Why MSFT Copilot+ and AI PCs are the final nail in the coffin of open computing
venturebeat.com
venturebeat.com
IBM never imagined the PC would be taken away from them, the vertical integration market we have been slowly going back to, was the common way of selling 8 and 16 bit computers, until PC's came to be, and Compaq took advantage of it.
The thin razor margins of PC components, the commoditization of computing, are most likely the main drivers for going back to that model.
Employees are going to have to start being very reticent to communicate in electronic form, likely will need to counter deploy AIs to generate communication and "work" that maximizes their appearance of productivity.
All of this with a magic handwave to absolve away any AI racial profiling or other HR exposures the business may have. "The great AI god said so" will be cover for any and all malfeasance.
AI is a deep war on labor by management. Not just in automation, which is what everyone is concerned with currently. That at least has positive efficiency/economics associated with it. The true war on labor will be a totalitarian one, and with it, everything necessary to do it at the state level at the drop of a hat, or a Gulf of Tonkin.
Your boss getting a copy of your Recall timeline is a thing that can't not happen. It's too valuable. It may be the strongest value prop of the feature.
It's amazing how industrially stripping privacy from people is "valuable"
So I don’t see why Recall is being cast as some disastrous new spyware that is going to ruin computer privacy. If you’re using a company device, you already might have effectively no privacy.
------
Data collected: All of it.
Uses it for: Whatever they want. Advertising, sells to your insurance company, sells to the government, trains AI models.
Price: Free
Kagi
----
Data collected: Your signup email
Uses it for: Giving it to the government if legally mandated
Price: $10/mo
I think a large double digit percentage of average people (in countries where $10/mo is cheap at least) would not choose Google. Maybe I'm being too optimistic though.
When it was just an application, companies could refuse to install it. They could say, nah, trust in our employees is part of our culture. When it's built in to everybody's OS, it becomes much harder to refuse to gather and use everybody's data. It's like the heat detector under everybody's desk: ostensibly it's there to measure the need to run A/C... but as long as it's there why not use it as an absenteeism detector?
Plus, the data gathered is much more extensive, and much more searchable, with the proposed Microsoft solution.
10K+ papers in two decades, https://scholar.google.com/scholar?q=lifelogging
Why? If anything, hardware acceleration should make it easier to run AI locally.
> February 1995
* You can of course run Microsoft's AI on Windows on this hardware.
* But I'll bet you can also use the NPU and the drivers that Windows has with different AI software.
* And sooner or later, I'm very interested to see Linux drivers for these NPUs so we can have nice hardware-accelerated AI without needing any proprietary software.
So I'd say that FOSS is an appropriate reply, but there is room for a continuum of answers.
That said, Microsoft has made it abundantly clear what direction that they are going in and it is an invasive consumer-hostile anti-privacy direction.
This will not change. For that reason, open source solutions such as Linux or better yet OpenBSD (in ideal only, I don't run it and wouldn't expect anyone else to) are the only chance of retaining end-user control and privacy.
I say 'chance' because it's not out of the realm of possibility that Microsoft, Google or others will social engineer consumer-hostile changes into the Linux ecosystem -I have no illusions about that. But at least consumers have a chance and that is something that can not be said of Windows.
It'd be nice if the ARM-based SoCs had the same openness expectation that x86 historically has had.
I could load my blog with it on a modern version of Firefox. It was not fast, but I could still do it!
Essentially we are at peak personal computing: End-users are just as well served by the laptop they had 10 years ago and the mobile phone they had 5 years ago, as they ones they have today.
Don't get me wrong; LLMs are useful and exciting and so are other forever upcoming things from CS - say microkernels ... but that doesn't mean they can justify an arbitrary large valuation. Ultimately these things are worth no more than the value they can generate for their end-users.
Right now big tech is betting it all on LLMs. It's a perfect match for big tech - requires massive amounts of capital and just a tiny number of highly paid specialists.
But after having spent billions and their best minds. This is what they come up with: An app running on your PC that takes a screenshot of your desktop every n seconds and feeds that into an LLM that you then can interact with. Requiring completely new hardware and massive amounts of computing resources.
So that you can go - "I looked at a brown leather bag earlier - what website was it at?" and similar.
All has value I am sure, but also probably limited value. And could have been supported much more straight forward and with much less computing power through traditional software.
They would be less expensive and you could upgrade then too over the years. Why spend 3k on a new one when all you need is a new GPU for 700.
But regarding the article or rather M$ PC"+", my 14900k has AI features, apparently. A NPU can come bundled with the CPU, doesn't have to be a separate component.
I don't think the article is correct, it's just trying to ride the announcement. Category spam.
Apparently, this is done all on-device, with no phone home, but that's only one concern addressed.
Just like people instinctively go to google for a search due good experiences with their search results in the past, new habits will be formed and people will use gemini, chatgpt, copilot etc.
So I wouldn't say any of these new AI platforms will be rendered worthless due to advancements in the field.
Enterprise knowledge context and a platform to currate that and feed it to LLMs, on the other hand...
Teams is not winning due to technical competence. It is Good Enough and bundled with the other Office products a corporation is already going to license.
With the Scarlet Johansson voice seem to be the same, too much noise in too much media at once for something that should had been obvious from the start, and it is not even the start of it.
It may or not be the next step in a progression? Ok. That increases what players are in the game? Maybe. But the game was already set, and with the main players of today.
There are more things to be scared about, with AIs and those players (and some that play the elephant in the room role), but that is not what they are pointing out.
(I do grant, we might need to get NPU drivers in good shape; not sure what the state of that is right now.)
OpenAdapt has similar functionality, except:
- it's open source
- it only records when you explicitly tell it to
- it has multiple PII/PHI scrubbing providers built in (see https://github.com/OpenAdaptAI/OpenAdapt?tab=readme-ov-file#...)
- the purpose for recording is to automate tasks in desktop apps
- it's cross platform (Mac and Windows now, Linux coming soon)
Full disclosure: I'm the primary author. Contributions and feedback welcome!
This is obviously crap/inaccurate/a misrepresentation, windows has always been a completely closed prorietary s/w platform.
The only "openness" (since we're not allowed to say "free software" on HN) has been the ability to load unverified binaries from unknown developers. In other words, a complete security nightmare of a s/w platform.
"Enterprise" has been "beholden" to M$ for decades. I would expect the M$U graduates in IT everywhere to embrace this abomination like they've embraced all the ones that came before it, that is with gusto 8-)