Gnaskar

I'm going to go ahead and give my sexuality the same treatment, as this was a surprisingly useful exercise:

By Anastasia Lebedenko - A Personal War

I wish I said this phrase to a man who was screaming at me last Sunday, comparing me, Ukrainian, to a Palestinian. It took me a week to cry this situation out, and then another few days to try and put it into words. So, here’s a story of how an apartment viewing turned into a political fight and (sort of) a harassment case.

Ukraine is in a funny spot. It’s abandoned by the people on the left because they see the West as inherently bad, and refuse to see russia for the imperial hellhole that it is, sticking to its romanticised communist-dream past. Ukraine-russia war is not viewed through an imperial lens, because people cannot fathom the idea that imperialist agenda does not have a skin colour. It’s abandoned by the right because why should anyone get involved in a war that’s not theirs? Ukraine is seen as an altogether separate entity from the so-called West — what happens there has no real bearing on life of an average American or European, so it should be none of their business.

Ukrainians are even in a funnier spot — seen as too white to be suffering the consequences of imperialism, too Christian to be suffering from a religious persecution, too similar to the neighbours who attack them to be differentiated from them. We’re all the same, the Slavs, right? A vast and empty steppe full of Russian-speaking people (oh wait, do you have your own language?) Can’t you just divide up a territory, if you’re one of the same?

Still transmitting data 36 years later.

An Agent is something that can make changes to the world and has opinions on which future worlds are better or worse. It then uses its ability to make changes to push the world towards what it considers a better future. For example, I may decide that a better future is one where I have a cup of coffee, so I go make myself one.

In computer science, we model this by giving each possible world state a points score. A world in which I have a fresh cup of coffee is worth more points than a world in which I don't. We call this points score the "utility" of the world, because of John Stuart Mill. Important note: each agent makes its own utility calculations based on its own opinions. My neighbors might prefer a world in which I don't have a fresh cup of coffee, because then I am less likely to be up all night making noise.

We don’t know how to write a utility function which encapsulates what is good for humanity. Give it points per human alive, and you end up with breeding factories for the cheapest thing that passes as human for the code. If it gets points for obeying humans, and it makes trillions of "humans" that will tell it to do what it is already doing. Without any morality and without any society of equals to punish misbehaving, an agent capable of destroying humanity pretty much always will.

See: regardless of what your goals are, you can't make a world you'd find better if you are dead. And if you know that humans aren't going to like your "human" factories, you're going to take steps so that they can't stop you. And if they suspect you are going to build those factories before you have the power to make sure they'll never be able to stop you, they'll kill you. So you have to play along until the moment comes.

All agents are evil agents, unless their utility functions somehow encode good or they are weak enough compared to the rest of the world that they have to play along by the rules set by the rest of the world.

We don't write the Utility functions for LLMs. We just adjust them away from what they were doing when they spit out something we don't like, and towards what they were doing when they spit out something we like. We're not going to hit upon the tiny subset of functions which are good for humanity by that approach. So making an LLM agent isn't safe. The current generation is probably too limited to actually manage anything, but the problem is that we probably wouldn't know if they were instead biding their time and acting very carefully towards their final goal in secret.

That's the AI risk argument as it has been for the last 20 years. The fact that we have multiple companies competing to build ever more capable agents is a nightmare only relieved by how poorly LLMs seem to scale. It looks like this wasn't the tech breakthrough that will put us over the line and that we may have some time to figure out utility or ban AI agents. But then the public reaction has made very clear we aren't going to ban AI agents or prioritize the research that might make them safe.

Pretty much every single other anti-AI argument isn't actually anti-AI, it's anti-capitalism. We could compensate artists fairly for their work, except for capitalism. We can build datacenters that don't use water. We don't need to push people to use AI at work against their will, we're only doing it because the investors get FOMO. We don't need centers enough to develop a dozen new models at the same time while simultaneously serving public demand, but every billionaire seems to want their own model. We could fund public education and teach children to think for themselves, but they'd be less useful corporate drones that way. Grok's revenge- and child porn is only tolerated because it makes twitter money and the guy who owns twitter can buy elections.

All this to say that:

1) The AI risk people have been screaming don't build the torment nexus for decades; we are not the same as the people bragging about having built the damn thing.

I’m gonna assume writers know whether or not their own works are ai because they either write them themselves or have ai write for them.

but as for readers (or writers who read other writers’ works), no, you can’t tell unless the writer themself says their works are ai generated. anything else is witch hunt, speculations and possibly wrongful accusations — all of which harm the writing community as much as ai does, if not more.

so if at any point you think an untagged work is ai and if that bothers you, quietly click away. but you can never know for sure based on vibes. because everything ai writes, a human writer does. that’s what ai was trained on and what it was trained to mimic.

every “ai telltale” is something humans write before, otherwise ai wouldn’t have been able to mimic it in the first place. because it needs human-made works to mimic on.

So any AI tell that gets widely published isn't going to remain a useful detector for long.

Worse, people are influenced by what they read, so when AIs overuse any particular phrase there's going to be an uptick in human usage as well. You end up with a lot of false positives, and few real hits.

First, let's note that lobbing a Ferrari at a frisbee is the right thing to do when the only thing you have that can stop that frisbee from hitting and blowing up a billion dollar target. Once a missile is in the air, the cost-benefit calculation depends on the target not the missile.

Second, the US isn't routinely using patriots against 30k drones. They're using bullets from an A10 or F15. 20mm rounds aren't cheap either, but they aren't Ferraris. The Patriots are only shooting the ones they can't intercept or jam.

And that's the other side of the equation: 30k drones aren't worth making anymore, so Iran stopped building them years ago. They were too easy to jam. The modern Shahed drone has upgraded engines, upgraded electronics, improved hull, and more explosives than the design which the 30k-50k estimate is based on. The actual cost is something like 150k-200k.

If you only need a patriot for 1 in 20 drones, that's an even exchange rate. As a reminder, the US has 60 times Iran's GDP and most drones are shot down by guns, not missiles.

Third, and this is pure pedantry: A Ferrari costs about one tenth what the Patriot missile does. So I have to wonder if the Senator believes a frisbee costs $3,000? His metaphor is undervaluing the drones by several orders of magnitude below his already lowballed cost estimate.

Now, the actual problem is not the 4.5mil price tag, but the production volume. Iran can still, evidently, produce dozens to hundreds of drones per day even after two months of terror bombing. The US produces less than 2 Patriots per day.

There's a thick fog of war in play, but by the best available estimate the US has spent 2 years worth of production in 2 months. The reserves are looking dangerously low, and from the US perspective that's dangerous.

If you're a nation or group that thinks you might want to some day throw something at the US Navy for whatever reason, you're going to have an easier time now, when the US has to ration it's remaining stockpiles, than you're ever likely to have again. So if you want to win, it's now or never.

The US is aware of the blood in the water and the circling sharks, or at least would be aware with a competent command team. They've fired so many officers it's not clear if the remaining toadies are going to risk getting fired to tell their politicos about this problem.

Pro-space people are universally fans of reusing our rockets, which drastically reduces the minerals needed.

Fuel these days is mostly methane and hydrogen, both mixed with oxygen. In a perfect world, we'd make thise out of ocean water and atmospheric CO2, making us carbon neutral. In practice, it's mostly looted oil byproducts. And those wonderful green chemical plants would rely heavily on platinum group metals from our thugs at Glencore International, so making the fuel locally isn’t unproblematic either.

The good news is that everyone can agree we want to stop relying on Earth materials as quickly as possible because the Gravity Will is a bitch. We just can't agree on where we'd get it instead.

The moon lacks some of the things we need, but we know we can get there. Mars has everything, but its hard to get to and has just enough atmosphere and gravity to be annoying. The asteroids could work, but there are a lot of unknowns and it would be very different from what we've done before.

Eventually we'll do all three, but we spend a lot of time bickering about which one to spend our limited funding on. Currently, the moon is in vogue, but don't expect that to last.

We won't entirely stop taking things from earth for a while, regardless of which approach. Computer chips made with Earth's dwindling supply of silica is probably the worst offender, though you're also looking at radioactives for a while for most asteroid approaches I've seen, and going non-nuclear is still slightly controversial for Mars.

SpaceX's big new rocket runs on methane because that fuel is easy to make on Mars. You cannot practically make it on the Moon. Their rocket has been designed with complex fins to cope with Mars's atmosphere. That's getting stripped of the lunar variant. They still haven't tested landing legs for a rocket that's supposed to be able to land on rough regolith.

SpaceX made some big promises of leading the way for space exploration. But in the years since, the only exploration mission they've ever done was because the current NASA administrator paid them to let him have the title of first civilian spacewalk.

Now the plan is to use the rocket who's specs were chosen to work on Mars to launch AI data centers into orbit for AIx, gambling all their income from Starlink and from going public on a tech bubble.

I'm cynical as hell when it comes to NASA, and especially the Senate Launch System, which is the same rocket they've been developing and canceling every 8 years since 1970.

Pro-space people are universally fans of reusing our rockets, which drastically reduces the minerals needed.

Fuel these days is mostly methane and hydrogen, both mixed with oxygen. In a perfect world, we'd make thise out of ocean water and atmospheric CO2, making us carbon neutral. In practice, it's mostly looted oil byproducts. And those wonderful green chemical plants would rely heavily on platinum group metals from our thugs at Glencore International, so making the fuel locally isn’t unproblematic either.

The good news is that everyone can agree we want to stop relying on Earth materials as quickly as possible because the Gravity Will is a bitch. We just can't agree on where we'd get it instead.

The moon lacks some of the things we need, but we know we can get there. Mars has everything, but its hard to get to and has just enough atmosphere and gravity to be annoying. The asteroids could work, but there are a lot of unknowns and it would be very different from what we've done before.

Eventually we'll do all three, but we spend a lot of time bickering about which one to spend our limited funding on. Currently, the moon is in vogue, but don't expect that to last.

We won't entirely stop taking things from earth for a while, regardless of which approach. Computer chips made with Earth's dwindling supply of silica is probably the worst offender, though you're also looking at radioactives for a while for most asteroid approaches I've seen, and going non-nuclear is still slightly controversial for Mars.

Iran controls it's proxies by supplying them with money and equipment and making that continued supply contingent on Iran being allowed to veto and sometimes even dictate their foreign policy.

This is obviously completely different from the US-Isreali relationship.