Andrew Curran says full recursive AI self-improvement may arrive in 2027

Andrew Curran, on X, says Anthropic co-founder Jack Clark now considers full recursive self-improvement likely possible in 2027, with creativity as the last missing piece.

Curran says the loop could close in less than six months. He cites OpenAI's unnamed internal model Aeon, which he says produced recent math results in one attempt each, averaging three hours of thinking.

These are Curran's claims as reported on X. The post does not include Clark's original statement or any OpenAI documentation of Aeon.