen's pre-publication report documents gains in pretrained base models through improvements made before reasoning post-training. Longer inference reasoning therefore was not the only route to better performance. DeepSeek's report supports the narrower point that longer reasoning can improve results. The evidence does not settle how much reasoning contributed to industry costs or how sharply training returns diminished.
Limitations
- The decisive evaluation is developer-reported.
- It does not independently isolate the contribution of each training change or provide a complete workload-cost breakdown.
- The verdict addresses the explicit claim of exclusivity, not a claim that reasoning is unimportant.
Evidence (2 items)
Goes against the claim: Qwen Team / arXiv,
Qwen3-32B-Base achieves 65.54 on MMLU-Pro and 39.78 on SuperGPQA, significantly outperforming its predecessor Qwen2.5-32B-Base.
Why it matters here: The report documents gains in the pretrained base model under a common evaluation pipeline, separately from its later reasoning post-training. Its architecture, data, and training improvements supply another route to capability gains, contradicting the claim of exclusivity.
- Publisher
- Qwen Team / arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: DeepSeek-AI / arXiv,
This approach has achieved significant improvements in various reasoning tasks, such as mathematics, coding, and scientific reasoning.
Why it matters here: The preceding sentence identifies the approach as increasing chain-of-thought length. This supports the narrower contention that additional inference reasoning can improve performance, but not that it was the only available route.
- Publisher
- DeepSeek-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, How To Argue With An AI Booster
Because model developers hit a wall of diminishing returns, and the only way to make their models do more was to make them burn more tokens to generate a more accurate response (this is a very simple way of describing reasoning, a thing that OpenAI launched in September 2024 and others followed).
- Where
- paragraph 245
- Published
- 2025-08-25
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-how-to-argue-with-an-ai-booster-d06e78-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-how-to-argue-with-an-ai-booster-d06e78-a1.