More to the point, it's unsustainable to just let anyone do anything just because they want to. There's a broad range of behaviors which fall into that non-category, and some of them are fine because they're not hurting anyone and some of them aren't because they are. Right to swing fist, end of my nose, etc etc.

I enjoy vroom vroom noises myself but I also appreciate that there are more important issues at hand than my own selfish gratification.

You literally cannot prevent that in an app which can make consensual nude images. Therefore the word nonconsensual is being used in order to trigger people into having a specific opinion. A better description is "an app which can be used to create fake nude images" since it can't literally show you what someone would look like unclothed.

The interesting thing about that statement is that it's the fascists that have brought us to this point where we have all these highly polluting coal plants.

They are if the end user can charge at home and aren't if they can't.
The intelligent thing to do would have been to mandate charger installation on new construction or any significant remodel or electrical system upgrade, years ago. But we can't agree to do things intelligently because we have to fight with each other over whether corporations have a right to make a profit. (They don't have one written into the law, but they seem to have a de facto one in that they can buy protectionist laws.)

Yes, that's my point. That limitation is crippling, and there was no good reason for the law to restrict their authority to navigable waterways, only a bad and corporatism-serving reason.

Cryptocurrency miners can buy land in a catbox state and put in a solar array. It's going to be cheaper than coal. Then they can mine while the sun shines. This isn't a move to stop cryptocurrency. This is a move to force coal plant operators to do what they claimed they were doing all along. We can find coal plants emitting more than they are legally allowed as fast as we can pay people to sample their emissions.

I disagree with you because I've learned to follow the money.

The phrase was coined in the 1976 film "All the President's Men", which by the way is pretty great if you're into things like that. You do have to have an attention span longer than the time between a SOCK and a POW in a Marvel movie, though.

You are welcome to catch up to the wisdom of over four decades ago.

Twitter only needed to show a profit for one year, and they only showed a profit for one year. What a remarkable coincidence!

You should kill whoever wrote that comment, then.

Quite possible, but Microsoft stole and included Stacker and that also caused quite a lot of data loss since it also sucked.

It actually made sense on the Mac, because programs were officially called "applications" and even had the type code APPL.

So DOS 4 is Vista, and DOS 5 is Windows 7... oh how history repeats

Creating the training dataset is the *last* step. I have dozens of TB of raw data which I use to create training datasets that are only a few GB in size. Of which I'll have a large number sitting around at any point in time.

Take a translation task. I start with several hundred gigs of raw data. This inflates to a couple terabytes after I preprocess it into indexed matching pair datasets (for example, if you have an article that's published in N different languages, it becomes (N * N-1) language pairs - so, say, UN, World Bank, EU, etc multilingual document sets greatly inflate). I may have a couple different versions of this preprocessed data sitting around at any point in time. But once I have my indexed matching pair datasets, I'll weighted-sample only a relatively small subset of it - stressing higher-quality data over lower quality and trying to ensure a desired mix of languages.

But what I do is nothing compared to what these companies do. They're working with common crawl. It grows at a rate of 200-300 TB per month. But the vast majority of that isn't going to go into their dataset. It's going to be markup. Inapplicable file types. Duplicates. Junk. On and on. You have to whittle it down to the things that are actually relevant. And in your various processing stages you'll have significant duplication. Indeed, even the raw training files... I don't know about them, but I'm used to working with jsons, and that adds overhead on its own. Then during training there's various duplications created for the various processing stages - tokenization, patching with flash attention, and whatnot.

You also use a lot of disk space for your models. It's not just every version of the foundation you train (and your backups thereof) - and remember that enterprise models are hundreds of billions to trillions of FP16 parameters in their raw states - but especially the finetune. You can make a finetune in like a day or so; these can really add up.

Certainly disk space isn't as big of a cost as your GPUs and power. But it is a meaningful cost. As a hobbyist I use a RAID of 6 20TB drives and one of 2 4TB SSDs. But that's peanuts compared to what people working with common crawl and having hundreds of employees each working on their own training projects will be eating up in an enterprise environment.

Taiwan reinforces its silicon shield.

This is all to produce a peak of 240k EVs per year. Production "starts" in 2028. It takes years for a factory to hit full production. Let's be generous and say 2030.

Honda sold 1,3 million vehicles in the US alone last year - let alone all of North America, including both Canada and Mexico. If all those EVs were just for the US it'd be 18% of their sales, but for all of North America, significantly less.

In short, Honda thinks that in 2030 only maybe 1/7th to 1/8th of its North American sales will be EVs. This is a very pessimistic game plan.