That largely depends on where the alfalfa is grown. A lot of the alfalfa grown in deserts is shipped as combination ballast weight for leaving cargo ships and to feed foreign cows and not used for domestic cattle. While nobody around or east of the Great Lakes irrigates alfalfa and is essentially a free product requiring no pesticides and no fertilizer and even adds extra nitrogen fertilizer to the soil. It is just plant, cut, and bail and is otherwise basically a wonder crop for sustainable agriculture anywhere that gets rain once a month or more.
The vast majority of cattle are raised in a pasture and eat grass, even those labeled “grain fed”. Grain fed just means they spend some time (e.g. 30 days) in a feed lot.
I hadn’t heard about cows eating alfalfa though, where is that happening? Wouldn’t it be more valuable to sell it to people instead of using it as feed?
So when there is no rain you let the cattle die? Feed on other people's land? By that logic even server's water can come from rain.
> Around 14 percent of the 3 670 km3 of freshwater withdrawn each year for the irrigation of crops and pasture is allocated to produce feed items for livestock
That's more a testament of how good Qwen3.6 27B is (it really is great) more than how bad this one is IMO. Gemma 4 31B was already good, but Qwen3.6 27B is incredible for its size.
Good models vs bad models are relative: if this was released in 2020 it would be earth shattering. But releasing a model today that's only on par with open-source dense models a quarter of the size and soundly beaten by open-source MoEs with active param counts a quarter of the size is kind of a flop. The niche for this is basically no one. It'll run at near-zero TPS for the few local model aficionados with enough hardware to try it out, and is lower throughout and lower quality for people trying to use it at scale.
I'm rooting for Mistral, I want them to release good models. This just isn't one. It's a little sad since they once were so prominent for open-source.
Who knows — if they have the compute to train this, they have the compute to train an MoE that's 3-4T total params with 128B active. Maybe they'll make a comeback (although using Llama 2 attention is... not promising). I hope they do.
Why don't they monitor average prompt and response token length(both cached and uncached) per interaction. Seems this could have solved all their previous unnoticed degradation.
Also bit surprised they don't have any automated quality check. They can run something like swe bench before each release. Both of these seem like a basic thing even for startup, let alone some product generating billions in revenue.
The site that he bought the crypto from to make a bet could trace it back to him, and many, if not all, crypto trading sites have shady ties with some governemnts around the world.
> I suspect that when things get really big, Google's systems will always be more cost-efficient.
In fact I am opposite of this hypothesis for two reasons. Google has artificially limited production. And because TSMC favours whoever could pay for the most capacity(as incremental capacity is very cheap for them). So Nvidia gets first slot for new process.
Also the second reason is that GCP's operating margin is very high compared to say Hetzner or lambdalabs and you can get GPUs much cheaper there compared to GCP. So students/small researchers are stuck on GPU.
Author took the most simple application and managed to make it so complex beyond belief. If he needed all this for 1 modal, I don't even want to imagine what would he had done if he had 10 different types of modal with 10 different rules as any half complex application has.
Funny I have modals on my mind this afternoon and boy does it drive me up the wall that mainstream React frameworks still aren't using the <dialog> element and without that you will fail WCAG because all the other schemes that are supposed to hide the rest of the page don't.
It isn't even that hard, like I've been able to update the parts of MUI and reactstrap that my applications use even to calculate coordinates for portalized flyovers correctly. I could send but I just don't want to do the 5x work to do all the other components I don't use and then face the politics of a community that probably doesn't care if disabled people can use their products or even if they can check the boxes and pass WCAG. <dialog> is easy to use with HTMX and just a touch awkward with React because React insists on being level-triggered in edge-triggered situations but once you have it coded up <DialogsThatReallyWork/> just work.
I tried to use <dialog> and found it to be a pain.
I wanted to close it when clicking outside, but Safari doesn't support closedBy.
Some Safari versions on iOS broke when trying to style my backdrop with tailwind.
The tailwind CSS reset didn't include <dialog>.
I get the allure of just using a position: fixed;
WCAG is lot lot more than just using <dialog> and unless you do a thorough review with a professional in this field, <dialog> won't solve anything. Do you know how many other WCAG guidelines your app or your site misses. No one knows that.
We do pretty good on Siteimprove and similar checkers although these are frequently wrong and in some cases I think think the specs are actually wrong or miss important things for political reasons. We have customers who send us bug reports and we fix them, we catch others preemptively ourselves.
It is a problem that the experiences of disabled people are erased by the current regime, I haven't once seen an organization actually ask a disabled person if they can use the site or how it can be better.
Also I think accessibility tools are trash. If I start NVDA on my dev machine I have to power cycle it to get control back. Microsoft Narrator sorta works but the more you use landmarks and other aria-markup the more it starts blurting out things like "LANDMARK NAVIGATION LANDMARK!" in the middle of reading something even thought Siteimprove thinks it is all peachy. Is it a fail? Or did the tool fail? My tester or myself can look at an application as a sighted user and test it in Firefox/Chrome/Safari and say "it works" but it is not clear at all what the "definition is done" for testing with screen readers.
I hear JAWS is better than the others but it costs about as much as a car. In the meantime though I know when we don't use <dialog> we fail and I've seen a lot of third party modals that don't use <dialog> that all fail.
It's just plain easy. It boggles my mind that nobody uses it.
But then again it boggles my mind that you can pass WCAG with those "is this a motorcycle?" things or that stupid anime girl you see on Linux pages or a GDPR popup and you can. People will say "what if you have to support WCAG and GDPR?" and I say "sometimes you have to make a choice", I mean a11y work is damned if you damned and damned if you don't, just damned all the time and personally the EU screws up my life a lot more than Iran ever did.
> We are only just now getting a taste of the “true cost” of these tokens
Why do you believe that? Better metric would be price per token of open models served by third party. Last I was tracking the price for similar level model was decreasing by more than 10x year on year, and they are 10-100x cheaper than top properietery models.
Sure you can say that you can't compare them but for sure you can compare the top properietery model of 6 months back to current open models and the gap in time seems to be constant.
No it is not. Yes it could be for your average everyday developer but if someone can run site with millions of active users alone, there is no difference in salary based on where the developer stays. Does Mistral pays $100k salary to researchers?
People working in non profits typically earn less, yes. In general, salary is not a function of how praiseworthy or important or even hard the work is. People who work for non profits have pay cut, because basically they are willing to be paid less in exchange of doing something they see as meaningful.
(Excluding purely "money washing for local mafia and politicians" non profits.)
Dude, Lichess is entirely funded by donations. There's only so much money to go around.
And Thibault iirc is the kind of person that's not terribly interested in earning lots of money. Of he wanted to, I'm sure he could make bajillions elsewhere in tech, because he's that good. But he apparently prefers to only work for a "measly" upper middle class salary and doing something he's really passionate about. And I thank him for it, because lichess is awesome.
Also power is not at all a limited resource as many top voted post on HN thinks it to be. Increased demand decreases the price of power not increases it in the long term.
And in any case ban doesn't make any sense. Instead they could charge different for grid electricity usage, and make the datacenter pay for grid expansion when they start building it.
reply