AI models may not be so opaque after all

Researchers have discovered that it's possible to interpret the inner workings of complex AI models, offering a glimmer of hope for those concerned about AI safety.