Posts by Blazde

Patch to end i486 support hits Linux kernel merge queue

Re: I miss numbers

It was always marketing words pulled out of a bag, but it didn't matter because as long as it was at least 3 months since you last bought a computer, you knew even the cheapest new one was better. I miss that.

Re: Milestone!

Going down this same nostalgia path myself, I can't help wondering if Linux now works better with the mid-90s Mitsubishi parallel port external CD drive I struggled so much to get 1.0 to install from? Probably. (As much as I remember the autodetect was way off but manually selecting the correct driver after deciphering some bizarre model number formatting, putting it on the bootstrap floppy, and then accepting 1x speed reads instead of the desired 4x speed eventually worked. Possibly resolving some IRQ conflicts too. I do not miss those).

Re: they are falling short of target

("more minor but still very significant ones between 486 and 586" obviously. Granted the oddball clones do complicate this, maybe there's a case for dropping Cyrix support but I can't see Pentium or K5 support being a significant burden)

Re: they are falling short of target

the same applies to i586

It doesn't though? The kernel list discussion mentions that the overwhelming part of dropped code will be FPU emulation to support the 486SX. There were absolutely massive architecture changes between 386 and 486, and some more minor but still very significant ones between 386 and 486. But by '586' the instruction interface had matured and almost all changes an OS needs to care about are flagged as extensions with, in theory, no guarantee that future chips will support them. Do you want to drop support for single cores? No. Drop MMX support for some truly negligible performance gain? Drop 32-bit support? Hell no. Or drop support for CPUs without a TPM like Microsoft (kinda) did...?

Cryptographers place $5,000 bet whether quantum will matter

Re: Quantum computer == nuclear fusion

Exactly. Whereas a quantum computer will be useful even if it only runs for a few minutes because it doesn't need to achieve some net benefit vs quantum computing inputness.

And because the bar is lower for quantum computers to be useful they attract a lot of private funding so the research easily snowballs (has been doing for 20 years). IBM are already selling quantum computers. (And D-Wave, at the risk of opening that can of worms). We can question their practical usefulness at this point but the fact there's a market means something. While fusion power needs what is effectively blue sky research funding from supra-nationals, which is in very limited supply. Plus it's just inherently more expensive to research.

It's a superficially appealing comparison but they aren't similar situations upon scrutiny.

Re: Quantum computer == nuclear fusion

You've been living under a rock for the past 20 years if you believe this.

The issue with fusion power(*) is it needs to be a commercially viable means of producing energy in a world where we're already very adept at producing energy.

Quantum computing just needs to basically work a bit better than now and it'll already be useful for all kinds of quantum calculations we have no alternative means of doing.

(*) Nuclear fusion itself is of course already very useful if you're into destroying things in obscenely over-the-top ways.

Microsoft locks out VeraCrypt and WireGuard devs, blames verification process

Re: Incompetence, malice

The way AI is driving everything now I wouldn't be surprised to see a 'redefinition' of the Microsoft employee metric followed by a healthy increase in numbers, a conspicuous decrease in total staff costs, and a new risk disclosure that energy costs might significantly impact those costs.

Re: Incompetence, malice

Not even really stupidity, just cost-cutting. This is why I stopped using anything Google: Their enduring belief that customer-service can be replaced with algorithms, which is backed up by their total inability to hear customers wailing because they have no customer service listening to any feedback. Microsoft were at least smart enough to somehow get humans to volunteer to do some of their customer service for free, and very badly, for many years, but I fear those times are coming to an end.

AI agents found vulns in this popular Linux and Unix print server

Re: Oh no!

We've also been denied the gold that could have been "2 flaws 1 CUPS"

Ex-Microsoft engineer believes Azure problems stem from talent exodus

Re: Boeing Disease?

It's a systemic problem. The root cause is a failure to value the essentialness of competition in market-based capitalist economies. The US is worst at this for a number of reasons: General corporatocratic setup with no financial limits on political lobbying and an entrenched two-party semi-democracy. Half-hearted antitrust legal history that tries to judge the effect of corporate corruption on consumers instead of just answering the simpler question of whether companies are acting in an inherently anticompetitive way. The fact that this has lead to the US being home to numerous companies whose business model relies on a moat backed up by a network effect, which is then defended at all costs by a nationalistic political body unwilling to cede that network effect to foreign companies and unwilling to see regulation as a means of mitigating the power of the network effect, both because of cultural inertia and the aforementioned unlimited lobbying. And then there's the consumer who generally thinks that because a company is big and profitable and sells products with competitive pricing then those products must be okay, which is a reasonable conclusion when we're all schooled on the logic of idealised free market capitalism driven by competition. Azure must be good, millions of customers can't be wrong? Boeing seem fine, their planes don't properly fall out of the sky that often after all, they must be doing their best or else they'd go bust, wouldn't they? (And besides the only alternative is a huge European company forced to operate to similar dynamics so who knows when their planes will start falling out of the sky, it could literally be tomorrow, let's just buy American and hope Boeing have turned a corner). And then when you're a company whose customers think and act like this, your profit-making motive forces you to constantly fire decent staff, cut costs, and see if you can take the piss that bit more than you were already doing.

Somebody will wake up eventually and realise how messed up this model is but I think we need to hit rock bottom first. There are good signs we're getting there.

Researchers didn’t want to glamorize cybercrims. So they roasted them

Even 'Scrawny Nuisance' could be made work. A name is a name, it's what you do with it that counts.

https://tvtropes.org/pmwiki/pmwiki.php/Main/AppropriatedAppellation

"When you bring me out, can you introduce me as Joker?"

Trump wants to take a battle axe to CISA again and slash $707M from budget

Re: Don't understand the criticism!

Absolutely. 707mil is enough for almost one fortieth of an Iran war (so far). Find another 39 departments doing bookish stuff no one cares about like this CISA nonsense and we can have more weeks of pretty televised explosions!

'People's Panel' to check if UK wants controversial Digital ID will cost £630K

Re: hooray

It's obviously not about cost. It's about creating a 'hostile museum environment', and it's primary appeal to Labour parliamentarians is that Nige didn't think of it first

Iran war drives urgent need to counter underwater attack drones

Re: Here's my revolutionary, never-been-tried-before solution:

People naively imagining they can predict how a civil war will go down are one of the main causes of civil wars. Even if it somehow started as armed Republicans vs unarmed Democrats it'd be something entirely more chaotic by about day 3.

Re: Might I propose?

Not a bad idea but they may counter by equipping their UUVs with scissors, and then the trawlers will need to wield rocks..

UK wants to know if banning under-16s from social media does anything useful

Re: UK wants to know if banning under-16s…

Just last month The Archers did a whodunit storyline that involved not a murder, but a bloke getting bopped over the head sufficiently that he couldn't remember who'd hit him, thus creating the standard unsolved-murder-in-a-tightknit-community dynamic without the usual nasty consequence for the victim. Judging by the trails that got as far as Radio 2 if I remember correctly, it was properly dramatic with most of the village being suspects at one point or other. There is a certain creative genius that you have to admire involved in writing a drama with a dogged rejection of any hint of melodrama. (Just not enough to prevent it still being boring sadly).

Re: Banning stuff

a chronological list of posts is an algorithmic feed. Any sort of sorting or filtering process is - by definition - algorithmic

The problem arises when sorting or filtering is based on the content, because it then affects the way people create the content in the first place, and leads directly to all the dishonest, attention-seeking, rage-promoting, propagandising, SEO-type keyword spam bot-liked, trash that these sites are full of and which old social media technology like forums and chatrooms still aren't to this day.

Only Trump can decide when cyberwar turns into real war

There can definitely be guidelines though and there's no reason they can't be a simple extension of regular rules and laws of war (except that the US has joined a growing list of countries discarding those).

The intent of the adversary matters (and this easily counters them abusing your criteria). And proportionality of the response matters. If a hostile state cyber-breaches a dam and murders 3000 civilians, then a kinetic response is not actually an escalation. In fact a kinetic response against military or government targets is preferable to responding in kind with a civilian cyber attack against their dam. The cyber/kinetic division is not necessarily relevant at all.

If instead it's an intelligence-collection op gone bad, it costs you billions to clean up and maybe even some tragic personal consequences due to systems being offline for months, then you probably need to take that on the chin and make you critical systems more resilient for future. And make sure you're all up in their stuff collecting intelligence too.

The problem the US (and wider developed world) has is when the intent is just to cause high monetary-value damage/gain alone, then there may be no good proportionate responses. N.Korea probably the best example because you can't even sanction them.

(The other problems of course are figuring out with high-confidence who the adversary even is, never mind their intent, and convincing sceptical outsiders who can't see the detail of your attribution intelligence that you've actually figured it out with high-confidence. The Iraq war looms large..)

RSA panelists cry out for help to missing feds

Re: Genuine question.

It's easy to get through to my GP surgery with just a little hacker-mindset: Go on Wayback Machine, find the number they used to use 20 years ago before the horrible 'Press star to hear those options again' screening bot, dial that, let it ring for a good 3 or 4 minutes and eventually a confused woman will answer with "Hello?".

Then your social engineering 'motivation' is also to casually act like it's still 2006 and set up a polite expectation they're reluctant to disappoint too much.. "Oh hi, I'd like to book an appointment with Dr Smith. Later this week if possible, does he have anything available Thursday or Friday afternoon?.."

Re: Genuine question.

Yea it is hard. Human beings by dint of their complexity, their inherent biases, and their self-interest that doesn't necessarily align well with their boss's or their corporate interests, are endlessly hackable. They all need individualised training and rigorous testing to rectify their vulnerabilities, and even once you've done that they will readily re-acquire old or new vulnerabilities sporadically.

CERN eggheads burn AI into silicon to stem data deluge

Re: Science needs better funding and higher policy urgency

The schedule is 4 years of downtime. Hopefully LHC boffin math is better than AC's.

Meanwhile the current dataset collection will still be producing important papers in 2030 because the analysis is arguably the physics bottleneck. Think of it as the LHC tick-tock model.

Annihilate-investigate

Telling an AI model that it’s an expert programmer makes it a worse programmer

"You are an economist, please answer this question..."

Re: Dunning–Kruger is an offline problem, too

I think it's relevant because an alternative/additional explanation (aside from extra requirements reducing the model's capacity for sticking to facts) for LLMs performing this way is that, among the training data, that which is explicitly marked as 'from an expert' is quite likely to be from self-styled experts exhibiting Dunning–Kruger while true experts will not feel the need to be explicit about their status instead letting reputation, content & potentially a countersignaling signal speak, and all that is very complex for LLM training to unpack.

Indeed it's a complex task for non-expert humans to recognise experts too, which is why being over-confident / faking until you make it / etc, works quite well for gaining people's trust.

Forget drones – the US Army just took delivery of a self-flying Black Hawk helicopter

Re: Horrendiously expensive boondoggle built for another age

I'd love to have been in the meeting where they somehow decided a tablet was the most appropriate controller for this thing.

"We want it to be so autonomous a 5 year old with greasy fingers could fly it!"

Edit: It appears to be a Getac K120

The drone swarm is coming, and NATO air defenses are too expensive to cope

Re: 'Be scared' sells newspapers

Both RAF & HAF shot down at least 4 other drones with jets, so that was the wide-area drone defence. It's not as if it was a totally open-goal.

Then the one that hit went into the literal side of a huge barn that's designed to protect it's contents from small explodey things, and by all accounts did it's job. For all we know they let it hit rather than waste an expensive interceptor (I don't think that's likely but possible especially since it was supposedly an American hangar) or it was in an area of the base with no particularly fragile things so wasn't covered by better close-range air defences.

In the grand scheme of the Gulf drama right now I think it's been played up a bit too much.

Turns out your coffee addiction may be doing your brain a favor

Re: Encouraging, but...

There is still that off chance that early dementia leads to a life style where people drink less coffee.

Yup, first concern. Ironically the longevity of the study counts against it here since it tracked participants well into later life when full-blown and sub-dementia levels of cognitive decline set in. If you're struggling with your day a bit then cutting out luxury tasks like making drinks that require a bit of time and effort is just what happens. At the extreme your daily routine has to simplify hugely. Where you might have made a Sunday roast you're instead struggling to pour a tin of beans over un-toasted bread, never mind delicately-wrestling a French press. Also sleep problems are associated with dementia onset so the afflicted may well choose to cut out caffeine early on.

It would be interesting to see caffeine intake over the first 10 years vs dementia risk after 40, but it's not clear from the non-paywalled abstract whether they've done that.

Second concern: We already know engaging in cognitively-stimulating activities reduces dementia risk and caffeine is a popular tool already associated with those activities. So, if drinking coffee helps you achieve the marginal extra concentration needed to complete a crossword without getting frustrated and giving up, then that might mean coffee leads to lower dementia sometimes, but it doesn't mean you can drink your way to lower dementia while watching tele-shopping.

But absolutely there are plausible explanations for biological effects. I find tea (but not coffee) helps my inflammatory condition to a surprising degree. The whole anti-oxidant angle has gone through a rise, and fall, and re-rise story-arc over the years, enduring to the point it can't be easily dismissed any more. It's a bit frustrating tea and coffee get lumped in with each other in these studies because they have hundreds of differing interesting compounds in them and yet caffeine is always the headline-stealing one. For some reason they did not study caffeinated-fizzy-drinks-in-a-can intake, even though that's the primary source of caffeine in the US. I note that the article says 2-3 cups of coffee but only 1-2 cups of tea were associated with the most pronounced outcome difference, if that's accurate it would indicate caffeine is not the (sole) cause since coffee generally has higher levels caffeine.

UK police force presses pause on live facial recognition after study finds racial bias

Re: May not be a racial bias at all

You're trying to somehow blame the environment for having the wrong brightness levels, yet it's the hardware(*) that determines how light signals are turned into data. Hence, if the system is tuned to generate more accurate data for white faces then the system is racially biased.

(*) If you're not convinced then think about how low-light cameras work, and then think about cameras which image the sun's surface. You may be correct that the human eye would exhibit bias but we have an opportunity to correct that when implementing the detection entirely artificially.

Starmer's digital ID reboot raises same old questions as its Blair-era ancestor

"a timestamped map of every pub a young person visited on a Friday night"

Stop it, you're making ID cards sound fun.

Challenge accepted!!

Re: Seriously why?

Not disagreeing but there are some extra checks in the opposite direction:

- The courts inevitably force Parliament to be explicit when trampling on civil liberties.

- Backbenchers & sometimes peers are not nearly so brainwashed by the LE agencies and quite frequently do kick up a fuss. We should applaud them more for this, but yea most people don't care enough to acknowledge it.

And it's attritional. MPs with great effort defeat one proposal by organising a government rebellion, siding with minor parties, etc only to find the same proposal reworded in a bill the following session. Or they succeed in watering it down so it only tramples on civil liberties quite a bit instead of a lot, but then the incrementalism still kicks in. The ratchet never goes backwards.

Iran's cyberattack against med tech firm is 'just the beginning'

"[Iran aggressive for] a while .. there's no secret weapon that they've been holding back"

Glad someone's saying this directly to counter the doomsaying that Iran is suddenly going to unleash a new cyber-hell, as if they'd been holding back in the grey-zone for some reason. (There was a similar line about Russia in 2022).

The counter-point is that in the medium-term, especially once their oil is flowing again, they'll likely spend a lot more on ramping up dispersed cyber capabilities (and drones of course) and a lot less on costly missiles that have proved undefendable. They do have a good starting point to do this relatively quickly.

Intel finds its Zen undercutting AMD with Arrow Lake refresh

Re: Cherry Picked Data

Power is the supposed rationale, yet we see Intel chips still as or more power-hungry than AMD's. I do agree power-efficiency is a very important aim (notably at the very low end and the high end, less so in between) but it's not at all clear why E-cores are a solution for that except marginally in quite well defined and serially-power sensitive devices like mobiles, when you really only have a few foreground threads and everything else is always just silicon given over to big tech telemetry and not really for the end-user.

Chip area is not a good guide to power usage. Clock frequency is the main determinant. Ideally to chase low-power you want the most capable cores with as much out-of-order compute as feasible to maximise instruction throughput, then you under-clock & under-volt them. That's server architecture where cooling $/hour is your primary metric. I don't mean to pretend it's trivial to achieve, and there are some trade-offs, but that was the entire direction of x86/x64 architecture for a long time and was mostly lead by Intel. Only when AMD seized an opportunity to ramp up core count did Intel flounder and see E-cores as a route to half-way match those core-counts within die area they'd constrained themselves to. If a Zen CPU wants an E-core it just throttles down a P-core, achieving ball-park 65% through-put with 30% of the power, yet it maintains the ability to be a P-core when needed and you can still anyway make efficient use of die-area by adding hyper-threading. So again, what is the actual advantage of E-cores?

Re: Cherry Picked Data

I see about 30% throughput increase from hyper-threading on Zen 2 with embarrassingly parallelisable workloads. But these modern high-core psuedo-NUMA chiplet architectures really suffer as you split up a workload that needs even small amounts of synchronisation between cores so it's not surprising if hyper-threading has become less useful for most, especially gaming-type, benchmarks. The contention goes up exponentially while you only get a moderate increase in compute. That's more often going to be a good deal if you're going from 2 physical to 4 logical cores, than going from 16 to 32.

The irony is Intel's E-cores achieve pretty much the same, while also adding massive headaches for OS scheduling and compiler optimisation, and for Intel's design and manufacture, due to their heterogenous nature. I can't really understand the logic of them pursuing that strategy. It's even worse when they have hyper-threading on P-cores too because the scheduler has to weigh up whether to put a workload on an already singularly-occupied P-core or on an empty E-core, where the former might be better in some cases but it's very difficult to identify those cases in general.

Anyway, step away from single-workload benchmarks we always see even in 3rd party comparisons, and into multi-tasking (where separate tasks typically have almost no synchronisation needs) and I think hyper-threading is still a no-brainer.

Swiss e-voting pilot can't count 2,048 ballots after USB keys fail to decrypt them

They do recounts if it's close, there's a strict process for that, so the integrity of the result is never sacrificed. If you watch the level of organisation in the quicker constituencies it's tough to conclude accuracy suffers at all compared to the lazier counts.

The catch is you need the right geography to even be in with a shout, hence it's only a few places that compete. Orkney and Shetland by contrast take pride in taking their time to ensure no ballot boxes fall off the boats in the dark..

One of the highlights of British general elections is that 2 or 3 constituencies always race to declare first. Not just volunteering but turning it into a sport. Human bucket brigades passing ballot boxes into the count building at 10:04pm, serious looking people with clipboards literally running out, barking orders and pointing fiercely. And hundreds upon hundreds of tables all pre-prepared for the hard work.

Another highlight is in all the other constituencies where you get to see the losing party candidates at dawn, after 24+ hours with no sleep, drained of any sense of motivation to stay awake by the scale of their defeat yet trying to maintain dignity just long enough to make it through the vote announcement and looking visibly pained as the opposition cheers their tally. Quicker counting would rob us of this.

Quicksort inventor Tony Hoare reaches the base case at 92

I had a sort-of premonition/pondering just last week that we must be getting to the time when a lot of these giants from the first generation of wide-spread computer science are going to be leaving us. Some have already gone of course, but we're facing the reality of a really important part of living-memory becoming pure history and that's a sad moment for the 'youngest science'. When I studied in the late 90s Hoare was in his 60s, and like all the textbook-name language & algorithm pioneers, still going strong. Aside from a very few early theoretical proto-computer scientists (Turing etc) we took it for granted that the field was composed of living people you could often just email there and then, if you had the guts.

I use Quicksort daily (and I genuinely smile when I get to use Quickselect because it's one of those algorithms that feels a bit magic), but I'll most remember him for his Billion Dollar Mistake talk because he eloquently and humbly leveraged his gravitas to shine a light on a very contemporaneous problem in a way that was accessible and thought-provoking. That people are still debating the value of that 17 years on is testament to how confident he was in the relevance of his decades of wisdom, even in his later years. Hope for us all. Don't get stuck in your ways, stay young, question your choices as the evidence rolls in.

RIP Tony.

FBI is investigating breach that may have hit its wiretapping tools

Re: International law

The sanctions on Iran & Venezuela are not even classed as 'international law'. They're almost entirely Western sanctions with one or two exceptions the UN agreed on with narrow focus on Iran's nuclear programme.

Firefox taps Anthropic AI bug hunter, but rancid RAM still flipping bits

Re: alternative truth

The processor should raise an exception if the upper 16 bits don't replicate bit 47 but I am not clear what the particular exception would be

It's just a general protection fault, but like you say it's hard to discern whether the root cause is a bit-flip or some bone-headed pointer arithmetic. It isn't clear from the article but the Firefox data quoted involves running a brief memory tester against the suspect addresses which I think should be more definitive than relying on triaging >48 bit vs <=48 bit changes, as potentially interesting as that is. It should also rule out cosmic rays - whatever their actual contribution - instead focusing on repeatably bad memory addresses.

Re: alternative truth

The current RAM shortage isn't an excuse, that's just standard supply/demand mismatch in a cyclical market. If ECC memory had been pushed by a handful of key players a few years back when it became sensible then, sure, we'd maaaybe all be paying anything up to 1/8th more for our RAM because of the extra chip demand but that demand would have been predictable and thus the shortage would not be proportionality worse than it is now.

Every other bulk movement of data has integrity checking. Networking, mass storage. Often at multiple layers. RAM is an outlier and we should all be outraged by it because there is no excuse other than than bone-headed fixation on more popular/better marketed metrics.

Re: alternative truth

In my experience it's very easy to see a crash (in code written in a non-memory safe language) and be so confused by how it happened it's tempting to blame a bit-flip until you eventually untangle the bizarre race condition that lead to it.

But, equally, browsers use vastly more memory than they did 10 years ago, and memory integrity during that time has not increased in line with the usage (I'm not sure it's increased at all) so we should expect more flips. The complexity of browsers with their sandboxing and layers of scripting interpretation and JITing also means an awful lot of memory usage consists of 64-bit pointers which have a very high likelihood of leading directly to a crash with a single bit flip (this was not the case with 32-bit addressing which might easily cause data corruption without a crash). So it is plausible.

Re: I'm not surprised about hardware failures

I'm sure I don't have a very representative experience, however, while I've seen a few sketchy ram sticks in my time, the way greater share of memory errors are 'end user attributable'. Either over-clocking/tweaking too much or seating DIMMs badly. (I mean: what do you do while you're testing stability of your new overclock? Maybe you fire up Cinebench or Prime95 but chances are you open a browser and get on with life too).

Still, I'd love to see a breakdown of 'probable memory errors' by RAM manufacturer, memory controller, etc

Chardet dispute shows how AI will kill software licensing, argues Bruce Perens

Re: Prompts?

APIs being copyrightable is not settled even in the US. It's fundamentally at odds with interoperability, which many courts in many jurisdictions have found to be an important goal in software and beyond.

For me, the concept of 'copyright' is butting up against the concept of 'patent'.

If I'm machining steel brackets and girders and I invent an oddball screw/thread spec (it's thread pitch is ya' know 1/16 of an inch plus 0.1mm, left-handed, and it has a troll-face as a driver head) and let's imagine the legal position is that nobody else is able to copy that spec and produce either bolts or holes the same, so that in practice my company's brackets can only be bolted into my company's girders with the same holes, and if you buy even one bolt from me then you're committing to an entire 'glorious' monopolistic ecosystem of steel that my descendents will happily milk for 70 years after my death.

Well, that's obviously stupid and people figured this out a long time ago. Meanwhile copyright was intended to deal with books, and music and other art-works which don't interoperate at all. (Arguably 'fair use' is the interoperability parallel but it's a bit of an after-thought people still argue in depth about.)

Software APIs are more complex than screws and good APIs *can* exhibit a sort of artistic beauty if they're well designed and so it's tempting to put them into the the copyright category, but in practice most of them are pretty shitty and either obvious or ultimately just one way of many, many thousands that 'this thing' could work and we don't really care which way it does, just as long as both sides of the system agree on the way, which means they're almost always non-innovative and non-patentable. They're nothing more than an uninventive thread with more parameters.

Cloning software libraries to some existing API spec has been bread-and-butter software development for a long time now, and I don't see any reason for that to change despite whatever the balance of 9 learned people in the US think on one afternoon. Obviously the API is in some respect a part of the original work but you can't avoid it influencing your implementation if the goal is to interoperate with other code that adheres to that same API. After all code which *uses* the API from the other direction is not itself copyrighted by the author of the API, duh! So the clean-room concept is plainly bogus. We live in one big dirty room, get over it.

The focus better at least be on the *internals* of the implementation (unless possibly the API is somehow truly ground-breaking - I can't think of any example). So, potentially even running unit tests from the original implementation to *check* an LLM-written clone is doing what you expect is going too far if those unit tests are doing anything more than checking the external-facing API. The issue with open-source is that the 'external-facing API' and 'internal implementation' are often not well delineated, but some parts will inarguably be internal and if a clone is replicating the *structure and performance* of those then possibly we should say it's copying the original implementation even if a simplistic token-by-token analysis shows dissimilarity.

Long-term I think LLMs are challenging whether even that structure is innovative or whether only patentable algorithms should really be subject to 'software copyright'. It depends a lot on how much faith you put in LLMs to write good, original code and to continue to, without self-enshittification: If the actual value of good human-writtten code is high going forward then that should be well-protected by copyright. If instead the LLMs can do in microseconds what a human took a week to do last year without seeing the human code then it doesn't much matter whether it did actually see the human code or not. If you study most 'implementations' of an algorithm they'll stick close to an original pseudo-code version perhaps published in a journal paper (patented or not) and if an LLM could knock them out from the original paper who cares if it's also seen a Python implementation, or a dozen of them? However many implementations do themselves have truly novel features which aren't documented, patented, or published other than via the source and the LLMs are seeing those too. Should the authors be forced to patent if they want to protect those, or should the copyright be automatic, and where do debates about novelty fit in? I have my own firm hunch where we're headed wrt to LLM-capability, and reliance on and legality of ingesting training data, but I also think anyone who says they're completely sure where we're headed is just being wishful. And I definitely don't trust US Supreme Court judges to know where we're headed.

UK watchdog eyes Meta's smart glasses after workers say they 'see everything'

Re: Dorothy Parker Quote

To be fair it takes some guts to make a pass at a librarian ("Ssshush"), and more so your teacher...

Re: describing the allegations as "concerning."

Cameras, microphones, and - through the 'Neural Wrist Band' - they "read micro-gestures from your hand". The mind boggles.

But they're being marketed to people who feel they're at a "significant cognitive disadvantage" without them so maybe it's not bleedin obvious to the wearers.

Iran all but vanishes from the global internet amid US-Israel strikes

Re: Another war for oil

By my understanding they're pretty concerned about any conflict in N.Korea, among other things because of the likely severe refugee crisis it would create in north east China. They have some levers to keep the Kims in check but it clearly wasn't enough to stop them getting the bomb.

(Anyway, yes. It's way more complicated than oil. The current attack taking place has much more to do with Israeli rather than US foreign policy, and we can't really say oil is the primary factor there)

Re: Just Weird

It's not perverse, it's part of the point of it. Even when the enemy know you're doing it, it's still divisive, because they turn on each other accusing each other of being bots or being brain-washed by bots and so on, and therefore dehumanise and distrust each other while reinforcing their own position.

Re: Another war for oil

Comparing the NHS to the IDF? One of the purest examples of Poe's law, or it's inverse. I'm truly lost.

Re: Another war for oil

the way IS did in Iraq. (If if weren't for the religious and cultural differences otherwise making this less likely in Iran, I'd have assumed it most likely that IS themselves would be the ones doing so to make an unwanted comeback).

For me, the more realistic outcome the US could heavily regret is that China now sees even more opportunity in Iran. They're desperate, well within China's expanding sphere of influence, big oil suppliers to China, no longer a nuclear state risk, and capable of relative stability probably more so than current most-valued buddies, Pakistan, and definitely than Afghanistan which Trump himself has already fretted about effectively losing to China's influence without actually mentioning that's his own fault.

The first signs of this are that while virtually all Western shipping is snarled up in the Strait of Hormuz, China-bound tankers appear to be seeing much less risk navigating it.

Re: Another war for oil

I've always thought the 'for oil' narrative was pretty lazy, and still do, but Bush's 2002 'Axis Of Evil' scorecard is looking a bit sus. at this point for those of us trying to argue it's more complicated than that:

Iraq WMD: Turns out, no. Oil output: Massive. US regime-change attempt: Yes

Iran WMD: They wish. Oil output: Massive. US regime-change attempt: Yes

N.Korea WMD: Nukes! Oil output: Zero. US regime-change attempt: No

(Libya: No. Yes. Yes. Venezula: No. Yes. Yes. Canada: No, Yes. Maybe...)

Re: "the good sense to leave the US & Israel to slog it out"

There's no way any war this close to us doesn't end up using our bases at some stage, and Iran have mindlessly lashed out at everything in reach, including us and our allies (Israel and the USA). There's nothing else he could do but allow our bases to be used.

I heard that a British-made military radio was slightly damaged after being dropped to the ground by an IDF soldier in Tel Aviv, during an air-raid warning triggered by an incoming Iranian ballistic missile, so I'm 100% with Keir. The gloves must come off!

(Err, the defensive gloves that is)

South Korea’s tax office apologizes for leaking seed phrase to seized crypto

'so perhaps this will still end in a win'

Fractional-reserve cash seizure. Clever. Good for hitting policing targets, but potentially inflationary.