We rebut alarmism about catastrophic AI risks and analyze flawed claims about how to make it safer.

Through our research and explanatory articles, we’ve shown what goes wrong when we fall for AI hype and how to avoid doing so.

Evaluating models on benchmarks was good enough in classic ML, even if imperfect. But in the LLM era, this has broken down. Chaos reigns.

The two of us are at Princeton’s Center for Information Technology Policy. We regularly contribute our technical expertise to AI policy debates.