> Some in <country> have X, while other friends in <city> have Y
Cool. What's your message here, though?
I got 10GBit/s down and 1Gbit/s up fiber in a medium size city in Germany as a non-enterprise customer. I even went to buy a capable router and SFP+ network cards for our main PCs and NAS to be able to enjoy it fully. A friend in Poland (close to the border near Cottbus) doesn't have an internet line at all and always relied on mobile data until they got their Starlink setup just recently.
Cherry picking and trying to make a point is cool until it isn't.
I have doubts in their reservation process, though. I signed up for the controller reservation as soon as it started (after they just gave the first batch to scalpers) and it was mentioned that they prefer genuine accounts. My account is over 22 years old at this point and has regular transactions going on. Countless games with countless hours, almost daily activity. I even got the Index back in the day and still use it.
Anyway, upon registration I was informed that my account was "not eligible". You are welcome, I guess.
This has been my dream ever since. Instead of encoding "all the knowledge" into those parameters, how about just making a model that has the same size, but all (or rather most) it does is reasoning? Just give it the ability to browse the net (e.g. language specifications, documentation and best practices) and just have it do its thing. Why does my coding agent need to know the population of New York, know a cheese cake recipe or the general lifespan of an ostrich? Just give it the bare minimum knowledge to think and reason about, and let it figure out the rest.
Sadly that's not how LLMs work, since all they do is "token prediction". At least the models we have to today ...
Is this really worth it, though? Throughout the years my experience with quantized models has been that they feel like a lobotomized version of the original. Doesn't matter if it's an LLM, dedicated diffusion model or some other dedicated task. Sure, they get the job done. But a lot worse. The only ones that can somewhat hold up are the ones provided by the vendor directly. Gemma4 comes to mind. However I suspect they have some secret sauce other than just "let's quantize this" since they have the original model and its data at hand.
There should be more native 4bit, 1.25bit and likewise models. Those actually work great while making them smaller in comparison. But I guess there is some reason for them being pretty niche.
How does this differ from electrobun, which they explicitly mention, but make no point about? I had a quick drive with deno desktop and don't see how it's better. If anything it's lacking in comparison in my opinion. But hey, we can build desktop apps with deno now, too. So they got that going I guess ...
This is some good marketing/news for Chinese models. I see the US government is making a lot of decisions which are in favor for China lately. Probably a smart move given the current political climate.
While I get your point, this kind of gives me "Old man yelling at cloud" vibes. Yes, all the AI talk and bullshit bingo became quite annoying at this point, and I also can't wait for it to settle. But AI is here, and it's here to stay. Wether we like it or not. It's like what dotcom was for the internet back then. We'll get through this eventually - with a bubble bursting here and there - but making fun of it with overtuned phrases like "Everything will be connected to the internet in the future, even your fridge, car and toothbrush!" won't age too well I am afraid.
It sold out in less than an hour. Scalpers are at it again. Units are showing up for as much as $300 in resell-value already. What I don't get is why their shop page just gives me an "out of stock" panel, instead of the purchase button. Why won't they just let me buy and pay it, and have it ship whenever? Is having customers to regularly check for new stock (and potentially missing it) better than ... just having them buy it in advance like a pre-order? I really don't get it. Like wouldn't it even make for a better forecasting indicator when it comes to resupplying?
> So supports structs, methods, interfaces, slices, multiple returns, and defer.
> To keep things simple, there are no channels, goroutines, closures, or generics.
Sure, slices and multiple return values are nice, but it's not what makes Go good. When people think about Go they usually think about channels and goroutines. YMMV
While I do kind of get what the appeal and target audience is supposed to be, I absolutely don't get why you'd choose a subset and still have it behave differently than the Go counterpart. For me that destroys the whole purpose of the project.