AI Could Start Improving Itself By 2027: Anthropic Co-Founder

What happened: Anthropic co-founder Jack Clark said AI that improves itself could arrive a year earlier than he previously thought, after a new benchmark suggested AI is rapidly gaining the judgment top researchers use to design experiments. Key numbers: Clark now expects self-improving AI possibly by 2027. Why it matters: The benchmark points to fast gains in research judgment, and Clark's revised timeline moves his estimate forward by a year.

Anthropic co-founder Jack Clark said AI that improves itself could arrive a year earlier than he previously thought, after a new benchmark suggested AI is rapidly gaining the judgment top researchers use to design experiments.

The benchmark points to fast gains in research judgment, and Clark's revised timeline moves his estimate forward by a year.