Re: Making deals with money you don't have
OpenAI is not a publicly traded company though.
820 publicly visible posts • joined 2 Jul 2008
I'm no lawyer, but I'd guess this lawsuit will fail. Unless the contract states exactly what proportion of the monthly payment goes towards paying off the handset, and that the handset is completely paid off at the end of the original contract term (or some specified period), it's hard to see what the telcos could get nailed with. Maybe you're effectively renting the handset, rather than buying it, and so it's never paid off?
This. No alternative reaction seems to be proposed in the paper (or mentioned in the article anyway, I confess I haven't read the actual paper), while what SpaceX currently does allows the constellation to continue operating effectively, so I'm not sure what the authors point actually is. SpaceX don't want to raise the orbits of their satellites unless they have to, because it uses propellant and thus shortens the lifespan. But letting them get dragged out of the sky by a bulging upper atmosphere would shorten their lifespans even more, so it seems like a reasonable tradeoff!
I'm keeping my ageing Huawei Mediapad M5 8" going as long as possible, because there just don't seem to be any decent replacements, the 8" tablet niche seems to have been abandoned by most manufacturers. If I saw an 8" that wasn't a downgrade to the Mediapad, particularly in terms of screen and audio, and didn't cost silly money, I'd buy it.
When an AI band can make it to number one on a Billboard chart [...] it's an insult to the human artists who rank lower.
Most music that's released is slop, that's the reality. Derivative, unimaginative, forgettable drivel, 99% of it. Up until now, it's been human generated slop, but it's really no surprise to find that AI generated slop is quickly reaching par with the other kind. And the AI generated stuff only gets better from here, so if you're a mediocre musician, this is your sign that you're about to get found out. I don't think it's an insult to the human artists; what's happening, or about to happen, is simply a re-valuing of human-generated slop - it's much harder to pretend that generic crap is anything special when AI can generate better with but a few clicks.
SpaceX and Musk called on to rescue China's Shenzhou-20 crew
Called on - by some rando on X, who almost certainly wasn't being serious - a low bar indeed to justify a headline on El Reg. I suppose it's an excuse to discuss the feasibility of such an endeavour, regardless the obvious 0% chance of it happening.
A healthy dose of scepticism is indeed warranted here. However powering datacentres on Earth is becoming a huge issue, and the availability of "free power" in orbit makes it something that I don't completely dismiss out of hand. Probably you're not just going to take existing server equipment as-is and punt it into orbit, that won't work - they aren't built to be mass-efficient or particularly energy-efficient. You're going to select or design components with energy- and mass-efficiency as a top priority, and your physical form factor could be very different. An obvious option would be to base them on something like a Starlink satellite - a Starlink V2 Mini is estimated to cost around $800,000 dollars to manufacture, and generate around 73.4kW with its solar panels, so that could be your envelope - how much and what type of compute can you achieve with 70kW and 800kg of mass? Launch costs per satellite if you made them exactly the size and mass of Starlinks would be a further $680,000 per satellite on Falcon 9 to LEO, that should come down once Starship is in the game, but that'd have to be factored in. It doesn't seem like an entirely crazy idea to me though, I can totally see Falcon 9s blasting off and deploying an entire orbiting datacentre's worth of flat-packed "computellites" in one go. Maybe there are niche workloads that would benefit particularly from being done in space.
I suspect Sean Duffy's blatant attempt at personal empire building didn't go down at all well with Trump or his inner circle; several White House sources have been reported as saying that Duffy totally overplayed his hand, and everybody is angry with him. So that will have opened the door back up for Isaacman, who has many influential supporters.
Have you actually looked at the package? As with the previous package, Tesla would have to achieve hugely ambitious milestones in order for Elon to get his shares. Market cap of $8.5 trillion. Operating profit of $400 billion. 20 million Tesla vehicles on the road. 10 million FSD subscriptions. Etc, etc. It's not as if he's asking for a handout. If Tesla hits even a few of those milestones, *all* Tesla investors will do extremely well. Which is why most supported the previous comp package, and I suspect will support this one too. In a world where CEOs commonly get massively rewarded for failure and cutting jobs, *more* CEOs should have their compensation tied to ambitious success targets like Elon.
You have to go a few generations back to not have AI at all, I think. I have a previous generation Google phone as of now, a Pixel 9a, and I believe it has significant AI features. I haven't actually taken it out of the box yet, but the reviews said while it's missing a couple of AI features that the latest gen phones have, it has Gemini Live and a bunch of other AI stuff. I think even the Pixel 8 phones had significant AI features.
My boss's mailbox is at almost 98%. I've enabled archiving, but not assigned a policy yet. I'm tempted to wait and see if this feature kicks in (it sounds like it's going to be on by default) in a few days. He basically wants emails going back as far as possible in his primary inbox, so this is really what he needs.
...while Cerebras has by no means abandoned training as a workload, that business has taken a back seat to inference...
So Tesla cancelled Dojo to focus on using AI5 and AI6 for inference (and maybe also for training), and now Cerebras has pivoted towards inference as well. It seems like the whole wafer-sized mega-chip approach may not be the way to go, at least when it comes to AI training. If they're pivoting towards inference, presumably they believe their chips are good for that, and there seems to be plenty of demand for the moment. So will the training business gradually be eclipsed by the inference side of things, or do they think they can do both well with the same chips? Or will they develop separate training-optimised and inference-optimised hardware?
Miliband painted Musk as part of a far-right cabal "who want to take away people's rights, take away people's freedoms."
Why not just classify X as a terrorist organisation and then arrest anyone who supports it, eh Ed? I guess taking away people's rights and freedoms is okay as long as it suits your agenda.
I happened to be watching Bloomberg TV the day that deal was announced, or maybe the day after, anyway I've rarely seen the Bloomberg talking heads as nonplussed - as they pointed out, the energy requirement of the deal, if implemented in full, would be equivalent to two Hoover Dams, which they did not seem entirely convinced would even be feasible. If the AI industry is *not* in a bubble, then I guess we're in for some crazy times.
Guessing might produce a superficially suitable answer. Telling users your AI can't find an answer is less satisfying.
This only mirrors the direction search companies went long before LLMs. Google went from proudly proclaiming it was an AND search, and that every result matched all your keywords back in the day, to the situation now where you have to fight to stop it from ignoring your most important keywords and returning utterly irrelevant results. Apparently lots of bad results is better than few accurate results. Same old same old.
...which perhaps explains why the DOE is only committing such modest sums, in comparison with the billions being pumped into the semiconductor industry.
The fusion industry is raising plenty of private capital all by itself, so it wouldn't make much sense for the DOE to pump money into already well-funded fusion companies that really don't need the help. Commonwealth Fusion Systems has raised over $2 billion by itself, for example. By contrast, manufacturing chips in the USA is *only* going to happen if the government pumps a ton of money into it (and uses tariffs and other coercions into the bargain) to make it happen, so I'd say it's more about putting the money where it's needed than anything else.
The fund values each pirated book at $3000, so it's not like they are just paying the cost of the books, most of that $3000 is punitive. However it is a huge win for the AI industry insofar as the principle of training AI on (legally obtained) copyrighted works was upheld as fair use. If that principle is upheld in other cases, that would be possibly the biggest headwind for the AI industry removed.
I'm honestly having trouble remembering all what I used to play on the Amiga. Dune 2, Civilization, A Train and Sim City 2000 for sure took care of many, many hours. Oh, and Syndicate. And I used to do a full length race in Microprose Grand Prix before every actual Grand Prix. Oh yeah, Sensible Soccer, Alien Breed, Speedball 2. Good times.
I think we can take it as read that they're not going to have half their constellation in orbit by mid 2026, but they don't need to in order to start offering services. SpaceX only had around 720 Starlink satellites in orbit when it started offering services commercially, ~6% of its initially approved 12000 satellite constellation, and less than 2% of the later expanded constellation of 42000. So if Amazon could do similar, they might be able to start offering service with only around 200 satellites or less, which seems a bit more doable than 1600 by next summer. So what they might do is start offering services with whatever satellites they've got, then go to the FCC and say, hey, okay, we haven't gotten half the constellation launched, but here we are successfully providing services already, please can we have an extension on that, and please don't take away services from the customers we're already serving? It seems to me, as long as they've got a credible plan to complete the constellation, the FCC is likely to go along with that. Since the requirement to launch half their fleet by mid 2026 is a US regulatory thing, I guess they might be able to position the initial satellites to better serve US customers to help make that case as well, I don't know if that's a possibility.
I did. It was a decent browser for a while. However I suspected from the start that it would get progressively less decent, and indeed I eventually elected to abandon ship - currently using Vivaldi.
P.S. I hope the "Microsoft edge is a Google Chrome downloader" joke is told with a suitable degree of irony.
Retailers used to be required to inform TV Licensing when someone bought a TV, however that was ended about a decade ago. However, I think the general assumption is that every household is going to have at least one TV in this day and age, so they probably send letters to any inhabited address that doesn't have a license.
So does this mean that intermediate-mass black holes are no longer theoretical? Or are they extending the range classed as Stellar-mass black holes? Up to now, I believe the intermediate-mass range of 100 to 100,000 stellar masses had a few candidates, but no confirmed members, so this would seem to be the first confirmed member of that class, unless there's some nuance I'm missing.
I booted a 32bit system from my GParted flash drive just the other day, not because I wanted to use GParted, as it goes, but it happens to have MemTest86 on it too - I was testing an embedded board with a Vortex86 SoC on it, a 32-bit 486-class unit. I'm not actually sure that the GParted option would have run on a 486-class CPU, 32-bit support or no, but I didn't try.
Haven't heard of Ventoy, but it sounds useful indeed.
Waymo has been operating for years, and burning vast sums of cash ($4.4 billion loss in the last fiscal year alone). Robotaxi is just launched and in a limited testing phase. So what, there are a few teething troubles. But given that the Tesla vehicles are currently basically unmodified Model Y's, if they can fix the minor issues we're seeing, it's hard to see how their costs aren't going to be a tiny fraction of Waymo's. And the dedicated Cybercab vehicle will be even cheaper to build and run, when that comes out. Shame Waymo isn't publicly traded, it'd be looking good for a big short right now.
I suspect that they are promoting their reactor as being an SMR in an attempt to have some of the latter's good publicity rub off on them
Mostly agree with your informative post, but the Natrium reactor is considered an SMR because it's small, 345MW, and modular, as in the reactor will be built in a factory and delivered to site, it's that simple. So I don't think they need an ulterior motive to promote their design as an SMR; the term does not define the technology of the reactor.