FlagThis

Publisher Description

As machine learning transitions from experimental labs to the backbone of critical infrastructure, the "alignment problem" becomes a primary systemic vulnerability. Brian Christian provides a masterful investigation into how flawed objective functions and reward hacking can lead to catastrophic failures, making this essential reading for security engineers and researchers tasked with building resilient, predictable, and safe autonomous systems.