AI memory evals were measuring the wrong thing
A new benchmark reveals that most AI memory tests mix up two separate problems, hiding where agents actually fail.
A new benchmark reveals that most AI memory tests mix up two separate problems, hiding where agents actually fail.