AISI’s findings show that Mythos isn’t significantly different from other recent frontier models in tests of individual ...
For us to trust it on certain subjects, researchers in the growing field of interpretability might need to learn how to open ...