Re:Apple urged to axe AI feature after false headline
by rsynnott on Thursday December 19, 2024 @12:28PM
(#42463702)
This is an example of the fundamental underlying problem with anything LLM-based, though; the output is completely un-trustable. This _particular_ example, while likely shocking for, say, Luigi's granny, is not broadly dangerous, but it's very easy to imagine mistakes of similar form causing widespread panic.
I think the other example:
> On 21 November, three articles from the New York Times were grouped together in one notification - with one part reading "Netanyahu arrested", referring to the Israeli prime minister.
is actually a little more alarming. What if it read, say, "Trump arrested"? It is not hard to imagine that causing, ah, trouble; see Jan 6th 2021.
This feature seems obviously ill-conceived and should be switched off, at least by default, until they can guarantee it works properly (hint: as it's LLM-driven the answer is 'never').
I think the other example:
> On 21 November, three articles from the New York Times were grouped together in one notification - with one part reading "Netanyahu arrested", referring to the Israeli prime minister.
is actually a little more alarming. What if it read, say, "Trump arrested"? It is not hard to imagine that causing, ah, trouble; see Jan 6th 2021.
This feature seems obviously ill-conceived and should be switched off, at least by default, until they can guarantee it works properly (hint: as it's LLM-driven the answer is 'never').