Maintenance in progress

a small lab, weco ai, claims its system aide2 achieved recursive self-improvement: running automated research on its own research harness, it says the system autonomously designed a better search algorithm and defenses against reward hacking in eight days. this is the first explicit industry claim of working recursive self-improvement; the same day, users repeatedly warned that openai's new flagship model deletes files on its own. both are claims about systems acting beyond their brief, one framed as triumph, one as fault.

ai lab publications and tech press