I have Nvidia Shield Pro.

I am pretty content with that. As a bonus, I can play my PC games through the Steam Link app in my living room as well, it even takes Xbox controllers.

The problem is, these statistical "AI" models actually do pretty significant factual errors.

If you let it, for example, design yourself a rocket engine there is a fairly high probability that it will explode when you actually build such an engine and switch it on. Most of the engine could be "correct" but all it takes is an error and you get potentially catastrophic failure.

Might not be a problem with a small and cute little toy robot. Might get someone killed with an industrial robot.

So yeah, sure, you can let one of these design you a robot. Or some action the robot takes. But you need an actual human expert who knows what he/she is doing to "proofread" the output and correct the errors in the design or action.

Why would they test the embryos for that?

As far as I understand in the US they would need to test their wallets, because no matter how intelligent a person is he/she is not going to college there if his/her parents can't afford it.

But ... ChatGPT and Midjourney, etc are not really AI as such. So I don't get where are you seeing that progress? They are just large data models based on correlation but do not have an *understanding* of the world.

Just, basically, autocorrect on steroids. A lot of steroids.

That was amazingly good. And I don't say it lightly. Been reading fantasy and scify for over three decades and it was something that was fresh and damn well written.

For some reason it reminds me a bit of Willam Kings book Death's Angels with a bit of Joe Abercombie's Blade Itself mixed in although admittedly neither of these books was about river spirits. I suppose it's the emotion of reading this that reminds me of these books.

Hmm, I'm pretty sure that for THAT price you could get a mirrorless camera and a telephoto lens.

Although it would be a bit bigger than even 7 Pro ;) so there is that.

The point was data pins. So that the new standard PSU's can communicate to the card how much power they can supply. Then on top of that someone figured that would it not be better if there was only a single cable.

Some optimistic engineer optimized it to the bone and it was decided that yep, it can handle 600W alright if the stars align just right and the user does not bend the cable in any way within the visual range of the card.

I'd expect consumer protection agencies getting involved and then proceed to tell to the industry consortium which is responsible for these power sockets that it is not OK at all for things to catch fire.

Meaning that after an investigation it is possible that regulators might withdraw CE and its US equivalent markings from any product having one of these sockets on them, making it illegal to sell or even import any of these into US / EU.

Fire hazard is a pretty damn serious issue.

As a short-term fix NVIDIA/AMD, etc would probably issue a driver/firmware update that limits the released cards to max ~400W. The connectors themselves might have to wait a few years until the next hardware gen to get improved somehow.

Too little pins, too many amps and almost no case in use has enough room above the card to leave 35 mm space from the connector until you bend the cable. The 4000 series are already pretty thicc then you plug in there a connector, that is itself some 10-15mm long, then you need 40 mm more plus whatever the bend takes as its not like these cables can be bent easily at sharp 90 degree angle.