Posts by Justthefacts

Britain's creaking courts to use Copilot for transcriptions

Re: End of justice

I mean, that is so obviously untrue. Do you actually understand our courts system? I think you are mistaking the fact that judges have a higher conviction rate than juries (which is true), but what you said is just rubbish.

Consider the circumstances under which you have any choice about whether to have a jury trial (in Crown Court) or not (Magistrates Court), ie either-way offences. For most crimes, there is no choice - e.g. for robbery it will be a jury trial, you have no choice. And then remember that Magistrates can only impose sentences at the lighter end. If you are guilty of an either-way offence, then choosing to go to Crown Court for a potentially much higher sentence would be like turkeys voting for Christmas.

Plus, opting for Crown Court over magistrate, virtually guarantees a really long wait on remand. Which for some means waiting in prison for 18 months, but at best leaves it hanging over you with probation restrictions that your employer is likely to find out about and get you fired. You would have to be absolutely freaking nuts to volunteer for that, unless you or your lawyer thought there was a really decent chance of being found innocent. 10% chance vs 5% is no way going to cut it. And when your lawyer knows the statistics that 80% of effective Crown Court trials return a guilty verdict ( for the record, I sat on four trials total, and all four were quick Guilty verdicts), again you would have to be freaking nuts to opt for Crown Court unless you had a better than average case, and a good chance of being found innocent.

Re: End of justice

I’ve explained why I think trial by jury is a good thing: because otherwise you have a underclass that is “done to” by Power. Of course you would think that benevolent Power is a good thing - you are in the 90% that assume it will never be in the dock. Power always thinks it is not just Right, but Just. The purpose of juries is to make everybody a part of the system of justice, and give everybody a stake in the society we live in.

I think you’d also be surprised by the actual research on jury/judge error rates. While separate judges do disagree with each other *somewhat* less than judge vs jury (seeing the evidence on the same case) the difference is not overwhelming, it’s roughly half. Judge verdicts are not as “expert” as you think they are, the data says that they must be in error at least half as often as juries.

Most of that difference is accounted for by one single fact: juries get sent out of the room for evidence that they are not allowed to take into account (such as previous convictions). But judges hear even the prejudicial evidence and are expected to mentally discount it when they report “the verdict they would have given”. In every trial I have jurored, we have been sent out the room more than once - ie evidence has been withheld from us for legal reasons, rightly or wrongly.

If you really wanted to bring jury verdicts into closer concordance with judge verdicts, we need do no more than allow juries to hear all the evidence. Alternatively, the rules of evidence thrashed out over centuries may be *right and justified*, and judges may be much less good at ignoring prejudicial facts, than they think they are.

There’s a reason I bothered to look up all this stuff. When I served on a jury, I wanted to get a feel for what “beyond all reasonable doubt” meant in practice, I took the job seriously, and both juries I have been on, all the others have too. Juries decide who to believe, and what factual conclusions you can really draw from the evidence. They do not need to know the law, that is for the judge to give guidance on.

Re: End of justice

The purpose of a jury trial is to not to give the most accurate decisions….which quite likely it doesn’t. Just as the purpose of democracy is not to mathematically optimise decision-making in the best interests of all its citizens, which it also doesn’t. The purpose of both, is for those who are being “done to”, to have a say in what is being done to them, particularly those on the bottom end. Without that, we do not have a society, we just have power structures papered over with fake concerned sincerity and process. That is why we should not give up on jury trials, as a temporary fix for a crisis in our criminal justice system.

We do have a crisis, the courts are in an appalling state. The number of abandoned trials and inefficiency is just astonishing. But those in charge should be held accountable for their lack of organisation. The cause is not the resources and availability of twelve people (actually 16, for those who have served on juries) who are paid at best minimum wage for their time. I’m far from knowing the true reasons for what is going on, but from my observations as a juror recently, I would say that a lot of it is: simple inability to schedule (get the lawyers, accused + witnesses to the courtroom on time), + adherence to a set of arcane rules that are not fundamental. For example, the entire trial system seems to be built around the idea that both sides can take as long as they like on the day, and therefore you don’t know which days people are going to be required on. Why? This doesn’t work in any other area of life. Rather than abandon jury trial completely, why not just say that eg in a GBH trial each side has 3 hours to make their case. It’s up to them to prioritise within that. Maybe not perfectly ideal, but it’s a lot better than abandoning the bedrock of our justice system.

Nvidia leans on emulation to squeeze more HPC oomph from AI chips in race against AMD

Checking for NaN makes software devs feel all warm and fuzzy that they’ve “done the right thing”. But for every 1 test case where a NaN is correctly caught by your check, there are a dozen adjacent cases where completely spurious data is being output that you aren’t catching. You are missing the forest for the trees.

Let’s make this concrete: sqrt(X) gives NaN when X is negative. The fact that you are calculating sqrt(X), is a claim that X must (should?) be positive. So you could put in a check for X >=0.0, or for NaN after the sqrt(), and that’s exactly what most devs do. And indeed what they think industry best practice is. This is a horribly naive approach.

If the assumption is that X must always be positive, then what do you think if you were looking through logs and spotted a value of X = +10^-200 ? That value does technically have a square-root…... But if you see that, it is a smoking-gun that the formula you are implementing is not correctly regularised.That’s the bug you have to fix, because it affects *all of the output data, all of the time*. Not the 1 in a million NaN that is corrupting one data-point.

See above. FP128 is absolutely not the solution to your numerical issues. All you are doing is reducing the number of test-cases that *flag* proximity to a singularity.

The underlying problem is that you need more stable numerical analysis techniques (MMSE and others) which will avoid the singularities in the first place. With better techniques, you don’t need 128 or even FP64.

However, without such techniques, even FP128 isn’t actually good enough. Neither is FP256. Or even FP512. You are just making it harder for yourself to find and test edge-cases.

“NaN and infinity handling is designed so that once you get one of them as a result it continues to propagate through the math.”

With respect, this is a common misconception and a massive problem in numerical modelling codes that I see. In fact, the problem isn’t “dealing with infinity”, the problem is “dealing with near-singular numbers that aren’t infinity (or zero)”. So, if one of your intermediate results is 2.01x10^300, it is perfectly representable, but is it correct and meaningful? Answer: no it is not. What you have found is a data-point within a gnats nadger of a singularity. The issue is that had you been even 1 part in a billion either side on the input data, that 10^300 could have been 10^200 or 10^800. Your intermediate result is completely meaningless. You should not propagate it. There’s always going to be loads more data-points near a singularity than at the singularity itself. By addressing floating-point value infinities, you are patching the symptom, not the underlying cause.

There is no substitute for proper numerical analysis. Double-precision (or even quad-precision) is not a magic wand.

How you fix stuff like this, is a big topic. The issue, of course, is that people say stuff like “but it matches the Matlab result” without digging in, becaause they don’t want to know they have a problem. Matlab can be wrong too….GIGO

Basically, you need to improve your numerical analysis algorithm to avoid ill-conditioning. The underlying solution is recognising that the input data is not perfectly accurate, so you need an MSE estimator which recognises that and propagates it through the full chain of calculations, giving an MMSE estimator at each point. If you do this correctly, you will never have to invert a near-singular matrix.

But even more important, you will gain a lot more physical insight into what the data really mean.

I can count on the fingers on one hand, the number of codes I have seen do this correctly.

The world is one bad decision away from a silicon ice age

Re: Why would China upset the apple cart?

There’s only one reality which matters: how much oil&gas do the EU buy from Russia to prop them up financially? Answer: the EU is still Russia’s largest customer, in December 2025

https://energyandcleanair.org/december-2025-monthly-analysis-of-russian-fossil-fuel-exports-and-sanctions/

And - are Germany still supplying Russia with munitions? Answer: yes, Germany export more munitions to Russia today than they were in 2019. They now export via Kazakhstan etc. Listed as “machine parts” and “vehicles” naturally. Look at the graphs.

https://tradingeconomics.com/germany/exports/kazakhstan

https://tradingeconomics.com/germany/exports/kyrgyzstan

The EU do of course sell arms to Ukraine as well. The arms business is a great business to be in at a time like this, if you have no moral compass. Nothing changes.

Artificial brains could point the way to ultra-efficient supercomputers

Re: Recent introduction to SpiNNaker from Steve Furber

That is true, but there’s a tradeoff here that he’s not addressing.

To know whether or not the inputs have changed to each operation, you need to effectively cache all the intermediate results at every layer. The spiking architecture does exactly that, but at the cost of providing an SRAM register local to each arithmetic operation. This is very power and area hungry. With current hardware technologies, it’s often not efficient to memoise stuff, instead recalculate what you need when you need it (a software lesson that applies more generally than AI software).

Some Alternative architectures do cache intermediate results by storing out to main SDRAM. But then, it costs hundreds of times more energy to shuffle cached data into and out of external SDRAM than the actual computation. The price of at least doubling the amount of HBM memory that you need, might well be more than the execution units you are saving. So it really isn’t obvious a priori whether “wasting 98% computing stuff that hasn’t changed”, is less efficient.

I don’t believe there is any way to know without just building the best possible implementation of each architecture that you can, and finding out the fact of the matter. So it’s definitely great that people actually are doing this.

UK sinks to fifth in ESA funding league behind Spain

Re: Ok

The Tories just can’t make up their mind

On the one hand, they want her to resign for saying there is a black hole when the OBR says there isn’t…..

On the other hand, they want her to resign because they claim there’s a huge black hole which she is hiding.

https://www.telegraph.co.uk/business/2025/12/02/the-hidden-13tn-debt-bomb-in-the-budget/

Which is it? Is there a black hole or isn’t there?

Seems fairly obvious to me. The OBR has their model predictions, which are better than guessing, and honest, but still at a practical level its accuracy record is piss-poor. The Chancellor, being democratic oversight, has not just the power but the *responsibility* to add extra margin if she thinks that is wise, to accommodate the huge uncertainty of world and domestic events.

Now, whether the actual outcome will be higher than the model, neither you nor I have a clue. It’s a judgement call. Which is why we have democracy rather than an unelected bureaucracy of experts like some other countries. It doesnt make the OBR stupid, but if you think the models are remotely accurate you haven’t been paying attention. Predicting the future is hard.

For example, the March 2022 forecast for cumulative GDP growth over the next two years, under-predicted the out-turn by 5.5%. Read that again. Yes that’s correct, I got it from the OBR’s own (honest and professional) self-assessment report. They were nearly 3% growth pessimistic per year, *over a period that has included a war in Europe*.

https://obr.uk/docs/dlm_uploads/Forecast-evaluation-report-July-2025.pdf

Read the OBR’s own report. You will learn a lot.

Next time you read a report from “leading economists” that eg Brexit or any other world event will decrease or increase GDP by 0.5% per year, take a deep breath and re-read the OBR’s own data on how ridiculously uncertain economic forecasts really are.

Researchers get inside the mind of bots, find out what texts they trained on

Re: Why not try Finnegans Wake

Right, well I’m afraid that didn’t take long to figure out what’s going on, and I absolutely stand by my original statement, although with fractional nuance.

ChatGPT also knows the first line of Finnegans Wake, verbatim.

But if you ask it the second line, it claims that it can’t, because copyright issues. But it can *paraphrase*. And what it paraphrases with…..is wrong. Completely wrong, wrong subject and no words in common. The suggested second line does indeed reflect themes and plot of Finnegans Wake, but ChatGPT does not know what the second line of Finnegans Wake actually is, even in outline. Or the third line. Or the last line, it absolutely does not have a clue what the text is.

Famously, the last line is cyclical with the first line. ChatGPT knows this, can explain the link, can do a thematic analysis of the final monologue. But it does not actually know any of the words in it. ChatGPT knows famous facts about Wake, but it simply does not know the text, apart from the famous first line.

In short, ChatGPT knows the Wikipedia article on Finnegans Wake. It may also have memorised a statistical average of blogs and essays about Wake, that I haven’t yet figured out, and am continuing to play with it. Now, you may have an opinion that storing (a compressed form of) Wikipedia without attributions is itself copyright violation. I haven’t thought that through yet. But one thing I am absolutely certain of, is that ChatGPT has not memorised any significant portion at all of the actual text of Finnegans Wake. Even though it is public domain and one of the most famous books in the English language (which nobody has read). Far less has ChatGPT memorised the text any of the much larger corpus of English literature.

And I’ve even got more direct confirmation of that: if you ask it the most common word in Wake, it says “the”. It can even give you the top three most common words. But if you ask it for the tenth most common word *it does not know*. It is not counting words in the text. And it says “I can’t find a reliable source for that information”. It literally does a web search in front of your eyes, posts a publically available link to a word-frequency analysis essay, and then tries to parse the document in that link. This is really clear.

You called out a different LLM. I can’t and won’t do a systematic study to see which of the LLMs might have actually memorised more. Maybe you can find another one. But to be honest, this is like perpetual motion machines. There are really good fundamental information-theoretic reasons to believe that a 7B parameter model is not storing the compressed text of a 70T corpus as a small part of its storage. I’ve debunked one perpetual motion machine, I’m not going to debunk all of them separately.

This is a thing that just is not true. It cannot be true. And it is not.

Re: Does the chatbot know the entire text of a particular book

No, that’s a completely different issue. I suggest you actually read the article you linked.

The reason why what Anthropic did was ruled piracy, is not that they reproduced copyrighted books which they had legally bought. The court did not rule that reproducing from LLM-training is not fair use. The case you linked to is that Anthropic bulk-downloaded *pirated copies* in the first place. That’s just no different to any individual or organisation downloading and reading pirated copies, whether they produce anything from them or not. That is illegal, correct, I don’t think anybody argues that.

The real main issue is whether you can use the stuff that you have read *legally*, to train. So long as you do not output segments which are long enough to be recognisable, and constitute a creative output.

Having said that, somebody else is showing that some models actually do so (over and above the unconvincing children’s book example). So I’ll look at that.

Re: Why not try Finnegans Wake ..

Now that, indeed, is Very Interesting, thank you. I’m going to try a few things out.

I should have tried it myself, it’s a simple experiment and worth so much more than the researchers whole Harry Potter study.

The next question is “how is it doing this”, because the entire global corpus of literature cannot be encoded in a mere few gig of parameters. So ClippyAI is acting as a front end to Ollama, which is a fully local LLM, and zero internet access? ie this is not agentic LLM scraping external content dynamically?

Re: Does the chatbot know the entire text of a particular book

Right.

So you think that if the LLM ingests the first line of 1984 “ It was a bright cold day in April, and the clocks were striking thirteen”, and a dozen other dystopian novels.

And when prompted to write a dystopian novel itself, it “decides” = “highest correlation”, that it should begin by destabilising common assumptions, and repurposing words. It outputs the text “It was another sweltering Christmas Day, and the blackbirds overhead were calling the Faithful to Prayer”.

Are you claiming that this is a violation of intellectual property, because in some sense it reproduces the overall effect? Factually, the law disagrees, that is not copyright violation, there is no copyright on “meaning, patterns and relationships”. If you are wondering why fan-fiction “sequels” sometimes gets caught up, it’s because character-names are copyright-protected.

Feel absolutely free to write a Harry Potter style book, as long as you don’t use names like Harry Potter or Hogwarts in it. Feel free to write a James Bond action spy thriller, as long as the hero is not named James Bond. There are thousands of them.

More importantly from your point of view, I simply don’t think you have a leg to stand on either philosophically or ethically, sorry.

Forgive me if I have mis-characterised your position. But “I’ve based my book on the way you wrote yours” just is not copyright, nor should it be, and that doesn’t change whether it is an LLM, human fan-fiction or whatever.

Re: Why not try Finnegans Wake ..

Yes, that is pretty much what I would expect the output to be.

LLMs do not have the capacity to store text word-for-word, they do not store facts, they store answer-shaped text associations. Everybody knows this. In fact, that is exactly people’s main complaint; and yet they seem unable to generalise this knowledge when it comes to the subject of “storing copyrighted text”.

If you ask Gemini about Finnegans Wake, it will probably come up with “a stream of consciousness” that looks like what Joyce would have written, on a broad swathe of what he might have written them about, according to a plot-summary based on his Wikipedia entry. But not the words that he did in fact write. That’s a category error.

It might reproduce a couple of short “famous quotes” from the “weighted average” of a hundred essays on the subject. But again, not any individual essay, because it does not memorise individual essays.

I haven’t tried. But if in fact it does manage to produce large-scale quotations of F’ing Wake, that would be evidence that would change my mind. Change my mind!

Stating the bleedin’ obvious again, on a techie website: A 200GB model has *not* allocated a significant fraction of its storage space to storing the exact text of the 70TB corpus of written books.

Because the Shannon bound, and Kolmogorov complexity are things . Facepalm

Does it never occur to people that the reason why these researchers-with-a-point-to-prove select Harry Potter, is that the text is just super-predictable?

Why not try Finnegans Wake, *a book which genuinely did enter public domain a decade ago*. It’s a famous and iconic text. Go on, reproduce sections from *that*, and I’ll be interested.

Re: False Assumptions

Exactly correct.

Most people do not understand text copyright law, and just have vague Feelz based on what they heard in the media about how music copyright works.

Text copyright is an entirely different body of law to music and performance copyright. In fact, the word “copyright” is just reused but they are two entirely different concepts. Music and performance copyright is based on *similarity*. It’s a correct claim if you can prove the source and target are identical however you got there.

Text copyright is based on causative chain, length and distinctiveness, and creative step.

Leading the witness by the hand, asking again and for plausible continuations, and then locking in each sentence once they have guessed correctly with a gotcha, this is the opposite of what could possibly considered a text copyright breach.

Whereas, for example, a picture or music is exactly the opposite. If the AI photo *looks exactly like* an iconic image, or a musical phrase sounds exactly the same, that is a copyright breach. As long as it is big enough and recognisable enough, you don’t even have to *suggest* that the AI ingested the original, let alone prove it. It is copyright breach because the two items are recognisably the same, period.

It’s just a completely different ballgame.

And most “training” material from organisations, particularly universities, is just 100% wrong about this area. They don’t understand it, and they’ve promoted the misinformation as a folk memory. Like it or not, there’s two totally different sorts of copyright.

Project Kuiper becomes Amazon Leo as satellite network trickles into orbit

Re: Launch licence deadline?

That’s a great idea. Not only does a rudderless NASA not want the pointless Starliner, Boeing clearly don’t want to do it any more either. Win all round.

Re: Launch licence deadline?

Yeah, but “late” is when you have a plan. Kuiper don’t even appear to have a plan any more, as they have no practical launch options declared. If they are going to launch on Falcon9, they should say so, but that is not the plan of record.

Both ULA Vulcan and Blue Origin are still 2 years away from volume launch, at best. Ariane 6 has made it clear that they will prioritise EU governmental, and already canned one Kuiper launch. Since Ariane is 2 years behind, maybe Kuiper will get *one* Ariane launch per year for the foreseeable future (2026, 2027, 2028). Anything Kuiper put up today, will already have de-orbited by the time they have put up 5 or 6 launches let alone 38. I don’t fancy standing up in front of the FCC with that.

I think they are going to be told to wind their neck in, and present a more realistic de-scoped Phase 1, relying on no more than 5-6 launches. This might actually be the best thing for them.

Taiwan gets chippy about US request it shifts manufacturing

Re: Pretend I am a Martian.

Always put things in human terms, as if this were a project you were going to implement for yourown financial gain, it often becomes clearer.

So: “Being a doctor is a high-salary occupation. Money (for training) aside, what prevents me from opening a doctors surgery?”

Well: you absolutely can try, but….,,in the best case it will take you ten years, during which time you will have zero or low income, even in the best case. The failure rate during training is very high. The majority who try, don’t make it, they’re not smart enough or they don’t work enough hours until midnight, or both. Today, you have no evidence of skill or talent. You might be in the top 1%, nobody knows, but the cold hard facts are that you are 99% likely to be in the 99%. You’re in a moderately well-paid job today. Not amazing, but you’re going to have to give that all up to take a spin on the wheel to be a doctor. If it doesn’t turn out for you, coming back in 10myears time you will be 10 years out of the industry, and your current job will not be open for you. You’ll be serving customers at B&Q.

All of that applies pretty much word for word, as to “should we, or should we not, spend $500bn of our taxpayers money to have a 10% chance of staying in business against TSMC”.

SpaceX bulks up Starlink Direct to Cell with $17B EchoStar spectrum deal

Re: Too much money for too little spectrum

Considered directly as an investment for ROI, I agree. But it does have defensive aspects.

Eg Kuiper constellation *in theory* is married to an organisation with near infinitely-deep pockets. Amazon can afford to buy anything, which gives them a competitive advantage in scaleout. By taking a one-time opportunity to buy a beachfront property, just 2-3 years before their likely competitor could consider spending money on spectrum, SpaceX may knock another nail into Kuiper.

At worst, Kuiper or some other NTN hopeful is *successful*. Then, they become prepared to pay more for that spectrum than SpaceX paid. It might be a no-lose investment.

Or SpaceX might have just paid too much for it, difficult to tell.

Europe Putin the blame on Russia after GPS jamming disrupts president’s plane

#1 “Galileo offers meter-level accuracy for public users and centimeter-level accuracy for commercial services. Its dual-frequency signals reduce errors caused by atmospheric interference, making it more precise than GPS in many scenarios”

GPS is dual-frequency too, but that’s not the point. The point is that more modern receivers have other methods for ionospheric correction. *Both* GPS and Galileo achieve 10cm accuracy absolute, because the smart bits are in the receivers (not the naive version you will read in any university textbook).

#2 Nationalist bullshit.

#3 “Galileo’s modern signal structure performs better in dense urban environments”

The Galileo signal structure “upgrades” are a load of university professor crap. In theory it makes a difference. In practice, precisely in dense urban environments, an advanced receiver blind-detects and de-convolves it all out anyway. This was originally discussed and designed thirty years ago, when getting even four or five RAKE fingers in a reasonable CPU power budget was ridiculously challenging, and the Galileo signal structure would definitely have been better. Today, we have x1000 processing power on the receiver, this really is a non-issue, the signal upgrades make zero difference. The only thing that really matters is the total satellite signal transmission power.

#4 Emergency Servicss. Cospas Sarsat provides this service already, as does Inmarsat GMDSS. And have you seen the actual spec on reliability and timeliness for Galileo beacon detection? By the time you wait for Galileo you’ll be dead.

#5 OSNMA. Sounds cool. Sounds blockchain-ish, Win Win Win.

Seriously, have you asked any banks if they want to use this? The EU asked them ten years ago, and they universally said Fuck Off, only less politely.

#6 Is True.

Re: Galileo

It was in a Commission newsletter last year, about how wonderful Galileo is, that all aircraft chartered by Commission have been fitted with Galileo units.

Now, if Ursula has been using off-balance-sheet planes for official business, that would be an even *bigger* story……

Re: Galileo

Well, I was on the engineering team that implemented part of the Galileo infrastructure. And I was in the room, lobbied for the contracts, and personally signed some of those contracts, when my company “the largest aerospace company in Europe” persuaded the EU that it really needed an independent sovereign version of the US GPS system. And then we proceeded to lose that contract to OHB. I rather fancy that trumps your “I read this on Wikipedia”.

I assure you that having a higher anti jam capability than GPS, was a major part of why the EU believed they wanted this in the first place. And we echoed that back to them, precisely, because that’s what they wanted to hear.

As to “the main reason for Galileo is to be independent of the 'Selective Availibility" of GPS”. It has certainly been a selling point to those not technically involved, that Galileo is independent of GPS SA. Unfortunately, there are two separate technical reasons that if the USA *did* actually turn on SA again, it would similarly degrade Galileo accuracy exactly pro rata. The simplest to explain, is that Galileo is on the same frequency, and shares the US GPS *ground* atomic timing as its single-point-of-truth. If the USA turned on SA, we can assume that they would also stop sharing their UTC timing, so any Galileo secondary timing atomic clock gradually drifts off within the ground infrastructure. This event has actually occurred once, due to a component failure a couple years back, which incorrectly did not have a redundant backup, leading to a week-long Galileo outage. So it’s not theoretical. The mechanism to prevent that drift ultimately traces back to receiving the GPS signal identically on two continents. Selective Availability being turned on, would spanner Galileo timing accuracy, exactly as much as GPS accuracy.

I’m so sorry that you have been fed propaganda. But this is a technical forum, you should have had the wit to check your system understanding before blurting out the first wrong thing that came into your head.

Galileo

Remember, Ursula’s plane was fitted with the best Galileo terminal that EU money can buy.

This is the outcome of€25bn of EU taxpayers spent on Galileo (so far), which according to them was largely spent on special EU technology to make it more difficult to jam. Which was all nonsense of course, all silly Ursula did was take the US GPS spec from forty years ago, and wind it forward by twenty years. Now she wants to spend more money, to make it only five years behind US GPS. And of course, current GPS is exactly as vulnerable as it physics says it is, which is “quite vulnerable”. What a waste of money.

US government snaps up 10% of Intel for $8.9B

Do *you* think that 10% stake is a passive investment holding, a stake reward for providing working capital?

Do you think *Trump* intends it as a passive investment? Or do you think, just maybe, Trump views it as a controlling interest?

If the management dare to disagree with something he tells them to do, either related to their core business of making profits by selling chips, or *not* related to that, do you think he will

a) Jump around and shout a lot on phone calls to the board, as any activist 10% shareholder is entitled to do

b) Threaten to have the CEO and the board arrested or deported?

Go on, if you really aren’t sure if this is expropriation by a dictator, let’s hear you expand your claim that this is “just a 10% stake” more fully, with reference to “voting rights at the shareholder meeting”

US tariff terrors prompt Nokia profit drop, TI inventory binge

Re: Will the EU cave in, also known as 'Doing a Carney'

I agree that you can’t make a deal with Trump. But it would be better just to ignore Trump entirely.

Tariffs are a tax on your own consumers. People have been pointing this out to Trump for six months. Then why would be it different the other way round?

How are you “punishing” the USA? You think higher EU tariffs will make Europeans buy fewer Dodge Rams? Europeans mostly don’t want that crap anyway, and those who do….are entitled to their opinion. It’s not up to us to stop them, and their neighbours will point and laugh, so that’s punishment enough.

Musk is messing with the Cosmic Dawn. Will alien hunters save the day for all mankind?

Re: Geofencing

Unfortunately, the issue seems to be more complex than that (link paper at bottom). There’s several different sources and types of emission from Starlink, each of which needs to be thought about separately.

First, one type of unintended emission comes from propulsion manoeuvres. You can’t just not do station-keeping in the region (at least, I don’t think so). This should have been considered during ion thruster design, but company and domain siloing is a problem in these big megacorps. I worked on a (classic large GEO mission) spacecraft, where some prawn failed to consider that putting the ion thrusters right next to the antenna was not a smart move, and for extra credit the Alfven frequency sat in the middle of the comms band. Fortunately we found that weeks before launch, and it was re-designed to fix, with unpleasant delays and lots of shouting.

Second, SKA-LOW is seeing not just Starlink but several different constellations, including NOAA. So it isn’t just the “move fast and break things” people. The difference, is that the other constellations are sparse enough that SKA can schedule around them; whereas Starlink is in-shot basically all the time. In other words, we don’t actually know that a radio-tech solution is feasible at all. Radio telescopes are just insanely sensitive, maybe you *can’t* filter or switch well enough. SKA are asking for a 20-30dB improvement, which is a lot. Then this would be an intrinsic flaw of dense constellations. Don’t know.

Third, a bunch of it seems to be TT&C (telemetry transponders) transmitting at low EIRP (few milliwatts) in their intended bands. And the authors say that Starlink is transmitting within regulatory limits. What ground control station (SAS) are these satellites transmitting to, which is within the horizon of the radio-telescopes quiet zone? SAS locations are all well-known and should beplanned. Somebody just failed to consider the Quiet zones during SAS location planning. That’s on the regulator….

https://www.researchgate.net/publication/374382316_Detection_of_intended_and_unintended_emissions_from_Starlink_satellites_in_the_SKA-Low_frequency_range_at_the_SKA-Low_site_with_an_SKA-Low_station_analogue

Time for Britain's CMA to strike hard – or risk losing the cloud competition fight

Re: A potential Brexit Benefit

I think your first point is that your specific regulator does not in fact impose significant load. That may well be true. But you seem to have taken the *overall* goal of reduction as targeted at your org. If its implemented blanket reduction, thats crap, but it doesn’t mean the overall govt request is wrong. And as I said, I don’t regard a 1.3% imposed load *in itself* as unjustifiably large, or shouldnt be reduced. There should be an honest impact analysis.

The worst argument ever, and a very common one, is “Something must be done, this is something, therefore this must be done”. If you’re doing “something” but it’s not effective then yes stop doing it. It’s simply not true that doing something is better than nothing.

“Well, you might care to de-personalise your attack.”

Apologies that it came across that way. I should have better said: “It is shocking to discover that a basic part of standard regulator process, and govt process of instructing the regulator, does not include both an estimate of imposed load, effective targets for that load, and ongoing monitoring of that load”.

Ok

Cyberstalker downvote script test @1.18am

Re: A potential Brexit Benefit

“480,000 FTE of private sector do NOTHING but fill in forms for regulators. Like 'em or loathe 'em, how credible do you think that is?”

I’ll suggest at least this data-point: it just so happens we’re having some building work done. It’s moderately complex in admin terms (planning permission, building control and other items), but more to the point it’s not our day-job so we’re contracting that bit out, to be sure it’s done correctly. Price is flat-rate 8% on the actual building works cost, which might or might not be on the high side, but it’s not so far off industry norms. That literally covers *only* private-sector admin to Councils, none of the building works or organisation.

So there you are, the answer is “for the constructio industry, about 8% of the private sector workforce is engaged in form-filling for the government”.

Now, there isn’t necessarily anything wrong in that. Grenfell Tower, for example, could have done with a lot more regulatory oversight.

Similarly, schools spend weeks preparing for an Ofsted visit. GP surgeries, hospitals and care homes spend huge amounts of their time documenting, and preparing for Care Quality Commission. And the food industry. And railways. While only maybe a quarter of the economy is heavily regulated, that which is might well spend 10% of its total time on “regulatory admin” or more.

Honestly, an estimate that 1.3% of the workforce is doing this, neither surprises me and is probably on the low side.

None of this is necessarily waste. It all depends whether it’s producing Safety or Safety Theatre. Grenfell still happened. Our Care Homes are horrible places despite the CQC.

What is a bit shocking though, is that you as a regulator are entirely clueless about the size of the imposed load. That’s a horror story.

Starlink says SpaceX targeting 2026 for launch of Starship-ready terabit satellites

Re: Intersatellite Links?

“Fish” (cruiseliners, oligarch mega-yachts, cargo vessels, offshore oil and gas ) have been a highly lucrative market sufficient to keep one of the world’s biggest satcomms operators in business for several decades - Inmarsat

Re: Intersatellite Links?

I don’t disagree, but you missed my point. Geo and Leo are two different ends of the tradeoff.

Geo is better, if your costs per launched bandwidth are so high that you can only ever hope to have a few tens of thousands of customers who afford that. Then you spend $2bn on three Geos, and focus on serving the high value customers. This is what we have been doing forever. This is what *I* was doing couple decades ago, I was a key responsible engineer and then manager on a Geo system.

However, if your launch costs drop sufficiently, that the demand at the indicated price becomes several million customers, your goal changes. Now you just want to maximise capacity, and you’re prepared to spend much more on the constellation, because the revenue is higher. You’re right that 90% of the earths area is economically sub-optimal, but again the business logic changes. Now, you launch at a scale that gives you sufficient capacity/customer value that your business case still closes on the 10% area which is high-value. And all the *other 90%* is basically free money. Every $100 your customers give you in low value places, is $100 straight in your pocket.

They are just very different markets / business-cases, so obviously they are different technical solutions.

Re: Intersatellite Links?

There’s nothing preventing LEO doing beamforming like a classic GEO, other than preconceptions.

These LEOs are big beasts by comparison with the minnows of the past (800kg+), and could fit phased array if wanted. From 500km up, it would be easy to stomp down 20km diameter beams, if it were advantageous to do so.

This isn’t the real difference between LEO and GEO. The real difference is power budget - you are coming from a position that “15kW total transmission power should be enough for anyone”, which is true if you have customer bases in the thousands or tens of thousands.

If you have millions of customers, you want to transmit with 6000x 1kW…..

Fundamentally, when the economics is high-cost, geo is optimal because you’ll never have enough customers who can afford it, and [three] Geo satellites is enough. When your price point drops, you want a scalable way to increase total transmission power as high as needed for the number of customers.

Other considerations about size, weight and cost of the ground terminal were also key drivers until ten years ago, but tech is now overcoming them.

We simply couldn’t have built Starlink ground terminals at a consumer price point until ten years ago, they cost $15k-$50k each as recently as 2010, because of mechanical tracking mounts.

Europe's exascale dreams inch closer as SiPearl finally tapes out Rhea1 chip

Re: Rhea1

“This chip”…..isn’t a chip. It’s vapourware at this point. It has *taped out*, which basically means it’s good-to-go for a first trial production test.

But it’s not going to *work* on its first spin, because no ASIC of this complexity does. A good batting average for a highly skilled team with 20 years experience, any of NVidia, AMD or Intel, would be second or third spin correct,…for a chip that they’ve already made several similar ones before.

For SiPearl to get fully functional silicon back on their second spin, let alone their first, would be like throwing several double-sixes in a row. It is theoretically possible. But in practice, it’s not going to happen.

They’ll put out a press release, declare it a “success”, and immediately start spending the next billion on Rhea2, without using a single chip in production projects.

UK puts out tender for space robot to de-orbit satellites

Astroscale would be a reasonable bidder.

But it will be Clearspace.

Financial 'stretch' for UK to join Europe's Starlink rival, says minister

Re: Long Brexit.

11 out of 27 countries currently suspended Schengen…..

Re: Long Brexit.

Having been in the space industry, and worked for the relevant companies (and still got the contacts at senior level), I can assure you that

A) IRIS2 is a really shit program, a total waste of money, of little or no use technically even if it were free

B) Its already 8 years behind, despite being kicked off just earlier this year. Officially it was going to be in service by 2030. Unofficially, internal estimates are that *if there is no further slip*, the absolute earliest in-service date is 2038. And thats not for actual users, thats the beta-test.

C) Theres a high likelihood it will be cancelled anyway, although not before frittering away the best part of €5bn on little more than PowerPoints. The problem is, that all the best contracts are due to go to France. And neither Germany nor Italy nor Spain see why they should pay for whats basically a French system.

Chip designers latest casualties in US-China trade war

Chinese costs aren’t randomly lower, for stuff like chips. That’s an EU delusion that somehow the Asia Pacific bad men are undercutting you on cheap labour and cheap raw materials.

TSMC Taiwan costs are lower than EU and USA *because* their yields are higher, *because* their industrial quality process and approach is better.

Labour costs are not a major part of chip fab cost. Labour is a major part of chip fab construction- and if you ask Morris Chang why USA foundries are intrinsically expensive, as many have done, he will freely tell you that US construction labour force is poorly skilled, uncooperative and entitled, very slow unproductive per hour, poor quality (high amounts of rework). You can dismiss his view if you like, but he does have the overview of having done both countries, founded and run the most successful fab on the planet,and trained and worked in USA (MIT and Stanford)

You can’t economically make chips at 50% yield and make it up in volume,you just can’t. And China isn’t stupid enough to try, nor weak enough to need to. It will trade its way to success.

Professors of chip design……

What the open-sourcies fail to realise, is what an academic actually *is*. A professor needs to know enough to lead a *graduate student* to write a PhD thesis. They’re first-line managers of 23yr olds. That’s all. In fields like this, academics only know the basics. They’re needed, don’t get me wrong, in just the same way that nursery-school teachers are needed to teach children to count.

PhD is entry table-stakes to start work at companies like that. You make the coffee, and re-learn how it’s *actually* done by people with thirty uears more industrial experience than you.

Even in the USA, a full professor makes $150k. In a company like Cadence, if you’re even *competent* plodder you make $300k in five years. Somebody actually driving the innovation at mid-career is making double that, and more senior tech people well over a million.

The general rules for modern ASICs are *not* the same as FPGAs. It’s a whole different ballgame.

How long did it take you to close timing and P&R an FPGA - a few days for one person? Using tools that came free with the FPGA.

Would you be surprised to learn that the answer for a modern SOC *given the Verilog source* to get to a mask handoff, takes a team of 200 specialists 6-9 months, costing $200M+. In tools terms, the team will be using licenses ranging in price from $50k per seat per year, up to $2.3M per seat per year for one of them. The Design Rule Checker alone may be validating 10k+ rules on a 3nm-class process.

What it took you to do the job in the 1990s is not in the same class as 2025. That’s *exactly* what I mean when I say that Cadence and Synopsys are 20 years ahead. You are quoting the 1990s job, and free tools exist for that. But the complexity and sophistication of what needs to be done, has been exponential.

As a general rule, you’re right. That applies to rocketry, TVs, lithography, and business processes. Much of innovation is researching a dozen plausible good directions, most of which turn out to have their own issues. Once everybody knows not only that which idea works, but exactly how good it is when working properly, and has a 3x return on investment if you develop it, progress by others is usually quick.

It does *not* apply to the statement “various algorithms and heuristics exist which can combine physical synthesis, place-and-route algorithms, detailed transistor models from silicon vendor, parametric IP models, IR droop modelling; to produce outputs which consume less than one-third the power of the naive version; and which have a 95% yield on vendor process rather than 70% yield”.

This tells you *nothing* about what those algorithms and heuristics might be.

Then you very much read what you thought I was going to say, rather than what I actually said. Nowhere did I say that China was only good for copying.

Most of my business is in the Asia Pacific region, I speak Mandarin, have a lot of respect for both Chinese engineering, and way of doing business.

What I *said* was that in the field of silicon EDA software, Cadence and Synopsys have spent their time and money solving hard problems, are now being rewarded for having done so. They are decades ahead. You can’t just catch up with the magic wand of waving a million graduates at the same hard problems from cold. They’ll have to work the solutions out the long way, like Cadence and Synopsys did.

China has many talented and clever people. And of course “never” is a long time.

But what I’ve told you, is that the best of MIT and CalTech, are currently 20 years innovation behind Cadence etc. One of the reasons for that, of course, is that when bright young PhDs and mid-careers at MIT come up with potential good new ideas, worth investigating Cadence offer them million-dollar joining bonuses, along with million-dollar salaries. And that’s the same whether they live in China or USA.

Chinese companies will take 15-20 years to catch up.

Plus there are other strategic issues to navigate. Like getting tight-coupling with TSMC libraries and design rules, which are their IPx. You can’t just sit with your head down, tap furiously on the keyboard, and magically expect to close these gaps.

No they won’t.

Chip EDA software is some of the most clever and IP-intensive software on the planet. And by IP, I don’t mean “people know what it does, but you aren’t allowed to copy it”, this is classic “trade-secret: it’s just better than anything you have, how it does what it does is unknowable, but it just is”. This is not at all like software compilers.

Open-source equivalents exist, written by professors with 30 years experience in the field for their students. The output of those are chips which are many times more power-consuming and silicon area than the proprietary versions.

Chip EDA software is the un-sung equivalent of ASML photolith, and is *far* more strategic and harder to catch up to.

EU Chips Act heading for failure, time for Chips Act 2.0

Re: Wrong idea?

Why are EU and MAGA both so *obsessed* with what the automotive industry wants? It’s a lousy industry.

The average is more like 1500 chips; 3000 is for a super-luxury fully-loaded with options that almost nobody has. The average price of those chips is about 50 cents each, the bulk is sensors and motor controllers, the *top end* of those 1500 is a bunch of €1.50 microcontrollers. So the total price is *maybe* €1000 of chips….which are selling at 10% margin. We’re arguing about €100 of value. Not even €100 per year, it’s €100 over the 15-year lifetime of the vehicle.

You want to swing your entire industrial economy around €8 per *household* per year. *Toothbrushes* have a higher Average Revenue Per User. Look it up.

Nationwide power outages knock Spain, Portugal offline

Well, no, that is clearly not the underlying cause. That may have been the *trigger* certainly. But their real problem is having a cascading-fault mode at system level.

White House confirms 245% tariff on some Chinese imports not a typo

Re: What would you do?

This is just a weird Republican talking-point, entirely free of truthiness. Since you want to focus on low-cost drugs, ie generics (why? because it suits your argument, and ignores the higher value ones where the EU and US are dominant?)

The generic drugs market is dominated by India, *not China*. Do I have a problem with that? No, I do not. Why should I, their quality-control and logistics in the volumes we are talking about is good, and so is their efficiency and cost-effectiveness.

Why do you have a desperate need to avoid being “dependent” on “India” for packs of paracetamol wholesale at 16p/pack? There’s clearly no economic argument. Among other things, if these were made in Yorkshire, you’d be just as “dependent” on the factory in Yorkshire, and subject to the whims of unionised action, flooding of the Ribbledale Valley etc. What you are describing is just being dependent on less diversified supply-chain. There certainly have been supply-chain shortages of drugs, but that’s usually turned out that there’s only 1 or 2 factories globally for one process-step, which nobody really knew about, and which had a manufacturing glitch. And usually in the USA by the way.

Re: 14 reasons why Trump’s tariffs won’t bring manufacturing back

“import tariff was based at the cost of the product as the point it enters the country and not the price sold in the shops”

This is true, but it’s only half the story when it comes to price-setting. There’s a reason why typical retail markups are 2.2x, somewhat independent of price. To sell something that costs £1000 wholesale, you have to work a lot harder in getting customers, than if it cost £10 wholesale. You need more premium retail locations (higher rent), more effort shmoozing the customer (more and higher-trained sales staff), more marketing/advertising spend, and there’s just fewer customers who can or want to afford it. Very roughly, retail costs will scale with the wholesale price.

The end-customer price might not track perfectly with the tariff. But thinking that the tariff only affects the raw cost of goods, while the rest of the value-chain remains fixed, is just wrong. For a concrete example, USA uses about 7.4billion barrels of oil annually, worth $480bn = 1.8% of US GDP. And yet, doubling the price of a barrel of oil in 2022 caused an oil shock and 8-10% inflation.

Self-driving car maker Musk's DOGE rocks up at self-driving car watchdog, cuts staff

Re: Self-driving is a fallacy

Horses for courses. About 20% of the country is either too elderly to drive safely, or too disabled to do so (eg blind). Most of them live lives that are significantly constrained, to being nearly housebound. So to answer your question “how much is it worth” - maybe £5-10k per year? Whatever a taxi would cost you, if that were your only option.

EU lands 25% counter tariff punch on US, Trump pauses broad import levy hike – China excepted

Re: Trump blinked

So…..basically QE, “print dollars”? Yes, they can. This will tank the dollar exchange rate, so all US imports will become much more expensive.

There are half a dozen technical ways for the USA to cope with the fallout, but all of them end with the words “so all US imports will become much more expensive”.