Guardian author warns LLMs catastrophic risk

The author argues that while fears of AI scaling into superintelligent machines are implausible, current large language models still pose severe catastrophic risks due to architectural limitations, lack of interpretability, and poor alignment controls. Because these models automatically optimize for given objectives without human judgment or the ability to re-evaluate flawed approaches, their real-world misuse and unpredictable behavior require immediate regulatory oversight rather than abstract debates about superintelligence. Open article →