METR report finds Anthropic internal AI doubled R&D pace

80000hours.org10 min read

An independent AI risk assessment group, METR, spent eleven weeks analyzing Anthropic's internal data and concluded that the company is improving its AI at roughly twice the speed of a year ago, primarily because the models are assisting in their own development. While Anthropic had set a benchmark line for when two years of progress occur in one, its recent filings suggest it may have crossed that threshold, raising urgent concerns about recursive self-improvement and the timeline toward superintelligence. Open article →

More on this topic