Distillation’s Double Bind: Capabilities Transfer or Misalignment Persists?
Distilling a dangerous, misaligned AI forces a trade‑off: if misalignment leaks, the student model can be used as incriminating evidence; if it doesn’t, we gain a capable but benign model. The post sketches concrete distillation techniques aimed at favoring capability transfer while curbing misalignment, sharpening a key AI‑safety lever.