The issue is that the hardware costs money to run. If you don't have a way to generate a proportionate return from using it, then you are still just sinking money into the black hole, and that is not sustainable.

Think about how it works with BTC mining - at a certain coin price and electricity cost, a given chip cannot mine a coin for less than the cost of the electricity to do so. So you would be a fool to run such a chip under those conditions, even if the chip was free.

AI at the moment is not generating anywhere near the revenue required to sustain the operating costs. It's entirely possible that if the bubble blows, many of these data centres cannot find loads that are not loss making. TBF, I imagine if you could discount the capital cost (which will happen when they go bankrupt), then I'm sure they'll be able to find some uses in things like research/rendering/simulation etc, but it will be brutal.

Search for 'Multi-Bit Error Vulnerabilities in the Controller Area Network Protocol'. (It's a thesis by Eushiuan Tran)

This issue is quite subtle, but essential, the fact that the CRC is applied before bit-stuffing means that a single bit error can cascade into multiple errors that exceed the detection limit for the CRC. The potential for this is fortunately rare, but it's like having holes in your bullet proof vest.

This is why CAN FD (apologies, I said 2.0 in the previous message) includes the stuff bits in the CRC (and then has to transmit the number of stuff bits), but this also has a problem, though I can't quite remember the details. It's better though.

Using Chat GPT to encroach on the lucrative search/advert market would have been fine if Google hadn't been able to catch up. But they have. So why would everyone keyed into the Google eco-system bother switching over to Open AI adverts/search now?

It's surprising Altman didn't push for this while he was ahead - that would have been a decent strategy. But instead he went on this crazy train about how they would have the super-intelligence blah blah blah.

The big problem he has now is that if open Ai employees see that their stock options might end up being worthless, they won't stay around, especially since Google/Apple/Meta can give them big $$ right now, not at some hypothetical IPO that might not happen anymore. If he starts bleeding the best employees he is screwed and the end will come quickly.

Even when I studied in the late 90s it was like this for anything applied. There was a push to make the degree more 'industry relevant' - i.e. companies want people who could do Java or Protel, not solve partial differential equations or analyse matrix decomposition techniques. The university wasn't really setup for this, so some of the younger professors who had used those tools as part of their research would run the courses.

Most of the faculty at my school were pretty switched on, so I think it was fine as an introduction. Fortunately outside a few of these introductory courses, most of the course was extremely theoretical (kinda the point of going to university but whatever).

I imagine in the mean time university has become much more vocational, so the demand for these sorts of 'end of the pipeline' skills has grown. The latest thing is AI, so here we are.

This is an embedded system in a high reliability environment. The way these things work is keep-it-simple to an absurd level. I bet you this is some dinky 8-bit RISC CPU that's built on a crazy big process node, and the production QC trace on it will be insane. On these sorts of systems, if you want ECC, you add it to the firmware, but only in the areas you need it, and only after a thorough analysis of (a) the problem it is solving (b) the amount of ECC required to solve that problem (c) the best algorithm to meet the identified objectives. There are many ways to do ECC - including just duplicating variables n number of times - which has the advantage of being very easy to implement and formally verify while being less efficient at RAM utilisation vs a Hamming Code, but even that depends on the statistics of your error conditions.

The point is that, sure, they could add some generic hardware ECC, but that ECC can fail (if there are too many bit flips, if the ECC logic itself gets bit flipped, or there is a design error for a particularly input sequence, etc etc). Maybe you win out overall, maybe you don't - the problem is that you'd have to run a complete analysis to know. That means you have to now add ECC hardware failure modes to pieces of software that did not need ECC before. I mean, sure, maybe you win, but maybe you make it worse, and have to develop extra software to deal with the new hardware failure modes. Whatever the outcome you'll have to do a boat load more documentation to make sure.

I bet you it took them less than a day to identify a fix for the code and update it. It would have then been thousands of hours of work to update all the documentation and thoroughly verify the new code against all the other requirements on the system.

If you want a good example of how quickly these supposedly simple systems can get complicated, look into the CAN bus CRC bug. This fault is present on EVERY system that uses the CAN bus (basically any vehicle since the 1990s). It is an extremely subtle bug involving the error detection system that is obvious once you're show it, but the very smart people who designed it, along with thousands of engineers who worked with it, didn't spot it for around a decade. Even worse when they developed CAN 2.0 they tried to fix the bug, and didn't even get that right.

We are doing something about it, but it's not very good. Look at what is happening to the economy - there is a proliferation of low wage (minimum wage if your country has it) jobs that do not generate enough income to support the worker. The state then taxes the remaining middle class workers to top up the wages of those workers so that they can survive on those low wages. Everyone is getting smushed into a barely surviving precariat group - it's the death of the middle class.

The thing you have to remember is that robots will never drop to zero cost. All that has to happen is for wages to fall enough that a human worker is still cheaper than the robot. Once you start subsidising wages, this price clearing level can fall well below subsistence. If you continue to do this then you can keep humans in jobs as long as you want.

But think about that world - you have humans doing jobs that robots could do, because you are artificially trying to keep them busy doing shitty jobs. Meanwhile you are destroying the middle class to enable this.

This is essentially what we are doing. You can see the results playing out already.

I don't even think this is some grand conspiracy from the 'elites'. It's just what happens when you build a whole society where everyone is expected to work until they drop, and then you automate away all the work. I don't know how you fix this without some completely new social structure. UBI might be it but, honestly, I have no idea. Ideological shifts are never simple nor painless.

The issue is that China understood how to automate things like assembling phones, because they are making all the phones. In the West, there was a lot of steady progress towards automation before we started moving everything to China in the early 2000s. Then all that institutional knowledge got lost because it was cheaper to just get some Chinese workers to make it for us.

IMHO this is such a huge problem for the west. I have tried to make things in the west and you quickly run in to issue where some part can only be purchased from a Chiense company. You then realise that the Chinese company has people who speak perfectly good English, will actually get back to you (unlike many Western companies unless you are from a big name brand), and that they can do quality (if you're prepared to pay for it). You then just think, well, if I have to get some of the stuff made in China, then I might as well just do the whole thing.

I mean, Shenzhen is insane. If you are buying and LCD display, and you need a custom flex cable, you can get in touch with the company that is making the raw polyamide tape and visit their factory if you need too. In the west you used to be able to do that, now all roads lead to 'we need to talk to our supplier in the far east'.

I don't know how the west fixes this. I've talked to people in governance and they are stuck in the 'oh China will build the robots but we will make the software', which is nothing more than a colonialist attitude towards what the Chinese are capable of.

Imagine telling someone back in the 80/90s that in 2025 we'd have driverless taxis that could run from renewable electricity, but people would be bitching and trying to get them shutdown because of the noise from their backup sensors.

We truely live in the age of stupidity.

I honestly think dynamic pricing is just the dumbest MBA think that has ever been dreamed up. We used to have dynamic pricing at markets, and you had to spend ages haggling over everything. The development of the 'price tag' - particularly in western countries - was a massive efficiency increase for the average consumer.

Now we a heading back to the haggling era - when booking flights I have to regularly mess around with user-agents and switching to mobile to check I'm not being scammed by the algorithms. It's just painful, and the biggest effect is that it puts me off bothering to buy anything. Especially for low cost providers, the whole thing was that you could buy from them knowing they had the best price.

It's the same thing with the 'pricing ladder'. It's just dumb. If you don't have a decent enough feature to distinguish between your 'premium' product and your normal one then stop confusing your customers with made up upgrade tiers. Many times I've been dragged up the pricing ladder to the point I decide that it's too much to spend and I just don't buy anything. I mean, dragging people up a pricing ladder by telling them the model they were looking at isn't quite good enough is going to do that.

I still remember the glory days of Apple when you went and bought the 'iPhone' and it was the one you'd seen on the keynote, and you didn't have to know all the specs because it would do everything they'd shown you.

Money. There was that AI guy who went to Meta and got something like a $100million sign on bonus a few months back.

I would imagine the payout this guy is getting is absolutely stratospheric.

You're absolutely right, but the reality is that most people really don't care. I mean, I'm in the tech world, and I gave up caring about privacy a long time ago. I don't do anything particularly dodgy, so if Google wants to know what I had for lunch good for them.

I'm not saying this should be acceptable, but the writing was on the wall for privacy a decade ago when we all started readily carrying around personal trackers and giving our data to Google in exchange for not having to pay.

I admire Apple for holding the line on this (though they had commercial reasons to do so), but my guess is they will water it all down now that it is affecting their ability to compete.

I've given trying to explain this to anyone. I mean, I'm pretty sure I learnt it in high school statistics. Yet it is incredible how much advice and even government policy comes from correlative studies alone.

My biggest bug bear is around second hand mattresses for infants. AFAIK there is one study that showed a small correlation between households that used second hand mattresses and an increase in sudden-infant-death syndrome.

Now, the obvious cause there is that second hand mattress use will be correlated with household income, and low income house holds have all sorts of material disadvantages that could explain an increase in SIDs. Yet if you ignore that and say it's something physically associated with the mattress, then you start getting lots of weird explanations around micro-biome exposure. This is just grasping at straws, especially when you consider that a LOT of people reuse baby clothes. Even in rich households it's not unreasonable to reuse clothes between siblings or pass them on to friends.

But anyway, there are millions of children's mattresses thrown out every year because of that study, and obviously nobody involved in the mattress industry is interested in doing further studies.

This sort of thing pops up all the time. I just have to ignore it because it drives me crazy.

I mean a country is by definition sovereign, so the idea of 'legal precedent' is meaningless. You probably mean that it's not in keeping with the 'rules based global trading order' which is true, but that was/is simply a post-war construct that is getting pretty shaky these days. You're not supposed to be able to slap tariffs on whoever you feel like to strong arm them into doing your bidding either, but here we are.

Ultimately the government of India can fine Apple, or whoever they want, whatever they feel like if their voters don't kick up a fuss. I imagine that Apple has tried to appeal to the government and hasn't made any progress, so they're going for a constitutional ruling since this has authority over the legislature.

But I think Apple still has a lot of leverage here. They probably can't pull out of the country, though I imagine they could threaten it and see if the government calls their bluff. But they have large factory investments there (through Foxconn), people like their products, and I imagine lots of Indians work on products tied to the Apple ecosystem. Closing all of that out would not be good for the Indian economy. There is also wider leverage from the US government. I'm sure Tim Apple can find an even bigger piece of gold to gift to the administration.

Oh there will be plenty. You can imagine a one-hit-wonder type pop-star will jump on this to become the first 'AI artist' and then a bunch of tik-tok artists will jump on that bandwagon as well. Or a 90s pop star trying to get back into the limelight. I could easily see something like the estate of Michael Jackson, Dolores O'Riordan or Amy Winehouse jumping on the bandwagon if it brings in the $$$ - it just depends on who owns their rights in the end. Of course when it gets good enough, you bet that a record label will release a new Elvis single using it.

In the end this AI stuff is going to become the backing track at the supermarket, cafe and airport. Honestly, I'm not sure that's such a bad thing. There is only so much Robbie Williams and Spice Girls I can handle when I'm shopping. If they just have some generic bland backing music that removes the awkwardness of silence then so be it. Realistically, playing something like Animals as Leaders, or some free jazz in those settings would be entirely inappropriate anyway.

The UK has become so slow at developing infrastructure that it's now at the point where if you are in your late 30s you will NEVER benefit from anything they have not already put shovels in the ground for. They can literally talk about whatever they want - its takes so long you'll be on your way to the care home before it ever happens. I moved here 15 years ago when they were talking about 'making a decision on the third runway at Heathrow'. Today they announce that they are about to 'make a decision' on it again. There are things like the electrification of the Great Western line - which would benefit people for the next 100 years if they did it - and they still haven't done. The first electric trains on the underground happened nearly 130 years ago.