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).

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.

I doubt it was done by a mastermind who knew what the target was and that it would knock thousands of people offline.

But nor do I think it was a "stray" bullet.

My guess, it was just a thing out in the open in an bright color that attracts attention and can be seen from far away. In some peoples' minds that automatically makes it a nice target.

More importantly has anybody tried it (or something that more or less equates to it) and what happened? Economists may be smart but they don't have a great track record for making predictions. (Actually I'd rather say, economics doesn't have a great track record of being predictable.)

+1

Looking into it a little further, the World Bank data on the wikipedia page excludes social security contributions (and some smaller stuff). That doesn't make much sense to me? I'd think 35.3% or so is the one to go with:

https://www.oecd.org/content/d...

Is this story even accurate? I thought tax revenue at 38% of GDP was astounding so I checked it and wikipedia quotes world bank the figure of 27%.

https://en.wikipedia.org/wiki/...

And I can't read the linked article due to paywall.

Still high compared to the US at 12%, although we are accumulating debt so fast it probably should be higher.

I think there is still a place for standardized tests.

You want to see not-inflation?

https://en.wikipedia.org/wiki/...

Sure, polluting is almost always cheaper that not polluting in the short run. The problem is we are also running up against increasing costs from a history of doing that. Getting hotter IS increasing cooling costs, and it IS increasing irrigation needs and costs, how could it not?

One could argue it will also decrease heating costs where it's cold and allow agriculture where there is permafrost. But civilization is built and optimized around what the climate used to be, so the more we change it the more ill-adapted everything is, without MAJOR dislocations, like "Let's say the US will be where Canada is, and Europe will be where Russia is!" That is not so easy to do.