Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Intel, Microsoft discuss plans to run Copilot locally on PCs

arstechnica.com

Post reply on HN
Print view View original

11–20 of 25 posts

Re: Intel, Microsoft discuss plans to run Copilot locally on PCs

#11

Post by zamadatix »

papichulo2023 wrote: post #9

Well... I hope this open to other parties or an antitrust may be on their way. Can you image getting the ai subscriptions money and just let the user paid for the operation cost? Google, OpenAI and endless others wont be able to compete.

Access to the NPU is done through DirectML which is open to anyone to use. The hard part is making useful AI people will pay for that runs on consumer devices themselves without a race to the bottom on the same few use cases (e.g. local translation). There may eventually be some bundling type concerns with the service being pushed by the OS over time though.

Re: Intel, Microsoft discuss plans to run Copilot locally on PCs

#13

Post by bottom999mottob »

rolph wrote: post #5

why would it be better to host an AI on your PC, rather than in its own local box, with ability to access permitted aspects of your PC ? i can tell you why a discreet box is better, you can firewall it, or physically unplug it when its not functioning in a desireable manner.

Same reason people virtualize dozens to hundreds of machines in one box, hardware efficiency.

I don't understand this sentiment.

Re: Intel, Microsoft discuss plans to run Copilot locally on PCs

#14

Post by rolph »

bottom999mottob wrote: post #13
rolph wrote: post #5

why would it be better to host an AI on your PC, rather than in its own local box, with ability to access permitted aspects of your PC ? i can tell you why a discreet box is better, you can firewall it, or physically unplug it when its not functioning in a desireable manner.

Same reason people virtualize dozens to hundreds of machines in one box, hardware efficiency. I don't understand this sentiment.

AI integrated into your PC, means cooperating with an AI to use your PC. this also restricts your[sic] AI.

your AI should go with you and saddle into task specific hardware, be that a home automation bus, LAN dock, automobile, or roomba-iod R2 like unit.

your AI should be your wingman and have your back. but being an agent capable of deciding without your input, there should be a way of training acceptable bounds into your AI. [yes that is looking like a mobile]

likewise there should be physical restraint when it gets a "bad idea"

Re: Intel, Microsoft discuss plans to run Copilot locally on PCs

#16

Post by mock-possum »

rolph wrote: post #5

why would it be better to host an AI on your PC, rather than in its own local box, with ability to access permitted aspects of your PC ? i can tell you why a discreet box is better, you can firewall it, or physically unplug it when its not functioning in a desireable manner.

Why would anyone strap a couch to the top of their car, when it would be better to buy a new pickup truck and transport it in the back of that?

Re: Intel, Microsoft discuss plans to run Copilot locally on PCs

#17

Post by petre »

amadeuspagel wrote: post #6

For a long time, before Apple announced that they'd integrate Gemini into their devices, people posted fan fiction on HN about how Apple was going to bring local AI to the masses. Now Microsoft is actually doing it, and ...

Gemini is a poor choice, rather useless. They recently bought some AI company, so scratch that.

Re: Intel, Microsoft discuss plans to run Copilot locally on PCs

#18

Post by wtallis »

rolph wrote: post #14
bottom999mottob wrote: post #13

Earlier quoted context omitted.

Same reason people virtualize dozens to hundreds of machines in one box, hardware efficiency. I don't understand this sentiment.

AI integrated into your PC, means cooperating with an AI to use your PC. this also restricts your[sic] AI. your AI should go with you and saddle into task specific hardware, be that a home automation bus, LAN dock, automobile, or roomba-iod R2 like unit. your AI should be your wingman and have your back. but being an agent capable of deciding without your input, there should be a way of training acceptable bounds int…

You're talking like we already have artificial general intelligence that can autonomously initiate actions, when at best what can run on a single consumer device is a very small, dumb LLM. So none of what you're trying to worry about makes any sense for this discussion.

Re: Intel, Microsoft discuss plans to run Copilot locally on PCs

#19

Post by throawayonthe »

papichulo2023 wrote: post #9

Well... I hope this open to other parties or an antitrust may be on their way. Can you image getting the ai subscriptions money and just let the user paid for the operation cost? Google, OpenAI and endless others wont be able to compete.

[dead]

Re: Intel, Microsoft discuss plans to run Copilot locally on PCs

#20

Post by wtallis »

zamadatix wrote: post #11
papichulo2023 wrote: post #9

Well... I hope this open to other parties or an antitrust may be on their way. Can you image getting the ai subscriptions money and just let the user paid for the operation cost? Google, OpenAI and endless others wont be able to compete.

Access to the NPU is done through DirectML which is open to anyone to use. The hard part is making useful AI people will pay for that runs on consumer devices themselves without a race to the bottom on the same few use cases (e.g. local translation). There may eventually be some bundling type concerns with the service being pushed by the OS over time though.

Forget about "useful AI people will pay for"; the limitations of these NPUs and the immaturity of the software stacks necessary to target them means it's hard to get anything running well enough to be a cool tech demo. Good enough that people would use it for free is still a ways off, and good enough to be worth paying extra for may not even be possible on the NPU hardware Intel and AMD are currently shipping.
Post reply on HN