ckastner

Landing can be: https://en.m.wikipedia.org/wiki/Autoland

In fact, it's pretty routine. Don't have the source at hand, but somewhere around 1% of all landings (at airports with ILS) are autolands.

I think it was Boeing that even requires at least 1 autoland per plane every 30 days or so.

You can find videos of this on YouTube. Completely hands-off.

> PayPal never did seek a banking license, arguing that it wasn’t actually accepting any deposits [...]

That worked in the U.S.

European banking regulators saw through this ploy and forced them to seek a banking license.

This is a bit misleading as the decline reported is year-on-year and of course sales haven't fully recovered yet.

Quarter-on-quarter or month-by-month would have been much more interesting, could have shown a change in trend. Especially after Musk's departure from the administration.

Oh this is neat. Took me a bit.

The shell treats the first line as a comment. It executes the second line, which eventually exec's the binary so the rest of the file do not matter to the shell.

And the compiler treats the first line as a preprocessor directive, so it ignores the second line.

I initially misread/mistook the first line for a shebang.

There was another 1981 paper that went into this by Boehm "Software Engineering Economics", but I can't find the details right now.

This NASA publication [1] cites a number of studies and the cost increase factors they estimate for various stages of development, including the Boehm study.

[1]: https://ntrs.nasa.gov/api/citations/20100036670/downloads/20...

This.

And often it's not an unhelpful upstream, just an upstream that sees little use for man pages in their releases, and doesn't want to spend time maintaining documentation in parallel to what their README.md or --help provides (with which the man page must be kept in sync).

Dimensions: 154,7 mm (height) × 73,3 mm (width) × 8,9 mm (depth) or: 6.1 in (height) x 2.9 in (width) x 0.4 in (depth)

Weight: 185,9 g or: 6.6 oz

A bit lighter than the previous versions, though I do miss the days where tha 4a was just around 145g or so.

In a similar vein: https://www.spyder-ide.org/ (MIT-licensed)

Yes, this. For example, `Ctrl+R makedo` would go back in history to the first command that matches this, to use the ` ./scripts/makedocs.sh` example from the article.

You keep pointing out European involvement as if that somehow displaced American involvement (and thus continued U.S. control, the subject of this thread).

Again, your first quote explicitly states that EUV-LLC was American. The second quote refers to the "Labs", which in this case were the Lawrence Livermore, Sandia, and Berkeley National Laboratories.

Apple may have developed the M4 from the ground up but they still licensed the ISA from ARM.

> It seems you are confusing the details and conditions of a contract never disclosed publicly

The details have never been disclosed but it is well known that this agreement fell within the domain of national security and export controls.

Here's a press release [1] directly from ASML that references these export controls. Even though this PR actually relaxing DUV controls with respect to the U.S., it reaffirms that "EUV systems are also subject to license requirements."

[1]: https://www.asml.com/en/news/press-releases/2024/dutch-gover...

> By 2018, ASML succeeded in deploying the intellectual property from the EUV-LLC

That explicitly says that the IP is with EUV-LLC (edit: which is of U.S. origin).

It seems that you are confusing ownership of the company with licensing of the IP it uses.

AMD produces their own x86 under IP licensed by Intel (and vice versa).

That still looks like something that a novice might trip over when reading this. Just describing what it does is already a bit convoluted: it tests whether any value of a list of booleans, produced by a generator expression by testing membership of a value in some other list, is True.

I would have gone with the following, as it requires less mental unpacking:

  if set(args) & set(["-h", "--help"]):
      display_help()

Also, note that any() works with any iterable, the outer brackets are not needed given the list expression.

  if any(a in ["-h", "--help"] for a in args):
      display_help()

> flutterfoundation.dev

It's not Flutter by own admission, it's very likely not backed by a foundation, and I highly doubt that Google consented to its "Flutter" trademark used for such a foundation.

Fair point.

> So if an automated decision happens, and the reviewer looks for a second at it, and says, good enough, that will be OK according to GDPR. Don't see what GDPR solves here.

The assumption is that a human review the conditions that led the automated system to make that decision.

I think it would be trivial to argue in court that rubberstamping some scalar value that a deep neural net or whatever spit out does not pass that bar. It's still the automated system's decision, the human is just parroting it.

Note that it's easier for the FAANGs to argue such a review has happened because they have massive amounts of heterogenous data where there's bound to be something that would be sufficient to argue with (like having posted something that offended someone).

But a single score? I'd say almost impossible to argue. One would have to demonstrate that the system is near-perfect, and virtually never makes mistakes.

Indeed. Quoting article 22 [1]:

> The data subject shall have the right not to be subject to a decision based solely on automated processing [...]

[1]: https://gdpr.eu/article-22-automated-individual-decision-mak...

> You might as well recoup as much money as you can on it.

Depending on how fast their value depreciates, selling them might recoup more money then renting them away. And being exposed to 3y of various risks.

Selling now at a 40% loss gets you back the equivalent of 60c/h over three years, and without having other costs (DC, power, network, security) and risks.

That's just the power. If one expects a H100 to run for three years at full load, 24 x 365 x 3 = 26280. Assuming a price of $25K per H100, that means about $1/h to amortize costs. Hence the unless they stole them, I guess.

Factor in space, networking, cooling, security, etc., and $2 really do seem undoable.

It would probably not stop all theoretically possible attacks, but it would stop many of them.

Say you find a bug in the GPU driver that let's you execute arbitrary code as root. That still all happens within the VM. To attack the host, you'd still need to break out of the VM, and if the VM is unprivileged (which I assume it is), you'd next need gain privileges on the host.

There are other channels -- perhaps you can get the GPU to do something funky on PCI level, perhaps you can get the GPU to crash the host -- but VM isolation does add a solid layer of protection.

I don't think I fully understand your question. If, with turnkey operation you mean virtualization, enterprise GPUs already officially support it now, and it already works with consumer GPUs, at least the discrete ones.

Amazon/AWS has no use case for VFIO in Firecracker. They're open to the community adding support and have a community meeting soon, but I wouldn't get my hopes up.

QEMU can work -- I say can, because it doesn't work with all GPUs. And with consumer GPUs, VFIO is generally not an officially supported use case. We got it working, but with lots of trial and error, and there are still some problematic corner cases.

> If the calls first pass through a memory safe language as what gvisor does, isn’t the attack surface greatly reduced?

The runtime may be memory safe, but I'm thinking of the GPU workloads which nvproxy seems to pass on to the device via the host's kernel. Say I find a security issue in the GPU's driver, and manage to exploit it with some malicious CUDA workload.

> To add GPU support, the Google team introduced nvproxy which works using the same principles as described above for syscalls: it intercepts ioctls destined to the GPU and proxies a subset to the GPU kernel module.

This does still expose the host's kernel to a potentially malicious workload, right?

If so, could this be mitigated by (continuously) running a QEMU VM with GPUs passed through via VFIO, and running whatever Workers need within that VM?

The Debian ROCm Team faces similar challenge, we want to do CI [1] for our stack and all our dependent packages, but cannot rule out potentially hostile workloads. We spawn QEMU VMs per test (instead of the model described above) but that's because our tests must also be run against the relevant distribution's kernel and firmwares.

Incidentally, I've been monitoring the Firecracker VFIO GitHub issue linked in the article. Upstream does not have a use case for and thus no resources dedicated to implement this, but there's a community meeting [2] coming up in October to discuss the future of this feature request.

[1]: https://ci.rocm.debian.net

[2]: https://github.com/firecracker-microvm/firecracker/issues/11...

Fair point.

I wonder whether there are any "outs". DE and PL subsidized because they want fabs ASAP; if Intel can't hold up their end, why not cancel the deal and go with TSMC or Samsung instead.

> It also said it would pause construction at its chip factory project in Germany for two years, a move Reuters had previously reported. The company plans to pause its project in Poland as well.

Just three days ago: Poland's Intel plant gets EU green light for $1.9B in state support [1].

Germany is subsidizing Intel with $11bn.

I'd be fuming.

[1]: https://news.ycombinator.com/item?id=41534763

No, that only thing clear is that the market expected better results.

Not achieving those results does not imply that results are going downhill from now, as you suggest.

Slowing growth and cooling margins are reasonable explanations for the stock being down even in light of continued growth.

If the market believed that demand had indeed peaked this quarter as you suggest, the stock would absolutely crater.

But nobody reasonable believes that. All the data from outside Nvidia (eg: Capex from hyperscalers) also points to continued growth.

> put it simply, this earnings release is as good as it will get for a while

They literally guided more growth for the next quarter and Nvidia is notorious for sandbagging this number.

Yeah, this is one of the more common failure modes. One meeting wasted, and possibly lots of time lost when the next meeting can only be scheduled much later.

All easily preventable if the person organizing the meeting prepares for it, and shares a brief agenda so that other participants can prepare to, or find replacements if they themselves are the wrong person for the meeting.