gravescale

Last I heard it was the back of the plane that was statistically safer. Most people still seem to want the front because it gets you off the plane faster and puts you at the front of the passport queue. Especially on larger planes going to places with lots of non-locals going through the slower manual channels.

At the top nodes and volumes, indeed there is no competition. They have a huge headstart handed to them basically by Japan screwing it up in the 90s and no one else being able to step (get it?) up quickly before the gap opened.

However, if they can't keep moving, competition that was trailing behind on larger nodes could get close enough that they become competitors for the high end. At first it would be slightly bigger nodes, but cheaper, which might be an acceptable tradeoff, as not everyone in the world is chasing performace over price. CPUs are so gruesomely overpowered these days that it may not really matter that much in many end applications.

Think AMD coming up behind Intel while Intel was thrashing around.

How long the AI hype continues may be important: if AI capabilities are needed in end-user equipment and that requires the real cutting edge processes, ASML/TSMC keep the advantage.

Anyone is a stretch, not everyone is so keen to take the US side in the trade war at their own expense. South American and African trade is still pretty brisk and growing as far as I know. Mostly they don't have the possibility of onshoring industries that they didn't have in the first place, as opposed to the West where it's more like attempting to re-shore previously-discarded capability. So they don't really have a horse in the race, they're going to be importing anyway.

"Anyone planning a major semiconductor fab in the 2020s", then quite probably, yes.

What's nuts to me is the time scale on the bottom. That is not a lot of time for something so complex.

Then again, if they do an Intel 10nm+++++ and fail to maintain velocity even for a couple of years, their competition will close that gap very quickly.

Sounds like the UK too. All my school friend doing non-STEM courses would ask me how many "contact hours" we had per week.

I can't remember exactly how many, but once you included lectures, scheduled labs and tutorials, was not that far from 9-5 (minus the standard university-wide Wednesday afternoon sport slot).

The friends would go quiet and say "oh, we have 12".

And that didn't even incude the "non-contact" hours in the labs grinding the coursework until security shooed us out late at night so they could lock the building. Good times (in retrospect, especially).

As a lifelong addict, I literally cannot live without my regular dihydrogen oxide. Even a day without is painful.

My prediction: a country which has the political and technical ability to do so and is located somewhere where global warming starts being a major threat to it. Likely candidates: India and neighbors (wet bulb events depopulating large area) or China (inundation of the biggest population centres). This country eventually says "fuck it" and just does it, international opinion be damned. Everyone else in the world will proceed to performatively freak the fuck out. But, they'll all be watching closely and if it works, they'll all be doing it fairly soon. Again, if it works, in the eventual biopic, it'll be mentioned as an aside that they weren't 100% sure if the atmosphere would be destroyed or not, but they forged ahead.

The good news, since an internet forum's commentariat opinion isn't relevant to whether it happens or not, is that humanity is a couple of teratonnes into a 200-year experiment of many kinds of atmospheric modification and nothing critically non-linear has happened yet. Though we certainly don't have many sigmas of confidence that it isn't going to go non linear (but in the "hot" direction) at some point!

We're even doing a cute little cross-over trial specifically on atmospheric sulphur dioxide injection (just entered the second placebo phase)

Don't look now, but humans dump a lot of oxides in the atmosphere. Millions and millions of tonnes. Gigatonnes if you incude dioxides.

Normal capacitors have to have absurdly thin dielectrics to get any appreciable charge storage. An aluminium electrolytic capacitor (the can shaped ones) use just thickness of an aluminium oxide layer. Any sandwich thickness that would be practical for thermal insulation will have essentially zero capacitance.

And even if you could have a super thin, delicate layer in the walls, and the builders don't break it putting it in, why not just roll it up tightly and put it safely in a can along with any electrolyte it needs in the first place?

The key feature of supercapacitors is that they exploit 3D effects (or rather a 2D effect around very tiny features in a 3D volume) that allow them to use the volume of the material, combined with a very tiny effective charge separation.

Models can be trained more and fine tuned, though, if we're going to stick to the analogy. But in the context of the analogy, the LLM won't be materially updated between two prompts in roughly the way that telling you that the answer you seek is in a book with a specific ISBN isn't materially affected by someone publishing a new book at that moment.

You are quite right that you're not convincing me of your original thesis that that a prompt contains the entire content of the reply in a way that some other reference to an entity in some other pool of information to doesn't. That's not the same as saying "ISBNs and LLM prompts are the same thing", which is a strawman. It's saying that they're both unambiguous (assuming determininism) pointers to information.

Of course no-one is disagreeing that a reply from a deterministic LLM would add no information to the global system (you, an LLM's model, a prompt) than just the prompt would. But I still think the same is true for the content of a book not adding to the system of (you, a book store, an ISBN).

In fact, since random numbers don't contain new information if you know the distribution, one can even extend it to non-deterministic LLMs: the reply still adds no information to the system. The analogy would then be that the book store gives you at random a book from the same Dewey code as the ISBN you asked for. Which still doesn't increase the information in the system.

Big things have to be expensive (houses and cars, they're working hard on education and health) or money will pool up down here and won't trickle back up to where it belongs.

I don't understand the distinction. If the book archive is electronic, like many in fact are, why can you not get a copy of the book with a given ISBN without altering anything? Even if it's not electronic, does the acquisition of a book by an individual meaningfully change the overall disposition of available information? If you took the last one in your local Waterstones, I can still get one elsewhere.

You can go and fetch the book from a book store using the information. Fundamentally there's not much difference between that and "fetching" the output from some model using the matching prompt. In both cases there some kind of static store of latent information that can be accessed unambiguously using a (usually) shorter input.

I'm not saying the value of the returned information is equivalent, of course. But being "just a pointer" into a larger store isn't, in itself, the problem to me.

Indeed you can losslessly "compress" an LLM's spew into just the prompt (plus any other inputs like values of random variables).

But you can also compress a book's entire content into just its ISBN.

It's just that books are hopefully more than just statistical mashups of existing content (some books like textbooks and encyclopaedias are kinds of mashup, though one hopes the editors have more than a statistically-based critical input!)

"Best X of YYYY" articles have been (mostly? fully?) automated mashups of tech specs for years too.

Or hacker for that matter.

Actually "engineer" as well in countries where it's not protected. Though the confused look on my German housemate's face when an "engineer" turned up to connect the cable broadband cable to the property was pretty funny.

I do wonder how investors who are supposed to be savvy so easily get scammed by BS that wouldn't survive even the table of contents of the due diligence.

Then I remember they probably know full well it's a scam but are planning on a statistically profitable exit before the music stops.

Maybe not the best source considering his own hubris got him killed.

To paraphrase another 2023 film: "That Russian a-hole is going down. What's he gonna do, snow on me?" -- Napoleon

Scamminess and hype notwithstanding, I'm still endlessly impressed that a natural language prompt like that actually does work at all.

Right, if this is true (possible I guess, but does sound a bit Syndrome-from-The-Incredibles), even if there's really no technical way at all to terraform Mars it's hardly the end of the world, uh, planet.

Not least, terraforming would take centuries in any plausible effort. If he goes to Mars to die, he'll die in a pressurised environment whatever happens.

Falcon 9s have 2 failures and one partial from 354 launches, so they're proving to be an extremely reliable vehicle on the scale of rockets. The current model is 289 of them and has a perfect record so far.

That's substantially more successful than Soyuz which crashed a lot more and killed someone by parachute failure and crew by depressurisation (and another crew escaped the exploding rocket by launch abort system and another abort subjected the crew to a 21G descent) in the early days and is still considered one of the safer rockets ever (Soyuz-U, the rocket, had 22 failures out of 786 launches, and Soyuz, the spacecraft on top, made 153 missions).

There's still a long future in front of us for SpaceX to screw it up by skimping on things or run into some very bad luck or unforseen issues, but at least as it stands now, pointing at test launches going bang, when they're weren't ever expected to not go bang, and calling the production rocket unreliable is not well supported.

Indeed, I'll believe a company isn't up to some repugnantly slimy shenanigans the moment they volunteer to accurately and carefully explain what data they want, why they think it's useful, how they'll look after it and why it's in my interest to opt in, all without being threatened into doing the very barest minimum by law.

Otherwise it's about as convincing as the school bully being forced to shake hands with his victim while the head teacher is standing over him.

Come to think of it, the mobile operators themselves have a large national network of radio receivers with lots of fancy time-of-flight, phased array multi-path, multi-band capabilities.

Maybe these days Ofcom can just ask them if they are interested in something specific.

You can't send a phishing SMS from a receive-only device. Any mobile tower must have an active transmitter (at the very least, so the handset knows what network it thinks it's connected to!).

Transmitting on a licensed mobile service band without the license is a very good way to earn a knock on your door.

Eavesdropping on tower-to-handset comms is illegal too (in the UK), but it's not very practical to find just a receiver, unless they already know almost exactly where it is and are able to do a TEMPEST-like attack on the local oscillator or something. So as long as you keep quiet and don't do anything to indicate you're listening, such as, posting on Twitter about it or you do crime based on it, you'll get away with it. However, a receiver can't bump a victim handset down to a primitive-enough protocol, so all you'll get is encrypted content and maybe a smidge of metadata (I'm not sure exactly what is and isn't encrypted for each "G").

I don't know if it was involved here, but "spectrum monitoring" (that's the keyword for more information about equipment you can get and who the expected customers are) is definitely a thing that is done in many places, and not only by the military, but also by police and regulators. It's not clear where the mast was installed (the CoL police aren't specifically limited to that very small area and one arrest was South London and one was Manchester) but if it was in central London, as a glaring terrorist target with an airport (London City) and heliport, I'd be very, very surprised if there's nothing watching the spectrum. Whether the operators of the spendy gear would share the information up for such "petty" crime, I don't know.

And there's always the Ham community who really, really hate spectrum abusers.

That's a pretty clever way to be very stupid! Anyone who reports the message (forward it to 7726, spells SPAM) to a network tips the network off that messages are landing on subscribers devices that didn't come through their system.

And I guarantee there are devices listening into, characterising and locating radio emitters in major cities at the very least.

In the relativistic inertial frame of the thrown baseball, it'll not be accelerating, at first. But it will be accelerating at 9.81m/s² (along with the Earth's surface, and hopefully you) after it lands and is just sitting there. The inertial frame will have continued along its original geodesic into the ground, though (or rather, the frame is only instantaneously defined at the moment of the release, but the ball would have followed that path). If you can convince the ball to continue to not-accelerate indefinitely, NASA or Vought International may have a job for you.

In the Newtonian inertial frame of the "stationary" Earth, it accelerates downwards at 9.81m/s² until it hits the floor and comes to a stop and is no longer accelerating ("gravity", which is a thing here, and the normal force now summing to zero).

This is fairly academic for sporting goods, as if working with these two frames results in substantially different outcomes for you or your ball, your day is probably going badly, or at least fairly energetically, and baseball isn't your priority right now.

Which also means that if a hypothetical company is indeed pivoting to AI, an AI-skeptical senior might be considering moving on anyway about now and a blunt-force RTO is one of the straws.

> Finding AI talent, which more and more companies are going to need, is already a search for the proverbial needle in a haystack. Adding strict RTO policies to that scenario won’t make the talent search easier.

Putting that question-begging aside, cynically, for a company on an AI kick it could be intentionally designed to make clearing out the old guard who aren't all in on AI easier and without severance.

In fact, it stands even without the everything-is-now-about-AI journalism gloss. The expensive, grumpy seniors with the pointed questions in meetings are largely the ones who feel most able to risk jumping ship in 2024 and treating them like naughty children or unreliable juniors really winds them up.

The wisdom of driving out senior staff, well, that depends on who you're asking and what are their metrics for success!

In the UK, only people on certain treatments or under non-interventional observation that require it or private patients.

Regular blood tests for normal children and adults (other than the elderly perhaps?) on a purely speculative basis are not a thing.

Edit: actually you do get a 5-yearly test after the age of 40 until 65, my mistake. This incudes a finger-prick cholesterol and blood sugar test and may incude a "real" blood test if thought necessary.