Not the mother, the breadcrumbs
If you follow the link, it's about the breadcrumb aspect, not the parenting.
C.
3533 publicly visible posts • joined 21 Sep 2011
Hi -- those are good observations. The point we wanted to make is that, in our opinion, SpaceX needs to reassure everyone that there isn't a deep-rooted systemic problem that will potentially affect all future launches, and that this is a one-off that can be identified and corrected.
Happy to make that clearer.
C.
Yeah, we know, we know. We just forgot to think it through, sadly. Although nested virtualization is a thing, you're right, that's not an accurate test.
Mea culpa. Now that we've read the documentation, we've tweaked the piece to better explain that the testing system involves a pool of bare metal equipment that can't be virtualized and easily migrated. If you spot something wrong like this, don't forget to drop corrections@theregister.com a note and we'll do our best to fix things up.
C.
FWIW the Group-IB report does take a few reads to understand, and we've tried making our summary of it more clear. It does start with the abuse of the Fortinet VPN to gain RDP access to the failover. And then it all falls down.
But to get into the backup servers, the CVE was exploited, and that has credential stores that are useful for other parts of the network. Yes, the network is compromised with or without the CVE exploited. The Veeam bug just seems to make the ransomware deployment easier. AIUI.
C.
You're not missing anything. Emissions are going up, making the goal of net zero harder, tho Google reckons the increases will be offset by greater efficiencies driven by AI. As the report states:
"As we further integrate AI into our products, reducing emissions may be challenging due to increasing energy demands from the greater intensity of AI compute"
C.
I get that in theory ensuring rules are set by elected lawmakers and adjudicated by judges is a fair thing.
But in practicality, can the court system really handle the load? Can watchdogs really afford to enforce rules if they're just going to be challenged in court every time? Can we expect lawmakers to pass detailed and technically correct legislation all the time? Isn't deferring to scientists and experts a fair thing, too, with parameters set by Congress?
Bear in mind the Supremes got nitrous oxide confused with nitrogen oxides in their separate Ohio v EPA ruling :( Baking those kinds of mistakes into law could have quite disastrous effects.
C.
It's like Microsoft took the i made this meme and absorbed it as SOP.
C.
I dunno if you've noticed around here but we tend to bend the language to breaking point, and we kinda enjoy it. It's fun to test the limits and play with writing. Yes, there are rules to follow. Apostrophes, plurals, commas, and so on.
But you're talking about a publication that writes about bonkers boffins, naughty Norks, and enormo electronics slingers. Masses and deorbit are par for the course.
C.
FWIW to clear up any confusion: AliveCor's Apple smartwatch app used the continuous heart-rate data from the Watch to identify potential signs of danger, and would tell the wearer they should take a test with proper ECG equipment to be sure (for the reasons you give).
And AliveCor primarily sold FDA-approved ECG monitoring devices, including an FDA-approved Watch wristband that did just that. So it's not like AliveCor was trying to do full ECG with just the Watch's built-in senor. Without the continuous feed, it couldn't even properly warn wearers of potential danger, and had to pull its app, hence the antitrust suit, or so it says.
The watchOS changes, as far as AliveCor is concerned, caused the biz to offer less to the market.
C.
As it's 60 days after _activation_. When the phone connects to its cellular network for the first time. Whether carriers would be OK with that is another matter, but right now, it's from activation.
The proposed rule, according to the FCC, is specifically "requiring all mobile wireless service providers to unlock mobile phones 60 days after the device is activated with the provider."
C.
OpenAI's CEO claimed in Feb that his super-lab's models output about 100 billion words per day, which is 1.2M words per second.
Let's say a typical response is on average 500 words. That's 2,400 requests a second. You're looking at about 10 bit flips a second, assuming one output run is one "inference," by Meta's numbers. Does seem a little high, based on that guesswork, tho that's the number quoted.
It's frustrating that Meta has issued a bunch of research on SDCs but presumably the lawyers and PR prevent the release of hard numbers on the number of bit flips its datacenters experience per unit of time. Other than the negative press it might generate ('Facebook gets it wrong X times a minute!!') it also partially reveals how many servers or how much compute resource Meta has, and hyperscalers hate revealing that info.
C.
Hey, we said "some black magic". Key word "some" – and black magic isn't pejorative. That's a nice way of saying it involves some internal bits of the OS that people generally don't go near.
If you know how that all works, great. But just as you're allowed an opinion about sudo, so are we.
C.
Ah, come on, give us some credit!
The words source code are in scare quotes ('source code') because that's how the leaker described it. In the article we call it internal data and assets. When you see 'source code', that's the claim: the article refers to what's actually been allegedly leaked.
C.
It's more nuanced than that as I understand it.
If you whitelist specific Azure services to allow them to access to your servers as you instruct, anyone using those Azure services also gets access.
Tenable argues Azure should prevent tenants from sending requests to other tenants via these services; Microsoft says you should put in your own levels of authentication and filtering to prevent cross-tenant access.
I believe in having competent IT staff who are aware of this, sure. I also personally believe in not handing footguns to IT staff.
C.
TOPS is short for a trillion operations per second. In the context of AI, it'll be INT8 or signed byte precision. So 40 TOPS is 40 trillion operations a second using signed byte values. It's a measure of AI performance by systems.
You'll typically see manufacturers say their hardware can hit X TOPS of AI performance, and software and LLM makers say you'll need hardware capable of Y TOPS for this AI application to be useful. You as a user will want X to be greater than Y.
Eg, Microsoft claims a PC needs a minimum of 40 TOPS to run its Copilot+ suite.
I'll define a page for it and link to it.
C.