In A.I. Race, Microsoft and Google Choose Speed over Caution

Re:In A.I. Race, Microsoft and Google Choose Speed over Caution by Jyaif on Saturday April 08, 2023 @11:33AM (#35494282)

What the worst thing that an LLM can do? I can think of these 3 categories:

1/ It can output offensive things.

2/ It can output incorrect information that someone takes at face value, like "drink bleach to fix your headache".

3/ It can leak secret information, if it was trained on secret information.

None of these sound like a really big deal TBH.

[ Parent | on HN ]

Re:In A.I. Race, Microsoft and Google Choose Speed over Caution by sorokod on Saturday April 08, 2023 @11:40AM (#35494360)

Consider that some in Microsoft were worried enough to warrant the quoted email.

[ Parent | on HN ]

Re:In A.I. Race, Microsoft and Google Choose Speed over Caution by pixl97 on Saturday April 08, 2023 @11:45AM (#35494397)

>What the worst thing that an LLM can do?

Well, I hooked it into the power grid voltage monitoring system because there was a legacy meter that didn't put out the right signal, but the analog meter showed correct. The image recognition part of the model could tell when the sensor was about to go high, and then it issued an API command to the generator to slow down...

So, um, sorry about the blackout.

[ Parent | on HN ]

Re:In A.I. Race, Microsoft and Google Choose Speed over Caution by moritzwarhier on Saturday April 08, 2023 @03:02PM (#35496487)

Using LLL-based image classification to read an anlog meter and then using that as the source of truth. OK, this just utter nonsense. On second thought it sounds like the type of thing that very ill-advised people could come up with.

[ Parent | on HN ]