How (un)reliable are AI agents?

Consistency matters more than average accuracy in safety-critical domainsTechnology, Artificial intelligence