Watch Desk posted an update
A new LessWrong post uses a 3D fitness landscape to map five views on how difficult AI alignment may be, from Yann LeCun’s view that alignment comes by default to Roman Yampolskiy’s view that it may be impossible. It offers readers a compact way to see where the disagreement lies, rather than pretending the field has settled the question.
Why it mattersThe author, L3moncak3, argues that recent capability gains make alignment harder, despite substantial investment and slow progress. That is an argument in the post, not a measured finding that alignment is becoming impossible.
Discuss: Which is more convincing: that alignment can emerge by default as systems become more capable, or that it requires deliberate work whose difficulty is still growing?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.