“Even if they were successful in somehow acquiring this much training data, and doing so in a way that was legally sound (which, to be clear, I do not think is possible), they would then have the increasing costs of training these models.”
Our reading:
Even if AI companies somehow acquired this much training data in a legally sound way, which Zitron does not think is possible, they would then face the increasing costs of training these models.
Assessment: Unresolved
What happened
Did the forecast come true?
Outcome: Unresolved
We checked, and the evidence so far doesn’t settle it.
Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.
Evidence up to . Checked .
Why
Epoch AI's October 10, 2025 analysis estimates OpenAI's aggregate compute spending in 2024, including research, training and inference [e1]. It does not establish lawful acquisition of the required training corpus or a comparable cost trend across successive frontier models [e1]. This evidence therefore leaves the conditional prediction unresolved [e1].
Caveats
- The required corpus, its volume and the applicable legal standard remain undefined.
- The evidence does not establish the legal-acquisition condition or compare successive training efforts on a consistent accounting basis.
- Aggregate compute spending does not establish total training costs per model.
Evidence (1 item)
Context only: Epoch AI,
renting cloud compute from Microsoft.
Why it matters: The analysis estimates aggregate compute spending and separates research, inference and final training runs. It neither establishes lawful acquisition of the required corpus nor supplies a matching per-model cost comparison.
- Where
- Opening paragraph
- Publisher
- Epoch AI
- Kind of source
- Independent measurement
- Measures
- OpenAI, Aggregate cloud compute expenditure, including research, training and inference, USD dollars, estimate, 2024-01-01 to 2024-12-31
- About events on
- Published
How we judged it
- Kind of claim
- Conditional forecast
- Deadline
- None given. We don’t add one.
Conditions
- even if they were successful in acquiring this much training data legally (their words)
- which I do not think is possible (their words)
Where we had to interpret
- Preserve the explicit antecedent: developers must acquire the required volume of training data in a legally sound way before the cost prediction can be scored.
- Read increasing costs as higher total costs for successive frontier-model training efforts, rather than higher costs per token or higher inference expenditure.
- 'This much' refers to the preceding discussion of data for further model generations; no complete corpus size is specified.
- No accounting basis is supplied for the claimed cost trend. Adjacent CEO projections do not define a comparable realized-cost measure.
- Treat the parenthetical doubt about legality as a qualification, not a proven statement that the antecedent cannot happen.
Right if: After the legal-data-acquisition antecedent is established, comparable records show increasing training costs for the relevant successive frontier models.
Wrong if: After that antecedent is established, comparable completed training efforts show that the claimed increase did not occur.
We wrote this test after the outcome was known.
What would settle it
Identify the required corpus and successive models, establish lawful acquisition, and compare completed training costs using consistent treatment of hardware, energy, staff, experiments and final runs.
We found 48 claims in this post. See them all.
Full record
Source and dates
Where's Your Ed At, How Does OpenAI Survive?, paragraph 91.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-28. The post may have been edited after 2024-07-29.
Where it was said (1)
Said once in the posts we read.
: Where's Your Ed At, How Does OpenAI Survive?
Even if they were successful in somehow acquiring this much training data, and doing so in a way that was legally sound (which, to be clear, I do not think is possible), they would then have the increasing costs of training these models.
- Where
- paragraph 91
- Published
- 2024-07-29
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Unresolved. Evidence up to 2026-10-28.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.
Held backBy us
Both reviewers disagreed with the draft’s verdict.