“The current models developed by both the hyperscalers (Gemini, Llama, et. al) and multi-billion-dollar "startups" like OpenAI and Anthropic are horribly inefficient, I had just made the mistake of assuming that they'd actually tried to make them more efficient.”
Our reading, the sentence that was checked:
Current models from Google, Meta, OpenAI and Anthropic are horribly inefficient, and their makers never really tried to make them efficient.
Assessment: Contradicted Evidence as of
Where and when Ed Zitron wrote it
Where's Your Ed At, Deep Impact, paragraph 66.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment wyea-deep-impact-ef9280-a3
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 current models
- Gemini, Llama
- OpenAI and Anthropic
- horribly inefficient
- assuming that they'd actually tried to make them more efficient
Our interpretation
- Read the factual assertion as an absence of substantive efficiency work implemented in the models available by the post.
- The broader post acknowledges efficiency teams and an earlier OpenAI attempt. Therefore, do not interpret this as a literal assertion that no employee ever researched efficiency.
- Read efficiency ordinarily as reducing training or inference resources while preserving useful capability. Lower API prices alone do not establish lower production costs.
- Use the latest model disclosures available by January 29, 2025. The retrospective assertion about development efforts has no stated starting date.
- No numerical threshold is supplied for 'horribly'; the verdict tests the claimed lack of substantive effort, not optimality.
Counts as supported if: Comparable measurements establish substantial avoidable resource use, and development records substantiate the claimed absence of meaningful efficiency efforts.
Counts as contradicted if: Dated records show substantive efficiency engineering implemented before the post in the named model families.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
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 pre-post disclosures contradict the claimed absence of substantive efficiency efforts [e1] [e2] [e3] [e4]. On April 18, 2024, Meta described training infrastructure changes and reported roughly threefold higher Llama 3 training efficiency than Llama 2 [e1]. On May 14, 2024, Google introduced Gemini 1.5 Flash in public preview and described distillation from the larger Pro model [e2]. OpenAI's July 18, 2024 GPT-4o mini announcement documented an improved tokenizer shared with GPT-4o [e3]. Anthropic's March 4, 2024 announcement described the compact Claude 3 Haiku, whose availability was still forthcoming [e4]. These records document efficiency engineering and model development before January 29, 2025; they do not establish how inefficient the models remained [e1] [e2] [e3] [e4].
Limitations
- The disclosures provide no common capability-adjusted resource comparison that settles the description 'horribly inefficient' [e1] [e2] [e3] [e4].
- These company disclosures document engineering activity but do not independently verify the reported gains or establish optimal resource use [e1] [e2] [e3] [e4].
- The March 4, 2024 Anthropic announcement establishes Haiku's development, not its availability on that date [e4].
Evidence (4 items)
Goes against the claim: Meta,
efficiency of Llama 3 training
Why it matters here: Meta describes concrete changes to parallelization, reliability, storage and training infrastructure, alongside inference optimizations. This is substantive efficiency engineering, although the improvement magnitude is company-reported.
- Publisher
- Meta
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Goes against the claim: Google,
more cost-efficient to serve
Why it matters here: Google describes Gemini Flash as an efficiency-oriented model and explains its distillation from Pro. This documents an implemented effort before the post.
- Publisher
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Goes against the claim: OpenAI,
improved tokenizer
Why it matters here: The announcement documents a concrete tokenizer improvement shared with GPT-4o, alongside the smaller model's release. This establishes engineering effort without treating API pricing as a measurement of underlying compute cost.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Goes against the claim: Anthropic,
most compact model
Why it matters here: Anthropic describes an intentionally compact model within a family designed around capability, speed and cost tradeoffs. This documents efficiency-oriented model development, not proof of optimal resource use.
- Publisher
- Anthropic
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Where's Your Ed At, Deep Impact
The current models developed by both the hyperscalers (Gemini, Llama, et. al) and multi-billion-dollar "startups" like OpenAI and Anthropic are horribly inefficient, I had just made the mistake of assuming that they'd actually tried to make them more efficient.
- Where
- paragraph 66
- Published
- 2025-01-29
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment wyea-deep-impact-ef9280-a3: 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.
HeldBy this site
The final-text check failed: proposition_faithful.
About assessment wyea-deep-impact-ef9280-a2, which was not published.