MATS and Redwood internship research experience sharing
Start here
Researchers doing empirical AI safety research at MATS and Redwood Research shared that the key to accelerating experimental iteration is building predictable, falsifiable models of the problem and continuously updating them with data. Their work involves fine-tuning-based misalignment audits and capability assessments for sabotage. Using LinuxArena as an example, the author noted that a weak monitoring model can capture 80% of attacks under a small audit budget, but if there are only 5 attack samples, one should not directly update a 50% prior to 80%.
Generated by AI from reports · Updated 2026-10-14
Progress timeline
- LessWrongLessons in AI safety research I learned at MATS and Redwood
A researcher doing empirical AI safety research at MATS and Redwood Research says the key to accelerating experimental iteration is building predictable, falsifiable models of the problem and updating them with data. Their work includes fine-tuning-based misalignment auditing and capability removal evaluation. Using LinuxArena as an example, the author notes that a weak monitoring model captures 80% of attacks under a small audit budget, but with only 5 attack samples, one should not directly update a 50% prior to 80%.