A new report reveals AI is accelerating Anthropic's research pace

An independent investigation reveals that artificial intelligence is rapidly speeding up Anthropic's internal research and development.

2 min read

A new report reveals AI is accelerating Anthropic's research pace

A comprehensive investigation published Thursday by the artificial intelligence safety research organization METR provides a rare look inside Anthropic. The findings show that artificial intelligence is dramatically speeding up the company's research and development, raising important questions about the future pace of technological advancement.

Key facts

Investigation duration
11 weeks
Productivity multiplier
12 times
Model 2 completion
July
Model 3 completion
August-September
Model 4 completion
September-October

Unprecedented productivity gains

According to the report, Anthropic's models are improving roughly twice as quickly as they were a year ago. Benchmark scores indicate that the lab has achieved roughly two years of historical progress over the last year alone. METR researchers expressed confidence that this acceleration was not simply the result of adding more computing power to train the systems. Instead, the majority of the progress is driven by artificial intelligence itself, with employee surveys showing that staff using the company's most capable internal model reported being roughly twelve times as productive.

Unreleased models and heavy redactions

Over the past six months, Anthropic has finished training three unreleased models codenamed Model 2, Model 3, and Model 4. While the lab granted METR access to confidential research and internal interviews, it also exercised its right to heavily redact the final public report. Anthropic cofounder Sam McCandlish defended the redactions on X, stating that sharing the specific drivers of the speedup would be commercially sensitive and could risk handing out a roadmap for competitors to move faster.

Safety and transparency concerns

The report has sparked intense discussion among researchers regarding recursive self-improvement, where artificial intelligence systems autonomously develop increasingly powerful successors with minimal human involvement. Critics have pointed out a perceived contradiction between Anthropic's public advocacy for slowing down artificial intelligence development and its own rapid progress, renewing calls for greater democratic oversight and transparency from frontier labs.

Source: Original announcement

Worth a read?

Report

Comments