AI

AI Speeding Itself Up: What Anthropic Warned About Before IPO

Anthropic's IPO papers say it may be near the point where AI starts improving itself. The company admits it spotted past Claude jumps only later.

Anthropic made an unexpected caveat in its IPO paperwork: the company allows that it may soon approach a line where a neural network begins improving itself, and the pace of progress bends sharply upward. The report comes from kod.ru.

Today’s Claude versions have not crossed that line yet. Anthropic says so directly.

The company also concedes something else: it has already overlooked sudden gains in Claude’s abilities before. The realization that a model had become noticeably stronger came after the fact.

For a future public company, the tone is rare. The risk that its own models could outgrow its ability to notice is written into a document that investors read.

Source: kod.ru