News for hackers. Stuff that matters.
Re:A.I. Companies Agree to Safeguards After Pressure from th... by candiodari on Friday July 21, 2023 @12:08PM (#36815543)
This video shows the issue with AI "safety": it kills functionality.

https://www.youtube.com/watch?v=PqcNqyd13Kw

The JSON example is the most damning. The model refuses to generate basic code because it refuses to "store personal information", which is not allowed due to laws such-and-such. This is FICTIONAL personal information (name + address).

But virtually every example is a lot better without safety measures.

If this is what AI safety means we may as well call it quits right now.

Re:A.I. Companies Agree to Safeguards After Pressure from th... by naveen99 on Friday July 21, 2023 @04:16PM (#36819072)
It’s not like human bureaucracies are any better. Most of the time they act like brick walls, it’s similar reasons.