3 ms·
Look up AI safety and THE aligment problem. This isnt a matter of training data quality.
by Xss3 11mo ago
Look up AI safety and THE aligment problem.
This isnt a matter of training data quality.
- skissane 11mo agoWe don’t expect 100% reliability from humans-humans will slack off, steal, defraud, harass each other, sell your source code to a foreign intelligence service, turn your business behind your back into a front for international drug cartels-some of that is very low probability, but never zero probability-so is it really a problem if we can’t reduce the probability to literally zero for AIs either?
- Xss3 11mo agoHumans have incentives to not do those things. Family. Jail. Money. Food. Bonuses. Etc. If we could align an AI with incentives in the same way we can a person then youd have a point. So far alignment research is hitting dead ends no matter what fake incentives we try to feed an AI.
- aswegs8 11mo agoCan you remind me of the link between alignment and writing accurate documentation? Honestly don't understand how they are linked.
- Xss3 11mo agoYou want the ai aligned with writing accurate documentation, not aligned with a goal thats near but wrong, e.g. writing accurate sounding documentation.