Actually I suspect this verification will roll out to all versions of Android currently pulling Google Play Services updates. In other words existing phones will see this rolled out sooner or later.

And at the same time they can get rid of popular youtube apps that avoid the advertising rubbish. There's no way NewPipe's developer can get a verified dev account.

If I can't run the apps I or others make without some stupid signing process, and side load what I want, what's the point of even using Android? Might as well go to an iPhone and have a fully curated "experience."

As for Europe, this is definitely malicious compliance. It does technically comply with EU demands even as it thumbs its nose at the intention of the regulation. The EU will have to continually fine-tune the regulation as Google is going to do its level best to end run around it every time.

A free tier is great and all, but suppose I want to make an app for others to side load that Google disagrees with? They will obviously pull the verification and prevent anyone from using my app. And if an app becomes popular, reglardless of what app store it's in, Google can bump the developer to a higher paid tier and demand payment.

I have to hand it to Google. They've on-upped Apple on this. They can claim to government regulators they allow side-loading, while still maintaining absolute control over what people do with their phones and they've found a way to charge for every app even if it's not on their store. I fully expect Apple to do the exact same thing. Sweet sweet 30%. Although by manipulating the developer fees they basically charge anything they want.

All in all this is very disappointing and unacceptable.

What's wrong with running a secondary coolant loop at hundreds of degrees? Assuming we have all this energy to play with (run a compressor).

The economics, on the other hand, yeah you wouldn't think so. But you would think Bezos would pay somebody to run some numbers before backing the idea in public.

In the old days I would have said the DoD (now hosted at http://war.gov/ - impressive!?) would surely pioneer this. But these days big tech is so much richer than even the military-industrial complex.

The data transfer part is basically just starlink.

You aren't going to remotely host your gaming PC this way. But for deep neural nets, which have a massive computation-to-IO ratio, that's a different ballgame. E.g. for an LLM the input might be a prompt of a couple dozen bytes, and the output might be a paragraph of a couple hundred bytes. But a lot of computation is needed to get from the input to the output.

(The AI would also need an onboard mirror of any part of the internet it can search for answers and citations).

Everyone expected e-books to kill off dead tree publishing. But it never did. Ebook "sales" never did overtake dead tree. It's still a big thing for some people (and obviously very profitable), but among the general populous, ebooks have never really gained much traction. And I don't know of anyone that reads books on their phones.

Audiobooks seem to be increasing in popularity, however. I'm not sure how that influences this growing trend of illiteracy. I do know more than a few people that don't like reading, or struggle to read, who regularly listen to audiobooks.

So is it your prediction that Waymo is not scalable, and will level off in number of trips per week at under, say, 10 million?

My prediction is that self-driving won't level off before handling a large share of all trips for several reasons: the cars are already somewhat adaptable and will become moreso; that increased trips will provide ever-increasing data to train on; that temporary roadway changes e.g. for construction will become handled more systematically to accommodate automation e.g. waze, google maps, waymo / tesla cars; and that the world's network of roads is actually pretty finite for most purposes.

Don't get me wrong, the average age of a car on the road is 12.6 years, so any kind of transition takes time. And self-driving has a lot of room left for improvement. But it is not facing any insurmountable roadblocks to a whole lot more growth, and running out of training data certainly isn't one.

"no you"!

Remember how waymo bootstrapped their self-driving cars - with sensor-instrumented human drivers. They can still do that as needed. In fact they are doing that in Tokyo as we speak:

https://www.reuters.com/busine...

Of course waymo never did rely on opportunistically scraping crowdsourced data in the first place, so "running out" was never going to be an issue.

AI's model weights (like a person's brain) don't need to store everything - the weights (the bare llm) need to store enough to read instructions, seach databases and documents, and recognize things. Which they already do. Once the AI is "out in the world," it will gather its own subsequent data.

For example, how does a self-driving car 'run out of training data'? They are gathering vast amounts every day from their already-deployed cars. Probably more than they can handle.

Same with call center AI. Where could they get more speech data? Well obviously it gets reams more data every day in the course of doing its job, to get better over time.

3GPP stated ambient IoT devices can gather energy from sources including "electromagnetic, thermal and solar", and operate with "limited energy storage capacity".

What do you want to bet in practice they get all their energy from electromagnetic, and how they get it is being targeted by an RF-emitting reader every time they are read.

I'm surprised they don't mention the aging population. Old people are frail.

Granted, having the hottest summer since the start of records 125 years ago is enough to explain it alone. But aging can only contribute.

At the $500K level you are missing the point. They are looking for people to advance the state of the art and to teach their company how to produce at that level. If training is to be part of the picture this person might be a primary author of it. They aren't looking for expertise they already possess.

Actually, I retract my earlier statement. I had read something about Tinker earlier today on another site, but crossreferencing, it appears to have been wrong.

This product is a 'me too' since there's already tools to help you fine tune various Big Player models.

That's not what this is. This is a tool for creating foundation models.