jakobson14

apple: "Out latest laptop has <magic maguffin>!"

consumer, at the store: "why doesn't this thing have <magic maguffin>!? bring me something that has <magic maguffin>!!!"

Intel has spent the last 25 years training consumers to look for the "intel inside" sticker. It's the only thing that's kept intel afloat while they butcher their engineering tallent.

People shouldn't care, but they do.

Guy is an employee of raspi, so of course he's going to write code that ships chips.

No there isn't. Let's take pytorch as an example.

CUDA is an API. OpenCL is an API. Pytorch is not.

Pytorch is very firmly GLUED to CUDA. It will probably NEVER support anything else beyond token inference on mobile devices. The only reason Pytorch supports AMD at all is because of AMD's "I can't beleive it's not CUDA" HIP translation layer.

OpenCL is a real cross-platform API and with 3.0 it's finally "good" and coincidentally, intel is....half-heartedly interested in it, except they're shooting themselves in the foot by trying to also cover useless CPUs for inference/training and spreading themselves too thin (OneAPI). Because all intel can think about are CPUs. Everything must drive sales of CPUs.

At this rate just about the only think that might save us from CUDA is rusticl. If a real, full-fat, high-quality openCL 3.0 driver sudently popped into existence on every GPU platform under the sun, maybe pytorch et al could finally be convinced to give a shit about an API other than CUDA.

They are only moving because AMD is finally providing that competition, but it's not fair to say that "if only AMD had gotten their act together sooner, intel would be a better comapny."

Ultimately, intel chose to let intel rot while AMD was out of the picture. (ignoring that intel's backroom deals with Dell and co. are a big part of what pushed AMD out)

ultimately it was a focus on marketing that led them to kill engineering

(if itanium and the many failures and lucky breaks that led up to it hadn't already convinced you that intel might never have had good engineering)

Honestly, they'd probably be better off if they ditched all the sed/awk/macro BS and just went back to bash scripts (or perl/TCL, if you don't like weird syntax issues) that spat out C code. Rust saw the writing on the wall and implemented proc macros.

Stop using tiny hammers and get a big one.

Oh, side note: Safe interps now have available time limits and a few other measures, though it seems memory allocation limits are a harder nut to crack due to all the places a memory allocation can occur in the TCL interpreter.

https://www.tcl-lang.org/man/tcl8.5/TclCmd/interp.htm#M46

Still blows my mind that TCL includes a firewall and virtual filesystem layer.

https://wiki.tcl-lang.org/page/firewall

https://wiki.tcl-lang.org/page/VFS

https://wiki.tcl-lang.org/page/tclvfs

You can even build a FUSE filesystem in TCL!

Continue forever? no. Be performed once at no cost? maybe.

I'm kinda sceptical that a computation to which the 2nd law is indifferent would occur spontaneously without immediately reversing. The 2nd law is what determines the direction that things typically progress.

Congratulations on supplementing one impossible thing we can't do yet with three more.

There are very good reasons these artificial hearts are strictly temporary while waiting for a donor.

The heart is massively more efficient. Just about everything in all of biology is massively more efficient.

Take a moment to consider that your entire body runs on an energy budget of only around 100 watts, of which your brain only uses a measly 20 watts.

Sounds a lot like TCL's Safe interps?

The verifier is mostly targeted at not letting a program run forever.

inside the TB link there's no such thing as "3.0" or "x4"

hector goes over this as well, but basically you've replaced all the lower (physical) layers of PCIe's networking stack. intel like to pretend that thunderbolt and USB4 just tunnel actual PCIe links unchanged (as if they're emulating everything down the PCIe lanes and PHYs), but they don't. PCIe over thunderbolt can run at whatever speed you please, limited only by the throughput of TB and the overhead of the remaining layers of the PCIe stack.

bots pay for blue checkmarks, and they'll pay to tweet too. twitter needs the money to stay afloat, and elon doesn't really have a reason to say no

his initial concern over bots was twofold: 1) he desperately needed a way to get out of a deal he made on ~vibes~ instead of due-diligence 2) elon was a mega-famous account. The #1 thing that impacts QoL for mega-famous accounts is a 24/7 firehose of bot mentions and replies

Ford knows how to build cars after building them for 100+ years. There's nobody who has ever tried to fly a helicopter on mars before.

NASA built in lots of margin to help ensure they met their objective, especially considering that it's really fucking hard to get a new helicopter to mars if it breaks.

Drag isn't relevant for a helicopter. Thin air means you have a very hard time just trying to hover in place without moving.

There's an odd dig at the FSF in there that I wish had more context. What must have it been like to watch GNU clone unix, having worked on it?

Read the article again, or better yet do a ctrl-f for the word "cyber"

https://www.csoonline.com/article/656427/over-40000-cisco-de...

With how shit cisco's security is and how badly they're having their ass handed to them, you'd think they could roll out any cybersecurity "hardening" in their special ukraniuan firmware to other models.

Read the article again, or better yet do a ctrl-f for the word "cyber"

https://www.csoonline.com/article/656427/over-40000-cisco-de...

With how shit cisco's security is and how badly they're having their ass handed to them, you'd think they could roll out any cybersecurity "hardening" in their special ukraniuan firmware to other models.

Nice dodge, but I'm referring to the dumb data-structures mimicking sentient human words.

Looks like you've fallen right into the corollary.

It's absurd to expect self-awareness in some system just because its parts are similar to the parts of other self-aware systems you're familiar with.

You are vastly over-estimating the competence of both cisco and the russian hackers

https://www.csoonline.com/article/656427/over-40000-cisco-de...

Read the article in full.

Timekeeping upgrades and cold-weather functionality are the flashy headlines that are easy to explain to the masses, but there's plenty in there about russian cyberattacks too. Russia has been attacking ukrane's infra over the internet for over a decade now.

Apollo was a scramble to get the absolute minimum tin can to the moon as fast as possible which wouldn't kill the 2-3 men inside it. Building anything secure is a detail-oriented endeavour. It is not solvable with a moonshot.

The question I have for cisco is: why are these special features for resisting cyberattack not standard on EVERY router?

Pure mythology.

Have you not heard of 500 errors? How about 400 for "bad request"?

Seriously, stop reinventing what is already an established standard. There is nothing more annoying than a server that returns a 200-class status when the thing I requested is not in the response.

Using gcode on a lathe or other subtractive machine is ORDERS of magnitude more productive because orders of magnitude less code is required because orders of magnitude fewer operations are required.

With a subtractive process you typically only need to put a few holes into specific locations using a cutting tool roughly the same size as the hole. Jog over, plunge, jog over to the next site, plunge. Maybe you need to mill out a few pockets or an exterior shape, in which case you switch to an endmill that's good for mass material removal and perform 2-3 passes around the outline's polygon.

And then there are additive processes that 3D printers use. Try to back-of-the-envelope how many passes of the tool head are required to build a 5x5x5cm cube using a hotend that has an orifice of 0.0004 cm and a layer thickness of 0.0002 cm. It's a lot of passes, and because gcode was designed for the subtractive processes where only a few passes are typically required, it contains zero macro commands for doing it in fewer lines of code. If you want to infill with a fancy material-saving pattern, the complexity of the movements required goes exponential and so does the LoC.

China is already doing it. Replaceable batteries are seen as the way to go over there and multiple companies are already offering it. Drive your car in and a robot will drop the old battery out and lift a new one in. No more hassle than a carwash.