Company Updates

Anthropic's AI development speeds up 2x in a year, report says

Share
Anthropic's AI development speeds up 2x in a year, report says

Anthropic is developing its AI systems twice as fast as it was a year ago, according to an analysis by Model Evaluation and Threat Research (METR). The report comes amid CEO Dario Amodei's calls to slow down AI progress.

TL;DR

  • Anthropic's AI development speed has doubled in the past year, per an independent report.
  • The acceleration is driven by AI-assisted improvements, raising safety and control concerns.
  • Anthropic plans to go public later this month, potentially the biggest IPO ever.

What happened

Anthropic invited METR to review its internal data and interview staff as part of its preparation for a Wall Street debut. METR found that Anthropic's AI development pace has accelerated significantly, with the company improving its models' capabilities around twice as fast as it was a year earlier.

The report attributes much of this acceleration to AI's role in its own improvement. Anthropic's Claude chatbot, for instance, has been helping develop newer, more advanced versions of itself. METR's assessment was based on analysis of Anthropic's internal data provided over an 11-week period, as well as employee interviews.

Anthropic co-founder Sam McCandlish acknowledged the findings, stating that AI is speeding up research but it's hard to quantify exactly how much. He also mentioned that the company plans to release its own analysis of the speed of AI progress in the coming weeks.

Why it matters

The rapid acceleration of AI development raises significant safety implications, as it determines how quickly the risks from advanced models increase. This is particularly concerning given that there appear to be no major bottlenecks, such as restrictions on computing power or data, holding back Anthropic's AI progress.

The use of AI to improve itself could start to compound the acceleration, which could be especially worrying. As Amodei noted in his September essay, AI advances driven by recursive self-improvement could outrun our ability to understand and control these systems.

For developers and startups, this news highlights the importance of safety measures and external evaluations in the rapid development of AI. For investors, it underscores the potential and risks associated with Anthropic's upcoming IPO, which could be the biggest ever.

Key facts

  • Anthropic's AI development pace has accelerated to around twice as fast as it was a year earlier, per METR's analysis.
  • The acceleration is largely driven by AI's role in its own improvement, with Anthropic's Claude chatbot helping develop newer versions of itself.
  • METR's assessment was based on analysis of Anthropic's internal data provided over an 11-week period, as well as employee interviews.
  • Anthropic plans to go public later this month, potentially the biggest IPO ever.
  • Anthropic CEO Dario Amodei has called for the AI industry to slow its work so safety measures can catch up.
  • Anthropic co-founder Sam McCandlish acknowledged the findings and plans to release the company's own analysis of the speed of AI progress in the coming weeks.
  • METR is an independent nonprofit that has conducted evaluations on model capabilities for Anthropic, OpenAI, and other labs for years.
  • The report was peppered with caveats, noting that the acceleration might only be limited to well-specified tasks and that there are no major bottlenecks holding back Anthropic's AI progress.

Context

This news comes amid growing debate over the potential for AI models to elude human control with catastrophic consequences. It also highlights the importance of external evaluations and safety measures in the rapid development of AI.

Anthropic's upcoming IPO, which could be the biggest ever, underscores the significant investment and interest in the AI industry. However, it also raises questions about the balance between innovation and safety in the development of advanced AI models.

The use of AI to improve itself, known as recursive self-improvement, is a key concern. As Amodei noted, this could outrun our ability to understand and control these systems, highlighting the need for careful pursuit of such advances.

Topics

Related coverage

Join the discussion

Have a take on this story? Weigh in with our community on Facebook.

💬 Discuss on Facebook →