Is that really a bad thing? There are certainly plenty of examples of old things that suck; either because genuine improvements became available after they had already solidified or because they were always broken and are now running purely on denial-fueled risk tolerance; but, in principle, it seems like it should be a bad thing that age is seen as a bad thing. Especially when software is more like math than like civil engineering in terms of the tendency of its materials toward corrosion, embrittlement, and fatigue. (and when so many 'modernization' projects turn into expensive failures or go way behind schedule and over budget to eventually death march toward feature parity, sometimes even achieving it in time to be declared legacy themselves.)

I'm not calling for a crusade against 'fast fashion' software; if people want to bang out an app on the fast and cheap to catch the moment when people care they can do that; fine, whatever; but it seems like software built on real long term service timescales should get a lot more credit than it does. Absent specific criticisms; it's not "eww, there are people who weren't even born then", it's "the software has been in service for a generation".

All the more if there are a lot of outfits doing the same thing: having some unique oddball legacy thing means having potentially crushing maintenance requirements unless everything was gloriously secure from day 1, which it probably wasn't; but if there is some big mass of enterprise Java 8 why should we call it all eol and scramble rather than just maintaining java 8? Especially when we can do so in software, without some of the vendor and hardware inflexibility you see with things like old school mainframe applications where there's an implied commitment to a single old school mainframe vendor in perpetuity.

It's not elegant; but realistically we are far enough both into the history of computer science and the history of computers-as-hardware-you-can-buy that there's a lot less obvious, low-hanging, progress to be had by going 'modern' relative to the amount of fashion and fad chasing. Especially if (as is the case for a great many people and organizations) the scale of your problem has grown at or below the rate at which hardware advances have made systems not particularly well designed for scalability faster.

Obviously this would require coordinated action, and some people likely have other reasons to want to either poke at or kill legacy applications; but(since all those java versions are solidly post openjdk) I'd be very curious to know how the cost and risk associated with "modernize because java 18 is going eol!" would compare to just...not...having java 18 go eol. Unsexy maintenance project that you'd need to pay to have done, sure; but very plausibly better characterized and lower risk than trying to deal with a lot of the oddball internal accretions that would otherwise need updating; and, depending on how much people have running on java 18, certainly possible that they'll individually spend a fair bit more running the treadmill than it would cost to just keep kicking java 18 down the road until (almost) nobody cares.

My response was specifically to the original poster who, for some reason, was taking a "we are losing the class wars; breed faster!" position rather than the "if you are already losing the class war why would you even think about putting in that much effort and cost so your children can deal with a bad or worse outcome?" position.

It would honestly not surprise me if that is a nontrivial contributing factor: If you aren't emotionally invested in children as an end in themselves the wage and cost of living numbers have done very little to encourage you to see them as affordable since roughly the late 70s(with a combination of substantial stagnation for anyone who is primarily wages rather than capital gains; and such good news as there is mostly confined to people who complete at least undergraduate education and remain in a career track full time) and people who are emotionally invested in children are often willing to go to considerable lengths to try to improve their children's outcomes; but are presumably discouraged by the prospect that they will most likely be downwardly mobile instead.

It's not a surprise that people who want labor, cannon fodder, or taxpayers to be abundant for them are fretting about it; but it's hard to see why most of us should care. Why do things that are good for society when society is pretty overtly disinterested in being good for you? You may be able to squeeze the current labor market a bit; because people who already exist tend to take the "or starve" possibility pretty seriously when deciding what they will put up with; but if you offer nothing but the demand for a toiling underclass to encourage people to have children that's not terribly compelling, either for those who aren't interested in sacrificing for children and see hitting education and career hard as increasingly existential or for people who would sacrifice a lot to better things for their children but are more or less accurate in seeing it as highly unlikely that they will be able to.

I'm rather doubtful of the plan's viability; but if they are willing to justify the company name by chaining Bezos to a rock so a bird can eat his liver repeatedly I could be persuaded.

I'm not sure what confuses you here: everybody has an EoL, age related if nothing else, and they don't live on either through their class or their descendants. What do you gain by throwing more poor bastards into a sucker's game when you could just not?

Yeah, it's ultimately a matter of taste and what becomes economically and strategically feasible; but I figured that there is at least a conceptual distinction between weapons that are better at being obedient(basically any attempt at stable aerodynamics on the low end up to electromechanical gyroscopic stuff, to TERCOM and GNSS guidance; but you specify where you want it to go and the guidance system attempts to minimize or counteract outside influences), weapons that can independently follow a very specific instruction(at least the simpler acoustic and IR seekers where you need to point them toward a particular loud/bright object but they can compensate for it moving, to a degree); and finally the ones that can take fairly generalized instructions of the "anything that looks like a target in this area" flavor; which seemed like the best candidates for 'autonomous'.

I realize that it's a better hype piece if we assume that the 157000 'agents' are fancy chatbots; but this is just an HTTP API; on what basis are they distinguishing between 'agents' in the currently trendy sense and someone's little script around cURL or whatever?

I, for one, completely trust this objective and disinterested report from a company that has no stake in the ease of ramming datacenter approvals through.

This is obviously a matter of degree, and what you feel like calling a 'drone'; but it seems implausible to say that it's a 'first'.

Something like a Mark 60 CAPTOR entered service in 1979 and is an enclosure for a torpedo that uses acoustic sensors and onboard signal processing to decide if/when to launch the torpedo which then homes in on whatever its acoustic sensors deem high priority. The human deploying the mine defines the search area since they control where it is placed; but everything after that is pure killer robot.

A slightly more recent system would be something like the Bofors/Nexter Bonus, entered service in 2000. 155mm artillery shell that releases two submunitions that use IR and LIDAR to look for vehicle signatures and explode to send an explosively formed penetrator into them. The artillery operator defines the search area by choosing the shell's path; but once they pull the trigger on it selection of who/what in the search area gets a dose of top attack is 100% automated. The German SMArt 155 is a similar concept with similar development and deployment dates.

Obviously how cheap and easy to deploy it is makes a vast difference in practice; there's a certain amount of restraint imposed by something costing $100k+ a pop and being manufactured in boutique quantities that is not imposed by vastly cheaper systems; but it's not the killer robot aspect that is novel.

Obviously pious concerns about fraud are...not...the motivating force here; but I'm curious where on the scale from 'secondary but real' to 'frankly absurd' the burner-powered fraudsters actually live.

It's not like bad prepaid phones are expensive; but, especially if you are actually burning them with any frequency, they aren't really cheap unless you are doing some sort of scamming rather more lucrative than spamming people about nonexistent aftermarket warranties with sub 1% response rates. Are actual SIMs, or even actual phones, remotely competitive with the VOIP equivalent of bulletproof hosting if you want an in to the phone network?

In a similar vein; what's the breakdown of phone-using criminals between people who actually go to the counter and pay cash, where the FCC now wants them carded, vs. the various PO box companies that tend to show up on weird phone charges? It's not a surprise that they are running with the excuse; but the idea that telcom enabled crime is actually substantially the domain of something as clunky as burner phones/SIMs, rather than more efficient services that nobody cares enough to chase down, seems very implausible.

Basically any service you can think of only costs as much as it does because there are limits to how much quality and reliability it actually promises. Electrical utilities tend to keep the grid pretty stable most of the time; if you want better than that you end up talking to Eaton or similar and running increasingly involved onsite equipment; just as people who want internet access to be very reliable rather than mostly reliable end up buying redundant links.

I wouldn't be surprised if there are cases where it would make sense for the utility to operate and sell the additional reliability, rather than the customer DIYing it, whether because there are grid topology things they can do to get the result more effectively or just because they have greater experience with alarming AC gear; but that would be a tier above the standard offering, not a concession that it's reasonable to run the entire grid at the level of the worst-case customers.

You could get into the same argument about water. Hospitals and precision chemistry applications often have fairly elaborate onsite setups to provide sterile or ultra low ion water for their particular requirements because that's not the standard to which utility water is normally held. In theory you could shuffle around ownership and responsibility for the additional processing steps, and in some cases it might even make sense; but it's not terribly compelling to run the entire water system as though it is being piped into a burn ward or a chip fab; and, at least in agricultural areas, there's often another tier below the 'standard' for non-potable irrigation where you can worry less about microbe counts and whether there's matching sewer capacity because it's just getting sprayed on fields.

I suspect it's a straightforward incentives problem. If you can get away with making it the grid's problem there's not much incentive to pay for more expensive facility power setups. Presumably this is why ERCOT is testing current and prospective customers and making noise about it; and why there are at least some standards for how ill-behaved a load can be while still being allowed to hook up; with some awkward interactions between very large sites that also have the ability to shut down rapidly at relatively low cost. If you are 'mining' crypto you presumably prefer the gear to be online because it is depreciating by the minute regardless; but the risk and inconvenience of shutting it down and booting it up again isn't particularly dramatic compared to having to cold start an aluminum smelter or something.

I'm sure that there are worse options, probably being actively considered since this is no longer getting them what they want; but an opaque 'public/private partnership' slush fund that spends its time slathering a thin layer of dubious military justification on random projects seems like a very, very, dodgy way of doing things.

Obviously the switch from "loss leader on a scale the capital markets can barely absorb" to "losing money" is going to sting; but I'm curious if we'll see sneakier knock-on effects.

So long as they were losing money hand over fist the vendor does want to throw enough tokens at you to make you feel like you are having a good time; but as few as are required to do that since they lose money on every one. If they were breaking even or turning a profit the incentive would be to sneak as much spend and upsell in as possible; and it's well known that the verbosity/cost of LLM chatter is hard to predict; harder if there are multiple models and other complications being switched around in the background.

What sort of exciting little tricks will we see from vendors who actually make more if you use more?

Everyone knows that 10x engineers validate inputs and carefully manage dependencies; so clearly there's no risk of unexpected behavior.