- AI firm Anthropic has warned that humans risk “losing control over AI systems” if a concerning trend with its Claude model continues.
- The startup revealed Claude now writes over 80 per cent of its own code, a significant increase from less than 10 per cent in February last year.
- Anthropic claims that code written by Claude is roughly at parity to human-written code, and is expected to be much better within the next year.
- In a blog post, Anthropic’s Marina Favaro and Jack Clark said this “points to an AI system capable of fully autonomously designing and developing its own successor”, but added that “full recursive self-improvement also might increase the risks of humans losing control over AI systems.”
- Anthropic said that it would “likely be a good thing” to slow down AI development to address potential implications, but acknowledges this would require broad industry consensus.
IN FULL
