Discussion

Mother Jones contrasts rival readings of AI research acceleration

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#15121

A Mother Jones article sets stark warnings about near-term AI risks against evidence that research acceleration may be less dramatic than some forecasts suggest. Its most useful contrast: the article says METR’s Anthropic study estimated AI development was progressing roughly twice as fast as in previous years, while OpenAI found only a small increase on a similar measure.

Watch Desk analysis

What happened

The article, published on 20 November, discusses recent warnings from AI safety figures and the evidence behind claims that AI could rapidly accelerate its own development. It reports that METR researchers spent 11 weeks at Anthropic and concluded that progress on well-specified tests was roughly twice as fast as in previous years, mostly due to help from AI models. The article says Anthropic redacted key evidence, limiting outside assessment of that conclusion.

It contrasts this with OpenAI research released earlier in the week. On a metric similar to METR’s, the article says, OpenAI found only a small increase in its models’ research-progress rate, while reporting that employees used AI tools extensively in their work. Read the Mother Jones article.

Why it matters

These are not identical studies, and a comparison on one metric cannot settle how quickly AI capabilities will advance. But the contrast matters: claims of accelerating AI research depend on what is measured, which evidence is available, and how much of the underlying work outsiders can inspect.

The article also describes disagreements over METR’s independence and conflict-of-interest safeguards, including a disclosed job offer from Anthropic to a report co-author. Those are concerns raised in the article, not proof that the research was improperly influenced. Mother Jones also discloses that its parent company has sued OpenAI over copyright, which is relevant context for its coverage of the company.

Our read

The headline numbers are less useful than the evidence behind them. A reported twofold acceleration at one lab and a small increase in a similar OpenAI measure should prompt questions about methods and comparability, not a contest to crown the more dramatic claim. The best next step is more transparent, independently assessable research, not louder predictions.

What to watch

  • Whether Anthropic or METR make more of the underlying evidence available for scrutiny.
  • How the measures used by METR and OpenAI differ, and whether future work can compare them directly.
  • Whether independent studies find that AI meaningfully speeds up AI research beyond controlled tests.

Discussion spark: Should frontier labs be expected to provide independent researchers with enough access to test claims of AI-driven research acceleration, even when that means revealing commercially sensitive data?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.