That is like me hoping I'll win the lottery. I don't play.

Well done, have a cookie! Unfortunately you failed to take the next step. If population density isn't uniform then literally none of the overall population density based arguments make sense.

Those areas to dense for even EVs? Perfect fit for efficient, high quality public transit, and good pedestrian and bike infrastructure because you have a population density of higher than inner London.

In order to know which weight to change, don't you need to understand the complete model first

Basically: no.

In the most straightforward case of training with SGD on labelled data, you don't need to understand the model, you just know what the output is, what it should be and what the differential is with respect to all the weights. Then you take a step in that direction. No understanding needed.

The same holds for other cases. Let's say you get a data set of things the network does badly on in a certain way (like weird repetitions). You can run the network and see which activations consistently pop up with high values relative to a control set. You can then go and down weight them which can suppress the behaviour, for example.

Not read the paper yet but those are two examples.

Unless you have offstreet parking

The USA is so spread out that everyone needs to drive gasoline cars everywhere yet somehow so compact no one has a drive to charge an EV.

Computers did not, steam engine did not, cars did not.

I don't think you are right about that. They've all caused absolutely massive changes. Steam engines drove the industrial revolution.

lol oops trying to avoid the lameness filter which hates the Nahtzees. We all know who invented the modern highway!

Apparently slashdot doesn't like mention of the third empire, Nahtzees or Hartler.

It's ironic that you complain about the military while using the internet; which is based on the work done by the US military when it developed ARPANET sixty years ago.

And the web was invented by civilians working on basic science research.

Food preservation

People have been working on food preservation for millennia. Refrigeration via artificial means (i.e. not using natural ice) was invented by civilians (one cycle even invented by Einstein). Freeze drying apparently is very old, since you can do it in very cold climates using the weather. Modern freeze drying was invented by a civilian but it a heavily adopted by the military from near obscurity.

GPS: sure.

Radar: sure, the military put a lot of work in, but it didn't start there.

Nylon: nothing to do with the military
Epipens: autoinjectors or adrenaline? Autoinjectors predate military use and while they were basically the first major adopter, those ones are nothing like epipens.

And highways? You know who ELSE liked highways :) The modern highway (special intersections, separated carriageways, type restrictions) was a creation of the for the greater glory of the as something of a propaganda/self aggrandizement tool. For it to work well the dictator pushed for a "people's car", or in German, Volkswagen.

Superglue... kinda, but only kinda.

And duct tape, sure, but a lot of the heavy lifting, like pressure sensitive adhesives and tearable cloth tape was done in the civilian world. But that specific formulation with the plastic coating? OK sure.

Most of these things were initially developed without military funding. In some cases, like RADAR and GPS, the funding advanced them hugely in a short time. And in some cases it was the initial impetus for building them in America (interstates). Though in most cases this is a deeply inefficient way to go about it.

Yay another perpetual motion machine!

If you hand me a working reactionless thruster[*] I can hand you back a perpetual motion machine. If you don't believe in free energy then you cannot believe in reactionless drives either[+].

[*]more efficient than a photon rocket.

[+]Well for certain values of "you". There are some people here who will just screech invective when presented with basic physic, so it turns out a person can continue to believe hi angel'o'sphere

Do you think every CVE out there has a fired human associated with it? In order to have a sensible discussion about how LLMs relate to security relative to people you need to be realistic about people.

Yes, and?

Even if you tell a human programmer to make no mistakes, they'll still write security holes. Yeah no shit LLMs aren't perfect. Neither are humans.

LLMs will cheerfully write egregiously bad ones unless you tell it not to. I find that a bit strange.

I predict that 3 years from now, the current DRAM shortage will turn into a glut.

Depends on how long it takes for the new fabs to come online. As of right now, there's about the right amount of capacity to supply the pre AI bubble world with DRAM.

Unfortunately unsold stock will be less useful because it's the wrong type for many consumer things. Though NVidia could rerelease some of the Quadro cards as consumer ones to make use spare HBM.

Robotics dweebs also deserve it, especially if their creations are used to (1) turf people out of jobs or (2) kill people - i.e. military purposes.

Meh. Firstly, I believe one of my creations is used to kill people, or as an important part of devices that do. But it's also used in a ton of other things. It makes as much sense to say steel is a bad creation because it's used to make guns.

Secondly, I'm not a pacifist. If someone's trying to kill me, I firmly believe in killing them first and I also do not think there is any such thing a a fair fight. Likewise I do not think it is evil of Ukraine to massively develop their arms industry and methods of killing people when it's in service of not getting killed or ending up in the dominion of Putin.

And if the world is going to develop weapons I really don't want it to be the sole preserve of the kind of dictatorial scumbags who don't jut not have your ethics, but positively revel in not having them.

The next AI Winter will be spectacular

One thing I am 100% confident in is that the that the AI winter following the oncoming crash will be every bit as bad as the internet winter following the dotcom crash.

Which is to say: not.

For a lot of things it works way too well and is way too effective to disappear. Sure we won't get frontier model inference for 1/5th of the cost. Some of the "growth at all costs" behemoths might crash as hard as webvan. But you know, online shopping and deliveries didn't die with them.

I mean take my futurology prediction with as much or little salt as you choose. But I'll lord it over you if I'm right and you can lord it over me if I'm wrong. Isn't that the true point of the internet?

While LLMs will probably make that worse

Why though? I'm not saying LLMs will not write security holes, they definitely will if you don't very explicitly tell them not to. But even the nerfed public ones are not bad at spotting security holes compared to people. Most people are notoriously bad at this, but there have also been enough that a huge number of CVEs have been well documented. These are all in the training data and LLMs can absolutely spot these patterns. And they can do it without getting bored on software that was written before those CVEs were found anyway.

The "not innovating" claim every iterative phone release is also kinda silly, since the "slab of glass with some cameras" thing is largely a solved problem

That and also Apple has not had a record of being a particularly innovative company. What they've always been much better at is finding ideas that they can now implement well and polishing them. The iPhone wasn't a particularly innovative device, but prior devices sucked or were curiosities because the tech wasn't quite ready, and the software was too clunky.

And we all know about the iPod here. No wireless.Less space than a Nomad. Lame. And good gried those early devices had excrable UIs.