ngomez
Joined 12 years ago
Karma
73
Submissions
25
Member since
Jul 10
Recent submissions
Comment · 1 month ago
Comment · 4 months ago
You can grind some of the raw silicon out of a finished wafer but I don't think it'd be suitable to use in another batch of product. So instead of having the weights on the wafer like OOP was suggesting, hardware inference has been trending toward having a wafer of lots and lots of smaller cores with fast SRAM.
Comment · 4 months ago
Using the space of an entire wafer for one chip would result in extremely low manufacturing yields. Even with state of the art silicon cleanrooms, there will still be defects in parts of the output.
With CPUs and GPUs, chip makers can disable faulty cores and bin them as lower SKUs to get some yield out of it. But if you're using an entire wafer to embed weights, and a speck of dust causes a printing defect that makes the weights wrong, the entire wafer is worthless.
Comment · 6 months ago
The CloudFlare blog discusses that idea when they talk about having an "agent process" to hold cryptographic material, but they list drawbacks like having to develop two processes, implement a well-defined interface, and enforce ACLs. I'm not convinced that "developing two processes" is a reason not to do it, since the kernel is effectively just the second process now, but everything else makes sense.
It's unfortunate though since this is one thing I think Windows does decently well. The Windows crypto and TLS APIs do use a key isolation process by default (LSASS) and have a stable interface for other processes to use it [0]. I imagine systemd could implement something similar, but I also know that there are very strong opinions about adding more surface area to systemd.
[0] https://blackhat.com/docs/us-16/materials/us-16-Kambic-Cunni...
Comment · 6 months ago
Interestingly, Microsoft has been trying to get ahead of this for a couple of years now with their National Partner Clouds program [0], which they describe as:
> designed for scenarios where full ownership and operational independence from Microsoft is required
In France's case, Capgemini and Orange have a joint venture to operate datacenters that Microsoft runs Azure and Office on top of [1]. Moving away from Windows and Teams would still reduce their dependence on Microsoft substantially. But if the core goal is to reduce dependence on non-European suppliers, I would be wary of the French government buying services from "Bleu" when it's mainly Microsoft and a couple of consultancies in a trenchcoat.
[0] https://learn.microsoft.com/en-us/azure/azure-sovereign-clou...
[1] https://www.capgemini.com/news/press-releases/capgemini-and-...
Comment · 1 year ago
I've been trying uBO Lite myself for a few months, and anyone who uses YouTube will absolutely notice that it's worse at blocking. Lite tends to delay playback at the start of a video for as long as the blocked ads would've been, making the site feel slower, and once in a while an ad will slip past the blocker anyway.
Comment · 2 years ago
Are you thinking of the Windows 3.00 Working Model?
Comment · 2 years ago
Not the whole thing: a lot of the Windows org chart is still under Rajesh Jha in Experiences + Devices, or scattered around Azure with Scott Guthrie. But they've already been pushing Windows Copilot and Bing Ads and widgets, so I imagine the plan is more of the same.
Comment · 3 years ago
It's not just the kids growing up now. I'm sure plenty of millennials who watched YouTube when AudioSwap was a thing will recognize "Dreamscape" by 009 Sound System:
Comment · 4 years ago
I think the IPv4 "evil" bit [0] already does this :)
Comment · 5 years ago
Whether it's proper depends on who you ask but you can use the singular they.
Comment · 5 years ago
Could you elaborate? AMD spun their fabrication arm off into GlobalFoundries a while back, and they seem to be doing okay. I thought that would be a good precedent for Intel divesting its fabs too, but I'd love to hear any counterpoints to that.
Comment · 5 years ago
Comment · 6 years ago
>if a Twitter user was logged in with a dongle but the attacker had access via social engeneered remote desktop access a dongle still could mean access to private data
It depends on the dongle. YubiKeys and similar devices require the user to physically touch/tap it to enable U2F auth, and it automatically powers down after a timeout to prevent remote desktop attacks.
I would hope Twitter already had this kind of setup, but their blog posts about this are all targeted at a more general audience, so I doubt we'll get that kind of detail anytime soon.
[0] https://openai.com/policies/how-your-data-is-used-to-improve...