Posts by doublelayer 11420 publicly visible posts • joined 22 Feb 2018
Remember that the analogy was about stock market activity, not everything they did. But the infrastructure analogy isn't as different as you're claiming. A railroad that built in a bad location didn't leave behind something that someone would take over, they left stuff that would have been used in a different place but was too expensive to retrieve and move to where someone else would use it, so most of that did not get reused.
I'll agree with that a bit. There are at least some people who believe, probably more because they want it to be true than that they think it is, that the unprofitability of AI will be so extreme that it crashes out of existence and we can all go back to not having to clean up from bad LLM junk anymore. These people are wrong, since no matter how unprofitable the big models are, there are lots of small models which are cheap to run. The software and its users will stick around in a similar form, bubble or no bubble.
But I disagree with the only frothy market correction prediction. LLMs face a pricing problem that none of the things you're comparing them to did. They face an absolute cost problem (training cost) and a marginal cost problem (inference cost). It's often practical to find a solution to one of those if the product is in demand, but both is a challenge that PCs and cars didn't have to face. Existing LLM companies used venture capital and funding from big tech as though they were normal tech companies, which means they acted like absolute cost was their only limitation, and this year has shown several examples of marginal cost being so high that they can't maintain that strategy anymore. Microsoft, which has billions to throw at something, couldn't justify continuing to subsidize that product (as much, there's suggestions that they're still doing it to smooth the plunge from functionally unlimited to true cost). If you want a parallel, I suggest you consider railroads. Some of them were and remained profitable and they changed history by existing, but the crashes were hard and many of the participants lost badly and permanently.
There is, but it's not organized. People do buy used memory on eBay, and if you search there, you can find what others are selling. Price yours near the low end for the size and speed of the sticks you have, a lot of those are people who set something ridiculous in the hope that someone would think it isn't.
This mostly depends on what kind of RAM we're talking about. If these are 4 GB desktop DDR3, it's probably not enough to make a difference. 16 GB DDR4 or anything larger or newer is probably worth it. You may have some success selling them inside the computers if you still have them. People who need the parts can handle the rest of them or ask you to split them out.
"If they try something and it doesn’t work, nothing is lost."
That's not how tech, or really anything, works. If they try something and they decide they can't admit it doesn't work, then we have to put up with it anyway, like user agent tracking. If they try something and it doesn't work to protect privacy but CloudFlare decides it does work for site access, then anyone who wants to access several sites has to use it because at least one of them is going to go through CloudFlare. But to be fair to you, you did say "doesn't work", so what happens if it full-on doesn't work for any of the goals? In that case, the only thing we've lost is the need to implement it in everything that speaks HTTP for the next five years until it gets deprecated, because they won't know whether it doesn't work until they've enforced it for a while.
We can only hope that they don't have the same attitude you're espousing. I hope they put a lot of thought into whether it could possibly work at all of the stated goals before jumping to shoving it in, because there's lots more pain experimenting on the public.
Because:
1. Do you want your ISP inspecting your traffic and automatically dropping it if they, and let's be clear by they I mean whatever random program they chose for the task, decides it's suspicious? Me neither.
2. Bot traffic doesn't look that different from human traffic, especially not to pipes through which the encrypted version is passing. CloudFlare and the sites that are using this are creating this specifically because they can't otherwise consistently tell the difference, and they've got the clear traffic which lets them establish patterns that the ISPs don't see.
3. There are still a few places in our modern internet that see it as not their job to police what you do and how you do it. From their perspective, if you retrieve data from a site using a script and the site doesn't want you to, that's for you and the site admins to figure out, not them, because their job is transmitting packets from address A to address B. These are the ISPs I want to use.
Speaking as someone who works in security, I think it's likely the people who wrote that did not intend to be nasty and probably didn't even think that they were at the time. That doesn't speak well of them/us either, but it is different than the image you seem to have about their motivations or attitude. Here's my guess about what went on when they planned that message:
One popular tactic among phishing messages is to promise a big reward. We should have one like that. The basic version is the fake lottery scam, but phishers are getting better at thinking things through so we should too. If this was a private company with jobs structured for it, bonuses would be a good lure, but this is a health authority, they don't do that. Even a small monetary reward would be out of character for a place like this. What kind of reward could be plausible enough to work as an actual test? Oh, I know, they don't have leeway on cash, but they can give days off work. Great, let's put that in the option pile along with that pseudolegal lure and the calendar invite with a link to a Google Drive file and send on of those when we've set up the endpoint.
What you're requesting is not something the security team can do. It may be impossible for the authority as a whole to do it, but it's definitely outside IT's powers.
Your conclusion that the problem will get worse is a puzzling one and I think it's likely incorrect. The people involved are not happy right now, but one thing that makes very clear is that phishing emails promising things can happen. Something like this is very memorable, in fact probably more so due to its unpleasantness. The narrow problem of knowledge of phishing probably benefited more from this than a normal test. As for the extra training, I wouldn't be surprised if they cancel that due to the negative reaction, but even if they don't, you're treating it as far more stressful than watching, or having on in the background while you do something else, a video before answering obvious quiz questions is in reality.
Your question has two independent elements. Why isn't there a single list of approved sites, and why are users responsible. Let's take the easy one first. Users are responsible because their actions can cause damage. That's true even if you have a perfect list of approved resources, as there are plenty of other things a phishing message can ask users to do which wouldn't be blocked by that. Users can cause damage and training is one of the ways, hopefully along with other technical measures, to try to prevent it happening as often.
Now why don't we reject anything outside a list? There are three possible reasons: IT is too incompetent to set it up, it's not feasible, or it would anger the employees whose anger is the reason for your original complaint. Usually, the reason is a combination of 2 and 3. It depends what things people have to access for their job and the costs of blocking those things, and it depends what employees are willing to put up with. Sometimes people need to access unknown sites to get information, and sometimes they don't strictly need to but want to and dislike being prevented. So tell me, which one is it reasonable to do to the overworked underpaid etc employees: send them training messages, or force them to fill out a form every time they access a new website?
How does that make the training out of date? That's the main reason for the training. People are already prone to accepting things that come through because they arrived in the right place and they're supposed to act on messages they receive there. One of the most valuable things phishing training can do is to make sure people are aware that they're going to get faked emails, some tools they can use to identify them*, and some reassurance that they should do that and it won't be held against them.
* The tools won't always work, but they often do. If an attacker has already managed to compromise one internal email, then checking the email addresses won't help, but it's hard to get such an account and many attacks don't bother, so checking addresses is a good thing to do and usually works.
Their proposal is helpfully vague, so it's not clear what level of computation they're looking for, but from what they did ask for, we have those already. CPUs are not the hard part; we've got mass-produced chips at all sorts of levels of speed, memory, and power consumption. Software's not the problem; we have compilers that can fit into the space on the larger ones and we can always write simpler compilers if we need to cram it onto more restricted chips, though why we need to is a question worth asking before telling someone to get started. One limiting factor if power consumption is the important element is the interface hardware since even e-paper takes a bit of current to refresh. Still, if someone wants a low-power computer badly enough, they definitely can make it if they're clear about what limitations they're willing to accept to reduce the power consumption.
I think you'll find your assumption is often incorrect. There are a lot of things, in games and other software, which are licensed, the non-exclusive distribution license kind. Only some things are purchased and completely owned by the people creating the program. Many libraries and some assets are licensed the other way, and you don't need the creators of those things as extra enemies when, at the moment, most of them don't care and some of them are on our side*. If you were under the impression that most software is completely written in house and by direct contract work, you're underestimating how many tiny companies producing things you never see are part of this industry.
* Sometimes, they're on our side for selfish reasons. For example, if someone builds a game and licenses music that's already been written rather than hiring composers to write new scores, the person who licensed the music non-exclusively stands to make more if that license is perpetual or needs to be renewed than if they sell a few years and the game is canceled if it's not popular enough to cover the renewals. Although that's not the most sympathetic of reasons to support the idea that games shouldn't exist at the whim of their author, it does help obtain that point so is better than pushing those people to the game companies' defense.
When you're considering what complaint to make, consider what discussion your complaint is being inserted into. I didn't say, nor do I believe, that this was the reason you should accept that games should stop working whenever their authors feel like it. I said this was a reason that trying to end copyright for a game was a bad way of solving the original problem. We agree that companies shouldn't be able to sell something as a permanent sale then take it away. I am willing, though probably others are not, for them to sell something they'll later take away but only if they tell you that and a minimum death date before you purchase it and in clear language. If you try to end copyright to solve this, it probably won't work technically or in the campaigns, but if you try to regulate accuracy in advertising and criminalize theft through unilateral contract changes, it's more likely to work.
I see, you're choosing the all assets involved lose copyright option. You can support that, but it's going to be much more difficult to pass than the original proposals are and you've just handed the evil game designers a powerful new weapon to aim at you. Before, it was the normal people buying games versus the companies that made them. Try as they might, the game companies were the unsympathetic side, because after all, it was them who sold you something and tried to take that away and it doesn't take much intelligence in a politician to understand that story. With this change, though, the game designers take a step back and find developers of physics engines, artists who earn all their money through game work, and people like that who will tell the politicians, honestly (if they only pick independent suppliers) that they did nothing to take your game from you, but you are doing everything you can to take their livelihood from them. Now, they're the sympathetic side and politicians, for whom game regulations are not high on the list, have more reasons to ignore you. There's already enough difficulty getting the original proposals passed. I'd suggest you don't make your life harder by asking for something extreme you're almost certainly not going to get when you could solve the problem without it.
How is that? If that copyright is not also released, then it's a problem between you, the person who wants to distribute the unlocked game, and the authors of that library who want you to stop and have the legal means to punish you, whether or not I (following your lead, the person who made and then broke the game) care or whether my copyrights still exist or have weight. If it is, then you have angered a bunch of people who can convincingly explain that they didn't cause the problem and shouldn't be punished for it, and politicians are going to care more about the person who says they spent a lot of time building that over the person complaining about a game, so you'd better plan out the alternative rather than expect that to pass. That you don't care about them doesn't make problems go away. Are you proposing that the copyrights for things not written by the game developer lose or retain copyright? Either way, you'll have a problem, but you do, at least in this hypothetical, get to pick which one.
It probably wouldn't work and it would cause plenty of extra damage which would get that reversed or blocked. If you're willing to look online, you'll find people who are already doing that, DMCA or no DMCA, and in practice, the company that ended it is not trying to fight them because they EOLed it, they don't care too much. Those people don't always succeed, proving that you may need more if you want to force that to be successful. If the server's doing much more than activation, then you also need to reimplement that behavior on something else, and copyright isn't the limiting factor to doing it since you don't have any of the code that ran there or the docs on how they communicated with each other.
Meanwhile, there would be more angry people than the original game devs involved. Many programs include licensed code, assets, etc developed by other people, and if you don't remove that copyright, then you can't distribute those pieces. If you do, now you've stripped copyrights from someone disconnected just because someone who used their data took an action. That gives you many more people to fight against your proposal, and more sympathetic ones at that because they didn't choose to take away a game. That makes this suggestion even harder to pass.
You're being too literal. The complete analogy was to medical equipment, some of which does beep a lot, but also, my computer does beep a lot. True, it's probably more accurate to replace "beep" with "ding", because that's what my Slack sound is, and "bong", because that's what the paging alarm sound is, and silence but little notification icon, because that's what another alarm system does. There's a lot of notifications of events, and when there are too many of those, it makes people less good at identifying real problems and handling them quickly.
Determinism is often observer-relative. If I have software which runs exactly the same operations, but whenever you ask me for output, I put in different input, then although the software is deterministic, the process you're using which involves the software is not because I'm adding nondeterminism and you only get the combination. Only one part of a system needs to be nondeterministic for the whole thing to be, and it only becomes deterministic again if you get to take it apart and use only parts you want.
That is why most LLM systems in active use are nondeterministic. The simple reason is that users don't get to set the temperature to 0. The more complex one is that system prompts change, you don't get to see them, choose them, or know when they've changed. For that reason, identical prompts on identical output using the systems most in this situation are using should be expected to give different output, and they do. That's nondeterminism, even though one part of it is deterministic (assuming aforementioned race conditions don't exist).
The main problem with human in the loop is almost what he said. I'll be charitable and decide he was saying it but phrasing it weirdly. To have a human in the loop, the human has to know whether the thing the LLM wants to do is the right thing to do. The problem is that doing that usually requires repeating the work the human would have been doing if the LLM wasn't there. Only a few steps of that process can be removed, and only with trusted automation in which category hallucination-prone LLM steps is not included. That's slow. So instead, what most "human in the loop" setups end up involving is that the human reviews what the LLM wants to do to check whether it's completely bonkers. To make this more formal, let's rename that "implausible". If the action says delete the database, they label that implausible and block it, but if it's to update a record which seems related, that gets approved. The risk, of course, is that smaller incorrect actions will slip through, and that, as he correctly points out, people will get worse at deciding what to approve if they're stuck in a cycle of review and approve all the time.
His suggested alternative doesn't fix any of this. Instead, it removes the checks for implausible actions but blames people when they run anyway. They didn't do the work to figure out whether the action was correct because he told them not to, they didn't scan to figure out whether it was going to go off the rails because he told them not to, but they're still to blame when stuff goes wrong. This might be an acceptable policy if they thought up the idea of the LLM that did the damage, wrote it, and used it by their own choice and without any pressure. If any one of those things are different, blame is no longer correctly assigned. If it was someone else's idea to use an LLM, they may be using an LLM for something it can't manage even some of the time. If they didn't write it, they have much less information about what it's going to mess up. If they didn't choose to use it, then they're being blamed for a decision to trust a program which someone else made. If the person who executed a command is blamed (correct choice) because the computer in the middle is not capable or permitted to countermand that decision, then the person ordering the use of the LLM is to blame when they order it used without review without permitting the person they ordered to run it permission to do it safely.
You seem content enough to post in this one. I don't think it would have much traffic with proven and public identities required on every post. Similarly, open source, the thing you cover and support, would be substantially weakened with that as well. You don't have to care if that's your opinion, but when you say these things, it really sounds like you don't recognize what would happen without the level of online pseudonymity we currently have.
I'm not sure what definition of "planning" you're using, but using the one I've used for some time, yes, they are. Planning something doesn't mean they have a perfect way of implementing it without any gaps. Planning just means they want it and have an idea of how to start, they do, and they're executing that. We can hope that, after they do this, they get bored and don't go any further, because I can easily come up with much more effective, though also not foolproof, ways to accomplish their goal. I'll not tell them what those are in the hope that they are too stupid to come up with those themselves, but that's unlikely if they really want their goal to be more successful.
You could dump things in one place, then when you're going to build a base on the moon, intentionally put it right next to that location. Benefit: people can go out and get whatever was dumped as easily as possible, only a benefit if it's practical to recover anything useful from that but I think we're all assuming that for any of the rest of the hypothetical to work*. Cost: after that base is established, you'll have to dump new things somewhere else to avoid pelting it with shrapnel. If the base has a major secondary industry as a recycling facility, that could make sense. Otherwise, probably not.
* Since we mostly don't have that ability, even here with all the nice heavy machines we can't operate on the moon, there's reason to question this idea. This strikes me as one of those ideas that works if we assume that new tech will just start existing because we want it without considering its feasibility. This is only valid in science fiction books, and even then only at the beginning when the setting's tech level is established. In other situations, especially when it's someone's solution to why they don't have to do something properly, it's unconvincing pseudofuturism.
The point of Google having written the code is that they should have the access and ability, and we know they do, to remove the Android parts, strip it down to the bare Linux kernel on which Android used to be running, and replace it with the version they've built here, even on hardware they didn't manufacture. I agree that you wouldn't want to keep the rest of Android while using it this way.
I think it's partially that they intended to run the CPU at high load for a long time, which would make a lot of heat. Android has software intended to limit that, and the phones were built around the assumption of that software, so running something hotter than it would normally go next to a battery that has already seen a lot of wear might eventually cause something to fail, and if one of them catches fire then having a bunch more batteries right next to it is not ideal. Neither happens normally in your pocket. It's likely neither was going to happen in the server rack either, but since the batteries weren't useful, taking them out made that risk zero instead of low.
High-end mobile chips are somewhat valuable, especially when they're being discarded so are cheaper than if you had to try to buy them from their manufacturer. Since they're talking about devices old enough that people want to replace them, the battery's value as a battery is mostly behind it and, since you don't need it for the compute part, it can be sent for lithium recovery. The screen is trickier to reuse anywhere else, but it's true that a lot of money went into making it and probably won't be coming back out.
They did answer your question. With no cooling, the chips will rapidly throttle and consume less power. With some cooling, maybe possible with passive heatsinks if there's a lot of space between boards and the room itself is cool, less throttling. With active cooling, no throttling, meaning more performance, more power consumed, and even more power consumed by the active cooling. Therefore, how you cool it makes a difference to how much power is consumed.
But fine, let's use some numbers. The Google Tensor G2 processor has a TDP of 7 W. Add some extra for the RAM and storage chips takes us to 10 W. So the boards themselves could consume 20 kW altogether, about as much as 6 of those Gigabyte pizza boxes assuming the power supplies in those are redundant. There are 12 cores per phone, so 24k in total for the Pixel cluster, and 40 nodes per box and 8 cores per node, so 1920 cores (but much faster ones) in total for the Gigabyte cluster. There is 12 GB of RAM per phone, so 24 TB across the cluster, and 32 GB per node, so 7.7 TB (but more available per process) there.
I suppose it depends what they do when a student submits something in both cases. Since they weren't clear about what they're doing on the phones, let's start with what you were doing. Was that just copying the submission for later grading or something more? For example, if your submission script was just storing the submission for grading after verifying that the files matched what you expected, but their submission script runs a few thousand test cases and a plagiarism check against all the other submissions and a bunch of gathered possibilities, it makes sense that theirs would be more compute-hungry.
I think it's greenwashing for a different but probably similar reason to the original poster. If I want to do that with old phones I've got access to, will I be able to? Will Google, who wrote the Android code on them, find a way to let me reuse that hardware, or does that only apply to theirs? But fine, I can't blame Google for Samsung not unlocking something or providing required code*. So if I've got a bunch of old Pixels, will they give me this? Or is this perhaps only going to be available for one user and one device, a device that incidentally still receives updates so they didn't need to be recycled anyway? Greenwashing is frequently done by doing something that would be useful but only to the extent necessary to get some articles published, not to the extent that it makes any difference.
* Actually, I can blame Google. Let's say Samsung refuses to let me unlock the bootloader. That's not a problem, as I don't need to. The Linux kernel is already there, complete with driver code for the contained hardware. I just need to delete the Android service stack above it, and Google software running as root has the power to do that, as do I if I have successfully rooted it. Once done, Google can provide the software that runs on top of that kernel. So, will they?
"Were she alive today, Mrs. Fowler would have opined that spell checkers and grammar correctors are for people who (for whatever reason) have not read enough."
Even if we accept that as true, there are at least two participants in any written communication and the reader can benefit from having things properly written even if it's the fault of the writer that it wasn't originally so. Someone who doesn't know how to spell but uses a tool to do it properly at least delivers something readable. Someone who doesn't know and doesn't have the tools to fix that without improving, unless and until they improve on their own, produces something that takes more time to properly understand.
Of course, such a complaint would be illogical for someone who told a lot of people they were writing badly, as people will learn from reading whatever style the writers used. While Fowler was less pedantic than some of his* contemporaries, he still opposed and attempted to define as ungrammatical things that people were frequently using in conversation and in writing. If those things were incorrect, then reading would not dislodge them.
* I have to wonder whether we're thinking about different Fowlers. Henry Watson Fowler was the only one that jumped to mind.
Cooper was trying to find some reason why the ignorance was acceptable. It's not necessarily that hard to imagine how that could happen, for example that the filler-inner did not use word processors during their education, worked a job that didn't involve them much such as something manual, and was therefore frequently using word processors for the first time in this new job later in their life. Your suggested alternative is just ridiculous, that they had used word processors for decades and were very familiar with them but somehow avoided having any familiarity with a very common feature in them for decades by virtue of never having misspelled a word, correctly spelled a word the word processor didn't recognize, or received a document in which someone else had done either of those things during their decades of thorough use. Cooper's assumption is possible. Yours is not.
Come on. You know what these terms mean. Code that exists right now under a permissive license is less restrictive on those who use it because I have fewer restrictions about what I may do with it at will. You think the more restricted version, under which I can't use or modify it without distributing my changes and a few other requirements GPL has, is better, but you are having to go to increasingly desperate and futile ends to try to define that as less restricted. Your latest one appears to be to claim that the code is what's restricted or not, when a) you know that everyone else is talking about the developers and b) developers are things it's possible to restrict and code is not since it doesn't make choices. The license doesn't restrict code's content or actions, it restricts developers' actions, and GPL puts more restrictions on those than MIT does by design.
If you argued about the benefits of those restrictions, it could work. Plenty of us either agree with that in all cases or at least in some (I use one of a few versions of the GPL on code when I want those restrictions to apply, though not always). The struggle to contest the correct use of "restrictive" by ignoring the thing it clearly means and trying to add something it doesn't is not helping you. Restrictive doesn't mean bad or good, it just means restrictive.
I'd suggest everyone calms down with the insistent terminology, especially when they're trying to warp words way out of their definitions to support their preferred license. For example:
Dan 55, stop pretending that GPL is less restrictive. It's more restrictive. You support its extra restrictions. You can easily argue it's better because of those restrictions, but not that there are fewer restrictions.
ChoHag, slavery involves a lot of forcing. GPL doesn't. It only affects you if you use stuff you were given for free which you have many ways not to use or to replace with something with doesn't use the GPL. While it can be annoying not to have as many rights with such a thing as you would want, you are not forced to use it, and you are not forced to use the GPL on things you produce if you don't want to.
Your theory is that non-text is rarely if ever efficiently diffable. This is not true; plenty of binary file formats can be edited in ways that only change some of the bytes, but I'll grant that in other cases, especially anything encrypted or heavily compressed, you can get exactly what you're describing where a tiny change still changes the file significantly. I think you'll find that most of the binaries people check into source control don't tend to be either of those things because encrypted files are hard to read for the people working on them and prematurely compressing assets causes problems if people are going to edit them so anything people expect to change is often stored in a raw or lossless format, but it's certainly going to happen some of the time.
Why does this matter to the original point, though? The storage and transfer process can easily try diffing the files to see whether it will be efficient, then proceed accordingly, whether the files being compared or stored are text or binary. In fact, that's exactly what their system seems to do, and their reason for not using Git is that it has limits on how well it can do that: "multi-gigabyte files are still handled out of band via LFS rather than as first-class content; [...] deduplication of binary content is bounded by pack delta heuristics". The advantages of a line as a unit don't have to affect the storage and transfer process and can be rendered at the display level.
If it was online, the everyone taking it at the same time option with questions unavailable until right then would become more practical. In fact, it is in person at a fixed set of testing locations. Still, I'd think that the extra cost of making people take it at the same time across all those locations is less than state-wide internet blocks.
How would they do that? Find out, and that's what you prevent next time. If it's important enough to switch off the internet for millions of unrelated people just to make sure they don't hear a rumor about cheating that's not even true, it's important enough to take a few steps to prevent actual cheating, right?
In my opinion, it would be less drastic to lock all those writing the exam into a hotel from the time they start writing to the time the exams have been taken. I don't think they have to go that far either, but we're dealing with some rather extreme reactions, and at that point there are lots of choices that work better than turning off one messaging app, which will do nothing, or switching off the internet altogether, which will but even more painfully.
That wouldn't prevent any of the things they mentioned in the article. It probably would prevent a lot of ways people might actually cheat, but when they go this nuclear out of someone doctoring an image to suggest cheating which they themselves declare wasn't real, who knows what they would do for someone actually cheating. They might actually subject test takers to an EMP as suggested by comments two weeks ago.
That's true. I suppose I should have said "unable to disable things and leave them that way if that's what they want". While I can recognize that there probably are people who want that level of blocking, I find it a bit difficult to imagine it as, if I don't want an app installed and I am allowed to remove it, it's pretty easy not to install it again. Still, since that's not what I want in a phone, I do worry how invasive something specifically intended to block things even if you try to use them will be against someone who doesn't need that, even though I don't think I want to use the things they're blocking. If, for example, they've made sideloading things difficult in order to make it harder to sneak social media in, that probably would be annoying to me.
Mostly people communicating internationally on a plan where either they or the people they're talking with will be charged international call or SMS rates. Combine that with people who simply decide they must use that, so if you want to talk to them, you also must use that. I currently don't use it, but I can understand why others want to and therefore reject a device that can't.
The OS is like a smartphone, the hardware is not, which will introduce some problems, and the software they've installed on top of the operating system might make things worse. I don't know as I've not seen it, but this is not the same as Sailfish, an operating system intended on building a smartphone, it's that plus software written by people intending on making a phone for people who don't like being distracted but are unable to disable things on their own. Since I don't have a problem not installing social media apps when I don't want them, I do wonder how many things there might be in that software which would get in the way of something I did want to do. Ideally, they just block Facebook, which since I have no intention of using it, wouldn't get in my way, but I don't know if that's likely.
This isn't just Sailfish in the same way that KaiOS isn't just Firefox OS. The code at the base is the same, but everything around it is intended for something completely different which makes some things less practical and some things impossible.
It does support other messaging apps. How well it does that is another question, as it's still designed like a flip phone. There are several AOSP-based things like that around, and you can sideload things onto them either because that's intentional or with a little hacking. I've installed Signal on a flip phone like that, and technically, everything works. In practice, the app isn't designed for a tiny screen, typing and navigating isn't convenient, it's functional if you have to but not as convenient as the SMS app designed for flip phone use, which itself is not as functional as any type of messaging app on a touchscreen.
It depends how you put a cost on that. For example, if my little desktop app* uses 53 MB of RAM when I know full well that, with a few days of effort I could make it run in 29 and, with a few weeks, 8, what is the actual cost to users? It's really easy to do that calculation incorrectly: I'm using 45 MB extra, let's say I have 10k users, I'm wasting 450 GB of RAM, which at the current costs of RAM according to the first search result means $3,600 in RAM. If my few weeks cost less than that, I should use them optimizing.
This is wrong because the extra usage only matters while the program is running and, when my program isn't taking up the 45 MB, they still have it to spare. As desktop users, they likely have plenty of RAM unused, meaning that they won't notice whether I optimized that or not. That depends whether I expect my users to be using typical desktops or laptops with 8+ GB of RAM or low-end Raspberry Pis with 512 MB, so if there are a lot of Raspberry Pis involved, I should spend more time on it. This is why optimization has to be tailored to the environment you're running in. Shaving off a little RAM usage can be an absolute requirement in an embedded system, very cost-effective in a multi-user server system, and unnoticeable on a desktop, depending on how much is being used altogether and how much can be optimized out. Optimizing when it isn't needed can spend a lot of valuable dev time on something that doesn't benefit users.
* Actual example of something I'm writing, numbers approximate. There are some low-end Pis involved, so I am going to optimize it a bit. It would be possible to optimize it far more than I'm going to, though, and I'm not going to bother because the environment doesn't need it and the users of this will prefer the features I could build with the same time.
I think they were responding to your "I can't see a reason why anyone would want the default to be background tabs can make noise". Clearly, they can work with that default. Anyone can. However, for people who are in the intersection of "frequently use background audio in tabs" and "don't have much problem with background tabs playing unwanted audio", having the default the other way means less manually unmuting things. This is my setup, mostly because I don't find background tabs getting noisy basically ever, so the less clicking around, the simpler things are for me. I don't care if the default goes your way, but I would appreciate being able to switch it to everything can make noise unless muted.
You're points are orthogonal to the discussion and irrelevant in any case. Let's start with the disconnect about performance. It's likely true that Manifest V3 is faster than V2. There's an advantage in moving compatible extensions to that. If you run only extensions using V3, the presence of V2 will not slow down page loads because V2 isn't being executed by anything. The page code doesn't do it, the browser runtime doesn't do it, so it only runs if you have an extension calling it. Therefore, the case that web office suites, not fast software in general, would be even slower with Manifest V2 is incorrect for anyone not using an ad blocker, incorrect for anyone who disabled their ad blocker, and not necessarily correct for those who kept it on. It's not as simple as everything slowing down if the code could theoretically execute, but that's a requirement to use the argument you did in your previous post.
Of course, the problems we're complaining about aren't that they replaced interfaces for speed. This is not people unhappy because V2 was nicer to write. It's because, when they did what ordinarily would be a normal improvement to interfaces, they conveniently left out things they didn't want to be in there and broke things that required them. That was not about performance, because they could easily have made a performant V3 with the power of V2. They chose not to.
And how confident are you that removing Manifest V2 will make any difference to the startup time of Google Docs? If there are no ads being removed, do the slower (claim not verified) operations that don't need to run to remove things still affecting it? Are there other operations that occur nonetheless that still affect it? Or is this just speculation that maybe removing code will speed things up, which is only true if that code was noticeably slow and running at the time?
They didn't say why they were going to collect that data. With citizenship restrictions, ID verification isn't sufficient, because an immigrant who hasn't been naturalized yet can still have local ID, more problems if dual citizens are also not allowed (I wouldn't put that past the US government right now).
If they're trying to prevent Chinese companies from copying Opus models by repeated prompting, and they think ID will do something about it (I doubt it), then they could easily do that to Opus too. Their statement does not suggest when or why they're going to do this, just that they might, so I wouldn't be so confident that it only applies to US users of Mythos.
They have a similar situation to what you seem to have. You've mostly seen negative comments about AI, so you assume that's what's out there for them to see. There is plenty of that, but another thing there is plenty of is people complaining that Apple missed the boat on AI, promised a development they didn't deliver, and really need to fix that in order to catch up. Those people aren't negative on AI, they want more and they specifically want it from Apple and will criticize Apple for not making it. Many of those people are the types who work in tech media for non-tech people. Perhaps, like me, you find that kind of thing annoying because they often swallow any marketing promise from a tech firm without understanding it so avoid it, but it doesn't change that some of them have much more readership than El Reg does. If Apple sees that a few million people hear that Apple must leap forward on AI or be left behind and a few tens of thousands see complaints about all AI, they have two reasons to follow the former. The quantity one is obvious, but the other one is that, if someone hates all AI, and everything has AI, then expect no change in how many of them still choose an Apple product before turning the AI off.
