“Putting aside terms like "performance" — as they are largely used as a means of generating things "accurately" or "faster" rather than being good at stuff — Large Language Models have effectively plateaued.”
Our reading, the sentence that was checked:
Setting aside 'performance' measures of accuracy or speed, Large Language Models have effectively plateaued.
Assessment: Unresolved Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, The Subprime AI Crisis, paragraph 68.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-subprimeai-8976bc-a1
We recorded 50 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 2024-09-16. 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
- putting aside terms like 'performance' (accuracy or speed) (qualifier, stated by the speaker)
From the speaker
- Putting aside terms like "performance"
- "accurately" or "faster"
- Large Language Models have effectively plateaued
- "More powerful" never seems to mean "does more"
Our interpretation
- This describes the state of LLM capabilities at publication, not a prediction about all subsequent development.
- With no reporting period specified, use the latest partial calendar month before publication, September 1–16, 2024, as the contemporary observation window. Earlier models provide comparison baselines.
- The ordinary reading is that contemporary models had stopped adding materially different useful task capabilities. Accuracy or speed improvements alone do not contradict the stated claim.
- A new training method or product interface is not automatically a new model task capability.
Counts as supported if: Contemporaneous comparative evaluations show no material extension of useful task capabilities beyond earlier models, after excluding accuracy and speed improvements.
Counts as contradicted if: Contemporaneous demonstrations and evaluations establish that current LLMs can perform materially different useful tasks that preceding models could not perform meaningfully.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
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 September evidence challenges a broad claim of stagnation: o1 introduced a different reasoning approach, and ARC Prize described improved generalization. But the claim explicitly excludes accuracy and speed, and the retrieved evaluations largely measure success on existing tasks. They do not cleanly settle whether useful task capabilities had plateaued under that narrower reading. Later releases cannot decide this historical observation.
Limitations
- The distinction between a new capability and better execution is consequential and unstated.
- The selected observation window is an evaluator interpretation.
- The search does not establish an exhaustive inventory of contemporary demonstrations.
Evidence (2 items)
Context only: OpenAI,
perform complex reasoning
Why it matters here: The announcement describes a different reasoning approach and substantial results on math, coding and science evaluations. Much of the evidence measures success rates on existing task categories, which the claim expressly distinguishes from doing new things.
- Where
- Opening paragraph
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: ARC Prize,
greater generalization
Why it matters here: The evaluators identified improved adaptation to novelty, which challenges a broad claim of stagnation. Their measured results also placed o1-preview near Claude 3.5 Sonnet on this benchmark, leaving the narrower claim about materially different useful tasks unsettled.
- Where
- Test-Time Compute
- Publisher
- ARC Prize
- Kind of source
- Independent measurement
- Describes events of
- Public from
What would settle it
Comparative evidence about models available by publication showing materially new useful task abilities, or a broad evaluation establishing their absence under an explicit capability definition.
Where it was said (1)
Said once in the sources this register covers.
: Where's Your Ed At, The Subprime AI Crisis
Putting aside terms like "performance" — as they are largely used as a means of generating things "accurately" or "faster" rather than being good at stuff — Large Language Models have effectively plateaued.
- Where
- paragraph 68
- Published
- 2024-09-16
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-subprimeai-8976bc-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.