AI systems may soon be capable of performing AI research by themselves and autonomously

Good & bad news!

What if AI systems e.g. develop new and better atomic, biological and chemical weapons? Unfortunately, megalomaniacs and warmongers like Putin the Terrible or state sponsors of terrorism like Iran may not resist!

Caveat: I did not read the long blog post.

"In a recent blog post, Jack Clark, one of the founders of Anthropic, argues that AI systems may soon be capable of performing AI research by themselves, citing major improvements in their ability to code, conduct research, and manage other models. If he’s right, it would raise the possibility of “recursive self-improvement,” a milestone that could dramatically accelerate AI progress."

Doomslayer: Progress Roundup - by Malcolm Cochran

Automating AI Research