The BaseEssayMETROctober 14, 2026

How an assessor could investigate evidence of acceleration from the automation of AI R&D

METR outlines how an assessor could test whether AI automation of AI research is speeding capabilities progress: internal benchmarks and indicators, attribution to AI versus other inputs, possible bottlenecks, and how reliably such methods would detect a dramatic acceleration.

Open the source ↗

Why it matters

The same thing, for different readers.

What it means for you

Asks whether AI building better AI is making progress speed up, something outsiders cannot currently check.

For lawmakers

Lists the data an outside assessor would need from a developer, such as internal capability measures and compute and labor over time.

For researchers

Separates acceleration in capabilities from its causes and asks whether the methods would catch a larger acceleration.

The question it raises

Raises who should be able to verify how fast a technology is changing itself.

In their own words

Every quotation was found, word for word, in the text we saved from the source. That shows the words are there; it does not make them true, or prove the speaker meant them as quoted.

From the source’s own text.

“Assessing this would involve analyzing capabilities of unreleased models, understanding the growth of key AI R&D inputs (labor, compute, data, and algorithms), and talking openly with relevant researchers about how AI is impacting their work.”

METRMETR on what an acceleration assessment would involve.

Found word for word in the source

“Overall, would the methods used reliably have identified much greater acceleration, if it had occurred?”

METROne of the limitation questions METR says the assessor should answer.