“As I've written about many, many, many times, the Large Language Models run by companies like OpenAI, Anthropic, Google and Meta are unprofitable and unsustainable, and the transformer-based architecture they run on has peaked.”
Our reading, the sentence that was checked:
The transformer-based architecture underlying LLMs has peaked.
Assessment: Unresolved Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, Deep Impact, paragraph 56.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-deep-impact-9b9bed-2-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 2025-01-29. 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)
From the speaker
- the transformer-based architecture
- has peaked
Our interpretation
- The ordinary reading is that transformer-based models had reached a substantive capability ceiling by publication.
- This is assessed as a historical observation, consistent with its past-tense wording, rather than converted into an indefinite forecast.
- Because no observation period is named, the target is the latest partial calendar month before publication: January 1–29, 2025.
- Progress before January does not alone disprove a ceiling allegedly reached by publication. Later model results are context under this historical contract.
- Benchmark saturation and limitations of pretraining alone are insufficient to establish a ceiling for the architecture as a whole.
Counts as supported if: Evidence applicable to January 2025 establishes a broad capability ceiling for transformer-based models, rather than merely diminishing returns on selected tasks or training methods.
Counts as contradicted if: Contemporaneous, reproducible evidence establishes that substantive capability gains remained available within transformer-based models, with sufficient information to distinguish architectural limits from changes to surrounding systems.
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 reviewed evidence does not establish that transformers had reached a capability ceiling in January 2025. DeepSeek documented reasoning gains, and later independent measurements show continued agent progress. Neither directly isolates an architecture-wide ceiling at the claimed date. Under the historical contract, later results remain context rather than decisive outcome evidence. This is a limitation of adjudication, not evidence supporting the ceiling.
Limitations
- The result depends materially on treating the sentence as a historical observation.
- A forecast reading would permit a different evaluation of later capability gains.
- Improvements in one model and limits on one training method do not establish the frontier of an architecture.
Evidence (2 items)
Context only: DeepSeek-AI / arXiv,
steady and consistent
Why it matters here: The paper documents large reasoning improvements during reinforcement learning, weighing against a general claim that model capabilities could no longer improve. It does not establish whether the best transformer-based systems had reached a ceiling by January 29.
- Publisher
- DeepSeek-AI / arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: METR,
1-4 doublings
Why it matters here: METR reports continued growth in the duration of tasks model-based agents can complete. This is material context, but its period extends beyond January and the study measures complete agents rather than isolating transformer architecture.
- Publisher
- METR
- Kind of source
- Independent measurement
- Describes events of
- Public from
What would settle it
Reproducible evidence applicable to January 2025 that directly tests whether transformer-based models retained substantial capability headroom, or a clarified claim specifying its capability measure and whether it predicts future stagnation.
Where it was said (1)
Said once in the sources this register covers.
: Where's Your Ed At, Deep Impact
As I've written about many, many, many times, the Large Language Models run by companies like OpenAI, Anthropic, Google and Meta are unprofitable and unsustainable, and the transformer-based architecture they run on has peaked.
- Where
- paragraph 56
- Published
- 2025-01-29
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-deep-impact-9b9bed-2-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.