News for hackers. Stuff that matters.
Re:Kenyan moderators decry toll of training of AI by version_five on Thursday August 03, 2023 @12:35PM (#36988147)
AI doesn't extrapolate, when you're training it your showing it data and saying "everything you'll see will look like this". And when you find edge cases you need to show more data that captures them, etc. For long tail events, you're basically just recalling what a person did, if that makes sense. Look at self driving cars and the seemingly infinite hours of footage, it's to try and capture every long tail thing that will ever happen (which scales the way you'd expect it to).

Anyway, unsupervised learning takes care of say 80%, the remaining 20% is really locking up cheap labor in an algorithm.

There are no comments on this comment yet.