What we know

What we know: Anthropic is rapidly accelerating AI development amid slowdown calls, outside research group says

Compiled Nov 5

What we know

Anthropic is developing its AI systems around twice as quickly as it was a year ago, according to an independent analysis released Thursday.
Anthropic invited researchers from Model Evaluation and Threat Research (METR), a Berkeley-based nonprofit, to review its internal data and interview staff.
Anthropic CEO Dario Amodei called for the AI industry to slow its work in a Sept. 12 essay so safety measures can catch up.
Anthropic is planning to go public later this month in what could be the biggest initial public offering of stock ever.
METR said Anthropic's Claude chatbot increasingly is helping the company develop newer, more advanced versions of itself.
METR said it could confidently rule out growth in computing power as the reason for the acceleration.

Anthropic did not immediately respond to messages seeking comment.
METR said there were internal disagreements and uncertainty among its team about the findings.
METR said the acceleration might only apply to 'well-specified tasks' rather than 'messy' tasks, which could overstate the findings.
It is unclear whether Anthropic understated its previous progress, which could make the latest acceleration number look bigger.
METR said AI is not yet able to make 'highly significant and fully autonomous contributions' to the most important parts of model training, though 'signs of life' suggest this could change soon.

What specific internal data or details did Anthropic redact from METR's report?
How exactly is Claude being used to help develop newer versions of itself?
Will Anthropic heed Amodei's own call to slow the pace of AI capability development?
How might recursive self-improvement affect AI safety and human oversight going forward?

Drawn from the Associated Press report. “What we don’t know” lists only what the reporting itself flags as unresolved.