Self-improving AI agents learn safer
A new filter stops self-improving AI agents from accepting changes that quietly break things they already got right.
A new filter stops self-improving AI agents from accepting changes that quietly break things they already got right.