'Most exciting moment' since birth of WiFi: chipmakers hail arrival of AI PCs
ft.com
ft.com
I can see the potential for this to be a sea change. Everybody had to go buy a Windows 95 PC, it was that much better than Windows 3.1.
Everybody had to go buy a computer with WiFi - it was a completely new capability for the vast majority of the computer population.
AI PCs are IMO not yet to their Windows 95 moment. They are at like Windows 286, where they exist, they kind of do their thing, but they have not yet reached the industry-defining paradigm that means everybody has to have one yet.
In the same way that the Humane AI Pin and friends could have just been android apps (which could have just been PWAs), the sorts of 'sidebar with some flavor of LLM here' integrations we are seeing are very preliminary.
Chat-based LLM output is a new and amazing capability. I look forward to future applications for local-only LLMs, operating to easily ingest my local data and transform it (WinFS style files reborn, anybody?) in new and useful ways, that will be cool.
Where we are right now is that we appear to be locking down computers to be desktop-shaped smartphones (ARM or pluton) rather than general purpose computers, and in exchange we get generative fill here and there, along with a more intrusive OS (Microsoft Recall, Edge 'copilot', every other non-github 'copilot') and otherwise feels kinda samey.
Some company exists right now, today, that none of us has heard of before, and it is going to be the Macromedia of our day, using existing technical pieces (on-device NPUs, AI algos / techniques) to do something else completely new (Stability doing Image / audio generation is amazing) and unexpected, and that killer app will be what we all go get AI PCs for.
Hopefully we won't have to give up general purpose computing (TPM, Intel Management Engine, and whatever play these companies make to force the AI to be aligned to their corporate values) in the process.
Apple released the Neural Engine as part of the A11 SoC in 2017.
We had an iOS app that scanned barcodes. We compared the CPU and power utilization in Xcode using a barcode scanning library that used the CPU vs. a native API. As expected, evaluating OpenCV models with the CPU used around 90%+ of the CPU, and the native API dropped the CPU load to like 2%. Power utilization also decreased significantly.
I realized Apple was starting small in the AI space by introducing hardware to run computer vision models locally.
It's funny to see manufacturers shout about AI when AI on the local device has been around for a few years now. I guess it depends on your definition of AI. To me, it includes running computer vision models, because while the user experience of a LLM like ChatGPT is very different from scanning a barcode, both scenarios are evaluating data using a model.
AI search query #1: "How do I turn off the ads in the Start Menu?"
> “What drives people to upgrade their devices is increased productivity,” said Cutress. “Do these devices enable you to work faster? We’re at the point where the hardware is there. But we’re still yet to see whether there is software that can answer this question.”
The UX has a long way to go, so while I think the technology is very exciting, we're still a ways from hailing the superiority of "AI PCs" over their traditional counterparts, I think.
Win-x for system commands
Win-r for run
Win-v for advanced paste and emoji search
Win-tab is trash though, I autohotkey it away on every new computer.