Forgot your password?
typodupeerror

Comment Re: Longevity (Score 1) 103

I wonder if you could get your speed up by slingshots around the sun or whatever else. Another poster says you can get about twice the speed of the voyagers that way - still pretty slow on galactic scales, but maybe it's the best we can realistically do. Your ion engine would make great steering though - clearly, if you're flying for like a million years, there's no way you're going to be exactly on target when you leave. Keeping any of it working for even a fraction of the journey would be the challenge of course.

Comment Re:Longevity (Score 1) 103

Of course - I doubt you'd fit a camera with an SD card either, even if you were expecting to be out of radio range. I was, perhaps clumsily, making a joke about aliens returning our stuff to us after our probe had been having a wild vacation.

That said, if you could get a craft to autonomously perform sligshots around the sun to get its speed up to get out of the solar system in anything like a short time, then presumably it could also, autonomously, slingshot around some other celestial body to slow itself down or to make a return journey. Of course, it'd have to stay working for like a million years to actually get to another celestial body... ;-)

Comment Re:Longevity (Score 3, Interesting) 103

Yeah, you've got to hand it to 'em, they're doing some good stuff there. Also to the original design that it's allowed for all this stuff to happen.

Whenever we get updates about the Voyagers, I wonder what we could achieve if we actually intended to go inter-stellar. Could we catch them up? What other science could we add? I know it's not very sciencey, but a (modern) camera or two would be cool, and a decent sized SD card so we can see what it's been up to when it gets returned to us ;-)

Comment Re:Using AI bug discovery and AI patches is great. (Score 1) 43

Couldn't they use the free models? It's not really a problem of "giving away your data" when it's all in the open anyway, is it?

FWIW, deepseek is actually a pretty good coder. It's also not bad at review. And it's free to use. I'm not going to talk to it about my personal problems, but for a bit of hobby code it's great. If it's not to your liking, OpenCode has a bunch of other free models you can use instead.

Comment Re:What's a TPU? (Score 1) 22

Maybe, but FPGAs are relatively slow. I'd also say that "inference" is a relatively static problem, and so could be solved by an Asic, at least in a lot of cases. I guess an FPGA might be useful for training, although I don't know much about that process, so can't really be sure. My assumption here is that you can figure out what a revision of Asic will do, make 1000 chips and then crank through work in your datacenter at less cost than it is to buy 1000 GPUs (or even 500 GPUs) - just doing that one piece of work (like my current TPU is good at image work, but isn't good at mathematics theorems, for example) . You can then iterate designs and refine what you're doing. Once you settle on parts of your desgn, you turn gate array into custom silicon, and get a speed/power boost for the next iteration. Asics don't need to be completely general purpose AI solutions necessarily - if you stop needing 10,000 GPUs and instead need 5000 asics, and 2000 GPUs, that's still a net-win.

As for the anchor problem - yeah, I agree. I reckon they'll get 3 years out of most of their kit though - maybe even 5 for some of the stuff they bought a while back. The non-GPU solutions aren't making much headway yet, so it'll take a while for them to displace GPUs. Whether 3 years (or even 5 years) is enough write-off time for the cost and the likely revenue is a different matter though - that maths is so far eluding most people, myself included. Say nothing of the cost of the datacenter itself...

Comment Re:What's a TPU? (Score 1) 22

I have a Tensor chip connected via USB to my CCTV recorder (it cost about $60). It's incredible - it can do all the object detection stuff without breaking a sweat. You could do with with a GPU, but then you need a decent bus, probably a fan, power supply, etc.

How TPUs aren't a much bigger deal in the market is rather a surprise. The main reason is (probably) that Google's software stack just isn't up to it, and so no AI researches will go near it. TPUs up to now also have been somewhat type-cast into being image processing only, I have no idea if that's justified or not.

Either way, I strongly suspect that AI is better served by an asic than it is by a GPU. GPUs were used because we had them, but if someone were to develop a suitable asic, along with the necessary software stack, then we could probably get out of this stupid GPU shortage thing we're in.

Comment Re:I get the feeling (Score 4, Insightful) 48

Probably.

"The nearest parking space available all day is 5 miles from your destination. Would you like us to book an Uber car to pick you up from the parking spot to take you to work?"

Whilst I'm joking about that, it's actually a tactic used by airport parking operators - that is, you 'valet park' you car at the airport, they drive it some miles to some cheap land and park it for 2 weeks. When you need it, they drive it back again. The same could happen here - instead of parking 2 minutes walk from where you need to go, you park 5-10 miles away, and uber the last bit. Those of us with public transport call that "park and ride".

Comment Re: Who. Cares. (Score 1) 105

> Somehow I doubt heâ(TM)s dumb enough to spill secrets to every skirt that comes along.

I'd wonder what secrets he knows, to be honest. He sure doesn't know the details of the chip designs or the research they're doing. He would know their sales numbers, next quarters forecast, cash position, etc. All that's useful for sure, but it's not likely national security.

What he's more likely useful for is waving through a customer without doing all the checks. She might well have got an order of the good chips, shipped to Singapore, but then onwards to Beijing. That is national security, but it's not him leaking secrets.

Comment Re:over 40 million ICE on the UK roads (Score 1) 189

I'd imagine that in half that time there'll be a change to the emissions laws that mean that a whole load of older ICE cars won't be legal or will be so expensive to drive that they'll be replaced pretty quickly. That'll accelerate the switch to BEV. This sort of thing is already happening - lots of cities are introducing emissions-based pricing, so it's only a matter of time before it goes national.

Say nothing about the price of diesel right now - you'd be mad to buy a new diesel car at the moment, especially as charging a BEV car hasn't really changed price in the last couple of years.

Comment Re:electric cars are gay (Score 2) 189

I'll just throw in some extra Octopus details...

Firstly, they've limited car charging via a smart charger to 6 hours per every 24. We now can't fully charge our car from near zero to near 100% in a single night. Not a problem for us, as we charge our car about once a fortnight, but it could be a problem for the road warriors.

On the plus side though, Octopus-enabled smart chargers will soak up any excess energy in the grid, so ours will frequently run for half an hour, or an hour during the day. When this happens, Octopus charge the whole house at the 8p/kWh rate, so if you've got some Home Assistant you can take advantage of that too.

If you lease a new car in the UK, you can use Salary Sacrifice to do so - which is a tax efficient and cheap way to get a new car. If you buy an electric car, you get £500 towards a charger - if you already have one, you'll usually get something else. Octopus Leasing, for example, gave us £500 of public charger credit - so not only do we get 8p at home, we get 0p in some public chargers.

Lastly, we used to have a PHEV. It was great at the time, but with only a small battery, we pretty much plugged it in all the time (and didn't have a smart charger, so it just charged at 26p/kWh every time). It was generally plenty to get the kids to school or pop to the shops, and of course we had petrol to go further afield. Nowadays though, there are so many chargers around you don't need to worry about range any more. What's more, our all-electric car, frankly, burns the living piss out of the lack-lustre PHEV we used to own. I'd drive an electric car over any sort of petrol any day of the week - it's so much better in just about every conceivable way - not least because it's considerably cheaper to run.

Comment Re:Its Going to Suck (Score 1) 48

I'm inclined to agree, but...

1) Google are weirdly crap at software. Everyone thinks they're the ubermeisters of software engineering, but they really aren't. A lot of their creations are pretty craptastic when you look beneath the surface (some also have a craptastic surface, but that's another problem entirely).

3) Samsung are supremely terrible at software. They always have been, and probably always will be. I don't know why, but they have excellent hardware, and terrible software. You'd think being good at one would at least teach you how to be good at the other, but it seems not. Their software is always terrible. They also try to lock you in to their terrible software world, which feels like they're in some sort of corporate delusion, but either way, it doesn't help their case.

3) We have AI now. I know it gets a bad rap, but it is a force multiplier, and maybe Motorola can use it to their advantage. I'd take your advice about waiting for a version or two though - especially at this price.

I have no idea about Motorola's software capabilities. I have a cheap motorola phone, and I have to say, I really like it. They've kept out of the way for most of my life, and added some nice things that actually work (eg. gestures and shakes and whatnot). Sadly, that's not likely a guide for how well they've done at this, but hopefully Graphene is doing most of the heavy lifting, like Android does most of it on my current phone.

Comment Re:It's easy for OpenAI to talk the talk (Score 1) 103

Yes, but they're going to actually pay their electricity bills - I'd say that's pretty noble of them. They're also going to look at ways to minimise their waste of resources in doing their jobs - again, pretty noble.

Honestly, OpenAI and God are so close together you can't see the joins. Why everyone's not queuing up to invest in them is beyond me. Those pesky markets just aren't listening hard enough.

Comment Re:Slop button (Score 1) 19

I'm not sure I see it quite as strongly as you, but unless they've got some decent stuff to recommend, I seriously doubt this AI recommender is going to be any better than the "Because you watched..." or "Recommended for you..." sections they already have.

Talking of not much content... I watched the first episode of Nocturne on Apple the other night. The story of a proper psycho in jail, somehow doing things "on the outside", a long back story of a nervous town, and lots of investigation and intrigue. Shaping up rather well.

Back to Netflix - they just put their prices up, saying "to help us bring you even more great content" (or some such claptrap) - that's not forgotten, so let's see what you've got, Netflix, or else dust off the 'cancel my subscription' button.

Slashdot Top Deals

The economy depends about as much on economists as the weather does on weather forecasters. -- Jean-Paul Kauffmann

Working...