Kliment

Sorry about the orange tint on the slides, I forgot to turn off the redshift function in KDE and the talk was late at night.

You might like https://en.wikipedia.org/wiki/Robot_%26_Frank

The only fruit that belongs on pizza is durian.

Nuclear plants need to be switched off for weeks/months at a time for the following reasons:

- refueling (regularly done, highly plannable)

- insufficient cooling water/ too warm cooling water (regular, seasonal, not very predictable short-term)

- planned maintenance (often combined with refueling)

- unplanned maintenance (accidents)

Whenever these things happen, the entire reactor is producing no power. This is why most nuclear facilities have multiple reactors, so you can rotate these tasks among them and still have some power. Unfortunately, the cooling issue is becoming more and more of a problem, and many sites that used to operate year-round now have to scale down significantly in mid-summer due to lack of cooling water.

There are many ways to improve the speed and accuracy, but it's really important in settings like these to not make changes until the input method is fully reliable, because there's the risk of complete loss of communication.

The full event streams have been published this time around, including the failed sessions. I recommend reading the article and watching the videos of the system operating - they have a video of the input calibration/training, and another video of the longest session. The sessions are few and many days apart, and not all of them demonstrate successful control, but on the days where the patient can successfully drive the system there is no doubt of the communication method's authenticity, and the researchers have no input or interpretation capability.

The previous controversy around the researchers is not about them scamming patients, it's about them publishing only the positive results of their work and not disclosing negative results, making the average success rate appear better. This is one reason they publish full event streams and list days with no attempts and failed attempts extensively in this paper. The code has also been published, and the raw potential recordings are available on request. I had a careful read and I can't find anything methodologically problematic in this publication. When you watch the video, be aware this is the longest and most successful session, and the researchers made this very clear.

Wordle, as we loved it, tracker-free and entirely non-evil, is still available to this day at https://web.archive.org/web/20220209234643id_/https://www.po...

I cannot imagine cooking with fewer than 3 rings.

You're allowed to own more than one standalone induction plate. And you're also allowed to position them freely over your work area :)

Get a standalone induction plate. They're cheap and portable and they let you try it out without a huge expense. I do 90% of all my cooking on a single induction plate I paid ~$60 for. Non-induction electric hobs are trash, and induction is much closer to gas, but more controllable, less messy, and less dangerous. That said, I would strongly recommend against a builtin induction hob. They're expensive as fuck, and eat lots of counter space. The standalone induction plates are amazing, much nicer to work with, and cheaper, and you can put them away when you need the extra space. Restaurant supply shops have standalone induction plates for ~3x the cost of the consumer ones which are phenomenally good, that's what the higher end new kitchens are going for.

I understand Russian. Looking at ria.ru over the past few days there is barely any mention of the war, certainly no details, there's much more mention of sanctions but only some weird stories about their origin and motivation.

The easiest way is to not use google fonts.

That's actually shown in the article as a mitigation strategy

The modern version of this is http://www.koruza.net/features/ which piggybacks on fiber transceivers.

"The Ugly American" has been claimed to have been an enormous wake up call to how the US does foreign policy. It allegedly motivated much of Kennedy's approach and was responsible for the formation of the Peace Corps. Before he was elected, he bought a copy for every sitting senator.

It means if you move to a different country you can probably keep your email, but you will probably lose your phone number.

implicitcad uses SDFs for modeling. It's weird but it works.

ARM has already succeeded. There is no "will" here. Most microcontrollers in the world are ARM. Most application processors in the world are ARM. It spans from the cheapest to the highest-performing parts. Nearly every smartphone in the world is ARM.

If you need help with redesigning your devices to be manufacturable with what's available now, poke me, maybe I can help. I've been doing this sort of shit for over a year now. It's not fun, but it's saved a number of products.

Component production cannot profitably increase because there is no increased total demand. Everyone is buying everything they can just to make sure they can produce, but they're not actually producing more. Component producers know this and that at some point this artificial demand due to stockpiling will fall off a cliff, and they'll be left sitting on underutilized massively expensive production equipment if they increase volumes. Besides, with the current stockpile-driven demand they can ask higher prices. This is why component production is not, and will not, rise.

GPU shortages are driven by a number of factors, and almost none of them are to do with component production (or wafers). There's a big scalper problem for GPUs, where some assholes will buy every GPU they can find and then resell them at a huge markup to either (increasingly desperate) end users and system integrators, or, increasingly, to cryptofuckers who will pay much more than legitimate users to ride their bubble. This means that even if there's plenty of production, unless the production can outspend the scalpers' total capital, it's all going to be swallowed up.

There's no general wafer shortage, but the highest grade wafers that get used in the highest end production have long lead times (because the fabs that use them reserve them way ahead of time). Wafer availability is not a bottleneck since the smallest nodes have such low capacity to start with. What has happened is that wafers have gone up in price in response to the general stockpiling behavior, but the high end stuff (including high-end GPUs) don't care as the wafer cost is basically negligible to them compared to all the process costs.

I work in this industry. Nobody is producing significantly more as a result of covid or anything else. Aggregate demand for electronics products has increased a bit, but it's not a significant change. The "skyrocketing demand" mentioned in this paper is not for electronics products, but for their components, and it's happening entirely due to a lack of visibility into future availability and demand, or, to put it more simply, a lack of confidence in the market. Device manufacturers are buying as much as they can because they don't know when they'll be able to buy again. Component manufacturers are not expanding production because they don't know when the rush buying will end and everyone will produce from stockpiled material. This is primarily a confidence problem (caused primarily by automotive manufacturers being absolute shits to their suppliers). The total production volumes of electronic devices have not increased by any significant amount. Many manufacturers are actually producing less now than pre-pandemic, simply because they can't produce more due to shortages.

That's exactly what made Krita so good. Every once in a while they'd shut a bunch of the developers in a room with a bunch of artists/users, have the users complain about everything that pisses them off, and get all those things fixed. The resulting application is obviously driven by what artists enjoy, and it feels that way. Strongly recommended.

Yeah, whenever I end up on google for whatever reason I get a nag screen to try to trick me into agreeing to their tracking crap (with no easy way to say "fuck off, no") and it seems to come back very frequently. I use ddg pretty much exclusively now.

Sadly they don't care. There's countless reports where the the reporting user gets pointed to answers.microsoft.com, and some asshead employee replies telling the user their issue is invalid and they should post on uservoice instead. For example, Onenote class notebooks can still not delete group folders, ever. A bunch of them get added by default, and can never be deleted. Adding one takes a few clicks and it can never be deleted. This was first reported in 2016 and hundreds of frustrated teachers clicked on "I have this problem too". Some microsoft employee replied telling people that that's not going to happen, and they should post on uservoice. As of this year, the entire uservoice instance with thousands of issue reports and hundreds of comments on each was deleted. Microsoft has used every tool available to tell its users their feedback is not wanted and they should fuck off.

You're missing that large feature sizes are often used because they need to handle particular voltages and temp ranges. If you change the material properties those characteristics are no longer viable with the original spacings and dimensions, so you get to relayout anyway.

The GPU market is ruined by fuckers who are buying up all inventory and reselling it to either desperate system integrators and users, or to crypto fuckers at an enormous markup. This is an entirely different problem compared to what we're talking about here.

No, the errors are inside the feature itself, since it doesn't fit on a single mask. You can't reuse design files on a new process. Everything will be wrong. You have different material properties, different gate capacitances, different parasitics. No matter what you do, you have to redesign to move to a new node. The problem with large features is, even if you do all that work, you still can't reliably make them with the smaller process, and all the benefits of the smaller process (smaller die size, faster switching, higher speeds, shorter interconnects) go away if you use large features.

Unless you're a giant customer with something big to beat up the vendors with, like high-level political connections, you're going to be paying up-front for delivery sometime in the distant future if you order anything today. Many vendors won't take any orders without payment up-front.

Because (very very roughly) the way you make small features is by scaling down your projection area to a tiny size and then stepping it across the wafer. Your projection area boundaries are where errors are most likely, and the bigger your features the more such boundaries they cross. The details are quite involved, but this is the general problem.

New microcontroller designs that would meet the requirements for safety certification in those conditions are also manufactured on larger process nodes.

It's almost never possible to do a function-compatible die shrink for anything that needs to work over a large voltage and/or temperature range. Many automotive chips operate at 80V or more, 5V and 12V parts are ubiquitous, and indeed lots need bulk area to be able to switch large currents. Making them on small processes is not only an enormous design effort where it's possible at all, it also doesn't make sense because there's still way more capacity on larger processes than on smaller ones, and the only way small processes win is by packing so many dies on a wafer that they can get away with their terrible per-wafer cost and awfully bad defect rate.