There are a bunch of problems with PyPI. For example, there's no metadata API, you have to actually fetch each wheel file and inspect it to figure out certain things about the packages you're trying to resolve/install.
It would be nice if they contributed improvements upstream, but then they can't capture revenue from doing it. I guess it's better to have an alternative and improved PyPI, than to have no improvements and a sense of pride.
There is a lot of other stuff going on with Pyx, but "uv-native metadata APIs" is the relevant one for this example.
I'm guessing it's the right PyTorch and FlashAttention and TransformerEngine and xformers and all that for the machine you're on without a bunch of ninja-built CUDA capability pain.
They explicitly mention PyTorch in the blog post. That's where the big money in Python is, and that's where PyPI utterly fails.
Overall the steak is cooling (as the outside air is cooler). At the centre of the steak the temperature will go up first (as heat from the outside of the steak which has been in contact with the pan transfers in) then down again.
Unless I’m mistaken (and that’s quite possible!), this should be trivial to test. Sous vide some meat for an overly long time to a precise temperature, then cut each piece at set intervals.
If you sous vide for a long time, the entirety of the steak will be at the same temperature, while pan searing cooks the outside to a higher temperature than the inside, which results in the core temperature going up a few degrees during resting.
Yes exactly. It all has to do with the thermal conductivity of the meat. Meat tends to be a fairly good insulator which means it takes a long time for heat to conduct through a thick piece of meat (such as a large roast or a pork shoulder). This means that cooking at a temperature above your target internal temp (most conventional cooking methods, not sous vide) will produce a fairly steep thermal gradient between the outside of the meat and the core. Resting the meat allows heat to flow down the gradient to bring the overall temperature closer towards equilibrium.
Another important effect during cooking is the breakdown of collagen fibres (present in tough connective tissues which make some cuts of meat very tough to chew). This occurs in the presence of moisture at temperatures above 170F. When collagen breaks down the meat tenderizes and the collagen itself turns into gelatine which absorbs moisture and retains it within the meat. A gelatine-rich cut of slow-cooked meat (such as oxtail) can be very sticky and packed with flavour. This is one case where food can taste juicier with longer cooking times.
This is not a very quick test because you can hold sous vide food up to 4 hours at the target temp. So you will need to account for that 4 hour gap in all the testing before it starts to break down.
Wikimedia is run transparently which is great but I dont really believe they need the money when you see their financial statement (link below) and think about what they need to run. Plenty of really deserving charities running on the sniff of an oily rag not paying 100m in salaries plus travel, conferences etc.
Keep in mind that the community aka the editors etc are all volunteers so the foundation organizes conferences, hackathons, grants etc for them (not as a compensation, but to help strengthen the community). Keeping "servers running" is only a small aspect of the whole. There's a lot of maintenance work necessary and there are also sister projects as well, like commons, wikidata, etc.
They have 82 million dollars in cash and 116 million in short term investment, why do they need to run giant screen sized popup banners a few times every year begging for money and making it seem like everything will be gone tomorrow unless you donate now? They don't even run these adverts by the wiki editors themselves, just impose them from on top. They are very controversial in the wiki community and always cause pages of arguing every year.
Because you don't have to pay and most people don't, + the reasons from my previous comment.
On the other point: Discussions are at the core the movement, and how to do fundraising "right" and how to use funds is worth discussing and gets discussed. But that it is needed in general is obvious I think. What else should be done? Let all the projects run out of funds and call it a day? That would mean the end - and today Wikipedia is more needed than ever.
As a liberal I've always had to fight the tendency we have to not see legitimacy sinks in the name of politeness. Lately I think people are willing to listen and I'm working on ways to explain this to people who don't bellyfeel them already.
Since I was a kid I thought that the endless fundraising drives destroy the legitimacy of public television. At the bellyfeel level it is visible moneygrubbing, but at a political science level these run side by side with ads promoting the sponsorship of the Archer Daniel Midlands corporation. ADM is notably the prime beneficiary of ethanol subsidies in the U.S. that wreck the environment and make farmers go broke spending money on nitrogen fertilizers that kill off life in the ocean off the mouth of the Mississippi River.
The trouble is that small donations don't give voice, but large donations do.
I can logically justify how I feel about fundraising drives on PBS, but I feel a resonance that causes me to feel the same way for Wikipedia -- I don't know what the Archer Daniel Midlands corporation of Wikipedia is, but it probably exists. Finding out that they don't really the money confirms this feeling.
a lot of engineering positions at WMF don't pay particularly competitively - you do take a pay cut working there to run/manage k8s clusters than you would elsewhere (even some public sector gigs pay better in big cities).
Its not just high income countries. Fertility rates are similar in a huge range range of other diverse countries (Colombia, Cuba, China, Vietnam, Thailand, Turkey, Iran, etc.).
Certainly I will. That's a great analogy, to the point that you could argue he WAS 'black midi', just on a related instrument. He was doing no-velocity black midi on a mechanical instrument.
I thought the main problem with recycling them were the fiber composite blades? If they keep those but just swap the metal tower with a wooden one they've achieved exactly nothing in practice.
Well replacing the tower reduces embodied emissions from the steel. Sure that's not as big as an issue if the steel was already recycled (and would be recycled again) using an electric arc furnace powered by renewables, but the wood is actually negative since it's storing carbon while it's not decomposing.
The blades themselves isn't really much of an issue if you actually compare it to fossil fuels - for example, coal fly ash was 18% of all waste generated in Australia around 2019 (this is likely a bit less now as one or two major coal plants have since been decommissioned).
I think it's astronomically unlikely that wind turbine blades would ever be that kind of proportion of a country's waste, but it was just a normal thing for coal. And gas and oil have a similar problem, it's just harder to see since it's fine particulate matter belched into the air instead of heavier ash that you have to deal with!
If you read the marketing content on their site, the intended method of recycling these structures is to cut up the wood sections and re-use them as structural beams in regular building construction. They aren't grinding them up or melting them down or anything like that. So between the lifespan of the tower and the buildings built (or maintained) with the recycled beams, the useful lifespan of the materials used are long enough to grow new wood to replace them. At least that's my understanding.
The other side of the authors discussion is the use of 'decibels' to describe 'loudness'. The big difficulty there is that 'loudness' is a sense perception that varies between people and in different contexts. The article touches on this 'weighted to mimic human hearing...' but doesnt mention the systems to do this - DB(A) and others, none of which achieve scientific perfection.
Our senses are all like this - for the same reason we have dozens of systems to describe color. And why perfume and wine makers can never agree descriptions.
This is a major reason (maybe the major reason) crypto keeps going up. Many people want money to avoid government oversight - for reasons both benign (e.g. avoiding having money confiscated arbitrarily) and nefarious (everything from tax avoidance to funding crime and war).
Not long ago we lived in a world where currency from anywhere other than the nation you were in (or maybe somewhere close by) was impractical to use on a daily basis. Things have changed now and the government's use of money as a tool to keep control of citizens is loosening. For better and worse.
2025 Chainalysis Crypto Crime Report ... The authors have so far tracked over $40bn of crypto transfers to illicit addresses made in 2024, though they reckon the final total will be north of $51bn.
That seems like a pretty insignificant number? World GDP is more than $100T. $50B is one half of one tenth of one percent, and even that is a significant over-counting because it's a measurement of revenue rather than profit and is counting all transactions to a given address regardless of their nature and potentially double-counting them.
Some drug dealer is making $20,000/year selling drugs, but the drugs are sold for $50,000 because they had to spend $30,000 on grow lamps and electricity and rent in order to produce them. The same drug dealer also uses the same wallet to sell ordinary lawful gift cards for cryptocurrency and they only make $5000 from that but it's against revenue of $200,000 because the markup on gift cards is small.
For that they're attributing $250,000 of "crypto transfers to illicit addresses" to this person but there was only actually $20,000 of unlawful gain. Overstating the problem to demonize the target.
This seems like a big number, until you compare it to the regular economy:
"The estimated amount of money laundered annually worldwide is between 2% and 5% of global GDP, that is, something between US$ 800 billion and US$ 2 trillion."
[0] https://docs.astral.sh/uv/
reply