Xkcd: Tasks (2014)

Re:Xkcd: Tasks (2014) by uejfiweun on Thursday June 01, 2023 @02:47PM (#36155559)

Although we've come a long way since 2014, I'm still waiting for a general purpose, high performance "image recognizer" that I can just import as a library and run locally. It's probably not even possible with current computer architectures. I do predict that in 10 years we will actually see some more low-cost energy efficient specialized chips built into computer designs from the ground up.

[ Parent | on HN ]

Re:Xkcd: Tasks (2014) by brucethemoose2 on Thursday June 01, 2023 @03:20PM (#36156044)

We are already there. Phone SoCs already have AI blocks that are pretty much purpose built for object recognition and image enhancement.

Intel has had this for some time too. They use it for stuff like facial recognition unlocking in Windows.

It just doesn't really feel like it because they aren't very accessible to users and hackers. The next gen blocks from AMD/Intel are supposed to be more accessible and generalized, but we shall see.

[ Parent | on HN ]

Re:Xkcd: Tasks (2014) by hnux on Friday June 02, 2023 @01:33AM (#36161406)

test

[ Parent | on HN ]

Re:Xkcd: Tasks (2014) by marssaxman on Thursday June 01, 2023 @04:11PM (#36156703)

> I'm still waiting for a general purpose, high performance "image recognizer" that I can just import as a library and run locally.

I wonder what additional criteria you must have for such a tool, because that description sounds like something which has been possible for years already. For example, I wrote this back in 2017:

https://github.com/jbruestle/plaidvision

> in 10 years we will actually see some more low-cost energy efficient specialized chips built into computer designs

Earlier this week, Intel announced that all Meteor Lake CPUs will include a Movidius VPU coprocessor.

[ Parent | on HN ]