makomk

In addition to the laws of physics and the engineering required to build TVs working against this, the EU deliberately changed the energy efficiency scale for things like TVs a few years back to specifically make an A hard to get and something that wouldn't be achieved by products currently on the market. They were probably too optimistic about future improvements too - a lot of TVs had to add special eco modes that aren't really designed to be used to meet the minimum efficiency now required by EU rules.

It's certainly not the only evidence of problems at NPR. For example, they managed to basically accuse Trump Jr of lying to Congress in a story that should not have survived basic fact checking: https://www.foxnews.com/politics/npr-issues-correction-after...

That Fox News piece actually understates how big of a screw-up this was. The key quote that supposedly showed Trump Jr claiming his dad's possible real estate deal in Russia had faded away by 2016, the one that was supposedly contradicted by Cohen's court testimony about ongoing negotiations, was in response to questioning about any possible deals other than the one Cohen was involved in - and in particular one specific potential deal with a different group of people. It's not just that it was brought up elsewhere in other answers that NPR missed. Merely looking at the immediate context of that key quote, the most basic thing we should expect of old-fashioned fact checking, should've been enough to flag the problem. The fact those other negotiations had in fact been brought up was literally the whole basis for that line of questioning.

The Chilterns, I think, and burying the line in tunnels was more or less necessary anyway. Some of the tunnels could technically probably have been replaced with cuttings, but apparently it would've cost more. The existing line used them mostly because tunneling was a lot harder back then. Mostly, the NIMBYs seem to have forced the existing tunnels to be longer than they need to be, which is expensive but not making building infrastructure in the country entirely non-viable levels of expensive.

Homebrew 6DoF tracking is definitely possible. I've had a janky and undocumented setup for a while that uses a standard smartphone as a display, paper and cardboard AprilTag markers with a computer and webcam for outside-in tracking, and homebuilt controllers. It requires a lot of improvement and is very sensitive to lighting conditions though.

I'm not sure it is. The EU has increasingly strict energy efficiency requirements for washing machines that are frankly stupid - they're pretty efficient already, so the way that manufacturers have improved efficiency through ludicrously long multi-hour wash cycles that keep on getting longer (there's apparently a direct relationship between the length of the wash cycle and how little energy can be used to clean clothes). The efficiency gains make washing less useful and consume more of people's limited time to the point that the cycles those numbers are based on don't really seem to be intended to be used.

That money has already gone out to people though. The reason that people's pay has bought less than before is that a large chunk of the economy was shut down and those goods and services were not produced. This means that there is simply no way for people to be able to purchase as much as they did before because it's just not there to buy in the first place. Shutting down the ports makes this worse by preventing US-based factories from getting the inputs to operate and non-US-based factories shipping their products to Americans. This is effectively making the whole country poorer in real terms as leverage to try and get a bigger share of what's left.

The basic reason this happens is that people's wages are only worth what they can buy, and in order for them to be able to buy those things they need to be produced and shipped. Sure, you could argue that this doesn't actually matter in practice because the corporations will always have enough excess profit to just pay people higher wages and employ more people to do less when the unions demand it, and that they'll have to do so rather than increasing prices, and that these cuts in profits will somehow always come out of the pockets of unsympathetic billionaires rather than ordinary workers' pension funds. That requires a lot more about the world to work out just right compared to the economics 101 version though.

As far as I can tell, this isn't actually true: "the DNA evidence that was not destroyed by the state is exculpatory". The Innocence Project are being very careful with their wording here. Initially, they relied on trace DNA on the knife that didn't match the accused murderer, but that ended up being from someone in the prosecutor's office handling it after it had been processed for forensic evidence. Then they tried to argue that this showed the state had destroyed evidence which would've proved his innocence, but the courts didn't buy it because all available evidence suggests the killer's DNA was simply never on the knife. (Which isn't that surprising - DNA evidence isn't perfect and gloves exist.) The other "forensic crime scene evidence" seems to be hothingburgers like a few non-matching hairs in a house that'd had a large number of people going in and out in the recent past.

All available evidence suggests there's nothing "classified" or "wartime ready" about these - they were your basic, cheap, totally unencrypted POCSAG/Flex pager. The same as any other pager carried by doctors and all the other people who use them - aside from the hidden explosives, of course.

In theory you should just be able to set -march to the lowest common denominator kind of CPU you expect your code to run on and it'll avoid relying on SSE if appropriate.

Yeah, that is a worry: maybe OpenAI's business model and valuation rest on reasoning abilities becoming outdated and atrophying outside of their algorithmic black box, a trade secret we don't have access too. It struck me as an obvious possible concern when the o1 announcement released, but too speculative and conspiratorial to point out - but how hard they're apparently trying to stop it from explaining its reasoning in ways that humans can understand is alarming.

It seems like the Guardian is spreading misinformation for election reasons anyway, judging from the comments - the increase in infant mortality it claims is well within the range of normal year-to-year variation, which is presumably also why an article that claims it's the "legacy of 14 tears of Tory cruelty" uses a claimed increase over just one year.

All of the stuff that they're saying in that response about how they won't abuse it is stuff that's already in the TOS. If you don't trust them despite that, then you should not be using their products regardless of what the TOS says.

The original 20-year-old version of golden rice was unfortunately basically unusable (not enough bioavailable vitamin A even when used as part of a much more balanced diet that was more favorable to vitamin A absorption than that of the people it's actually aimed at and substantial yield reductions). Fixing the yield reductions at least required more modern GM techniques that are likely going to be under patent for a good while.

They're not just philosophically correct but probably correct in practice as well. Golden rice - especially the original version of it - is not just an easy drop-in solution. It requires the development of specific variants suited to the climate where it's grown, the original version didn't supply enough vitamin A even in an ideal scenario and had pretty major yield reductions which made the rice more expensive (which is a huge problem when poverty is one of the major reasons people are so dependentt on rice in the first place), and this was compounded by licensing restrictions which blocked both cross-border sales and most growing in countries which were self sufficient for food production.

Those licenses made it effectively unavailable both to most countries which imported their rice and most countries that were self-sufficient. I think the two countries that had early trials may well have been the only two that were both eligible to make use of it and able to do so, and in at least one case that was a result of an error which resulted in them being counted as eligible when they weren't. They mostly seem to have been a PR stunt, something big biotech could point to and claim that they'd given the world a free solution to vitamin A deficiency that was being blocked by evil anti-GMO campaigners that wanted kids to go blind.

Oh huh, a Guardian article about a protest about unsustainable tourism in Tenerife and the Canary islands and the environmental damage it's creating. Wonder if they're going to mention that an awful lof of those developments on Tenerife are illegally dumping all their sewage in the sea rather than building sewage treatment... nope.

For context, there's been a huge amount of anger in the UK that was basically kicked off by the Guardian about how the UK's seas and rivers are catastrophically polluted with sewage by our water companies, we're supposedly the dirty man of Europe now, privatisation and Brexit has ruined everything and it's all getting worse, to the point it's likely to influence the next election. One of the big headline stories involved a British windsurfing champion leaving because it's like "surfing in a sewer"... and moving to Tenerife, where sewage treatment is objectively far worse and as mentioned there are sewage systems that just dump all their sewage out into the sea basically everywhere. (This is something that was fixed in the mainland UK decades ago post-privatisation, and in fact seems to be one of the big infrastructure improvements that privatisation was done to get funding for.) The rest of the coverage has been about as founded in reality as that.

When looking for information on this, I even ran across people that were presumably from there complaining on a forum about this and how it'd probably only get fixed if British tourists found out about it and started objecting. Guess there's little chance of that.

Yeah, the key reason this enables so much faster time to first fix is that the precise ephemeris parameters are transmitted much more often than the full almanac, but only from the satellite they apply to whereas each satellite broadcasts the entire almanac covering the whole constellation. If I'm understanding the info out there correctly, every transmission of the ephemeris data comes with only 1/25th of the almanac.

Most decent modern-ish receivers tend to have pretty speedy aquisition time without any assistance data. For example, the reasonably ancient GPS running watch I use can usually get a GPS lock in a couple of minutes from cold with no internet access (in a wrist-sized device running on battery!), and even the two decade old SiRFstarIII chipset is specced to have a sub-minute cold start time without assistance and much shorter with - though I think that chipset was pretty advanced for the time.

I think even these days, a lot of what the author is doing here is still generally done in hardware for power efficiency reasons - it's just that nowadays, the hardware looks a lot like the software-defined architecture used here with a conversion to digital at the frontend followed by hardware accelerators for doing cross-correlation in the fourier domain. Actually, this kind of approach of implementing the architecture an SDR-based receiver would use as specialised hardware seems to be pretty common in general nowadays.

Reduces the attack area through which this could be found, I expect. Without all the obfuscation someone might spot suspicious data in the test data or at some other stage, but this basically forces them to find the single line of suspicious shell script and follow the trail to find the rest of the stuff added to the build process.

If I remember rightly, Starbucks' profits two years ago were down fairly substantially compared to their pre-pandemic levels and their long term goal seems to be to try and return the business to that same stable level of profit.

Yikes. For context, the 7900XTX and 7900XT are the only consumer GPUs that AMD officially support running AI workloads on, and the XTX is the one they recommend out of the two. Other cards in the same generation and any from previous generations are officially not supported or tested and may not work. So it's not like they have a particularly large amount of hardware that needs to be included in their testing environment.

The actual compiler output from the autoconf feature test is one of the things I'd probably look at fairly early on if some feature is disabled when it shouldn't be, but I maybe have a bit more experience running into problems with this than younger folks.

AMD would explain that ROCM is an abbreviation of the ROCm Software platform or something unspeakably stupid. And that your graphics card wasn't supported.

If even that. A few years ago they managed to break basic machine learning code on the few commonly-used consumer GPUs that were officially supported at the time, and it was only after several months of more or less radio silence on the bug report and several releases that they declared those GPUs were no longer officially supported and they'd be closing the bug report: https://github.com/ROCm/ROCm/issues/1265

At least here in the UK (which has the second most offshore wind generation in the world behind China) it's both - onshore wind got stuck in NIMBY planning hell and there's not really enough land in the right places to make it an easy option. Also, it looked for a while like with sufficient investment offshore wind could actually wind up being cheaper due to higher capacity factor, larger turbines, etc. Frustratingly, a lot of the British press has been pretending that the NIMBY issues didn't exist and the current government just made them up for ideological reasons or because their super-rich donors didn't like wind turbines, along with presenting onshore wind as the solution to all our problems and essentially ignoring the offshore wind and pretending the government has done nothing on renewables. (This seems to happen with literally everything - whatever the current governemnt didn't do or cancelled is spun as a magic solution and any problems it was having before are written out of existence.)

As far as I can tell the HDC-402 wasn't even the Viking computer that got reused on Voyager anyway - it was something NASA developed in-house from the orbiter side of the Viking mission and that presumably has even less documentation available.

It's low effort, talking point nonsense. The whole reason this is interesting is that it allows the recovery of non-public information that was never released at all - about the closest analogy I can come up with is if it was somehow possible to reverse engineer part of an author's notes from their published work, though that's a very imperfect analogy. In order to tie this in to a pre-existing talking point about LLMs being bad, the current top comment has to basically completely ignore all of the reasons why this is interesting and surprising and considered an attack in infosec terms - indeed, not really engage with what it's doing at all.

It's not just a question of shareholder value and profits always needing to go up - there are real society-wide opportunity costs to companies like Sony continuing as before. All of the workers they employ and the resources they use are workers and resources that cannot instead be producing other things, and the increase in interest rates that's lead to all these layoff is essentially a signal that those opportunity costs are higher and the bar for economic activity to be viable is higher than before.

Part of the problem, of course, is that the pandemic and the economic shutdowns hugely impacted economic output in a way that couldn't be magicked away by printing money and handing out furlough payments - all the furloughed and unemployed workers and shuttered still produced nothing during that time, that unusued capacity could not be saved up for later, and so it wasn't possible for everyone to continue buying the same amount of stuff as before even though goverments in the developed world tried to create the illusion they could. This also created a temporary increase in demand in certain sectors, like video games, which has since vanished.

Pretty much every article lamenting the seeming inevitability of capitalism and the lack of coherent alternatives is going to misdiagnose the problem and have no concrete solutions, because the fundamental problem is that the problems people care about are not simply consequences of capitalism and cannot be eliminated just by getting rid of it.

For example, no amount of monkeying with the economic system will make it possible to fight global warming simply by cutting off fossil fuel supplies and destroying fossil fuel company profits without huge impacts on the demand side that affect ordinary people - that'd require consuming fuel that was never produced. This effect on ordinary voters would make it horrendously unattractive to politicans even if they didn't have to care what corporations thought at all. Similarly, here in the UK there's huge public anger over sewage in our rivers and seas but no amount of nationalisation can change the fact that rebuilding the sewage system not to do that would take huge amounts of resources from other things that people care about and cause lots of unpopular disruption from the work required.

Because the things that people care about and want fixed aren't actually caused by capitalism, the most politically effective anti-capitalist ideas are populist lies that promise false solutions to those problems which cannot actually work. This also means that any leader or would-be leader with any sense isn't willing to go along with those ideas because they're creating a whip for their own back when they fail to live up to that. Any concrete, achievable anti-capitalism also has to fall so short of the populist version to be entirely unattractive.

Apparently most stuff that relies on the weights being resonably similar to SDXL don't work - control nets, LORAs, commonly-used inpainting patches, the lot. It seems to go well beyond fine tuning and be substantially retrained to the point it's a good chunk of the way to being a different model entirely, and the amount of training time is on the order of what went into SDXL originally too from what I can tell.

As I understand it, diffusion-based video generation models simply are not casual in this way. They work by modifying the previous frames in the video to be consistent with future frames just as much as they do later frames to be consistent with earlier ones. That's why Yann LeCun can argue that they do not have to be able to generate plausible continuations of a real video, just generate some arbitrary sample from the space of plausible-looking videos, and that the latter does not imply the ability to do the former. It's also why it's not possible to just generate videos of arbitrary length and lots of VRAM is required to create even a relatively short clip.