Yes, I've seen two documentaries about the Human Safari in Kherson. Here's the one I was able to find again [Content Warning!]. It strikes me that it's unrealistic for the Russians to be able to retake Kherson (given how hard they've been struggling even to take Pokrovsk, one of the much easier and higher-priority targets) yet they've still been trying to drive civilians out of Kherson with pure terrorism. And these tactics aren't limited to Kherson, they're just concentrated there.

And this new reality where the landscape is covered in fiber optic cables is kind of expensive and autonomous AI targeting comes next. We were warned in 2019, people!

they are not going to fund things that they disagree with - which every administration does.

No, don't pretend Trump is like past Republican presidents. Legally, Congress has the power of the purse. Once a bill to spend money is signed into law, the executive branch is supposed to be bound to spend it. And before Trump, presidents allowed the law to be executed because they had some modicum of respect for the law. They didn't go around trying to illegally abolish entire independent federal agencies, or refuse to say who was in charge of the organization that did that, or for that matter, deport Americans and non-Americans without due process, or earn a billion-plus dollars from selling memecoins in exchange for pardons, or accept "free" 747s from the middle east for their "presidential library".

It never ceases to amaze that people treat the current administration (or the current Supreme Court, for that matter) as being in any way normal. Trump may not have invaded any countries yet -- although he's talked repeatedly about invading Greenland and Panama and taking over Canada through "economic coersion" -- but look, before GW Bush invaded Iraq, he at least went on a world tour trying to convince several other countries that it was a good idea. He even worked on building bipartisan support! Bipartisan! Remember the days when "owning the libs" wasn't the only thing that mattered? The new alt-right Republican party is way too unhinged to even try to convince any other countries of anything, and they don't want to work with Democrats because they're simply The Enemy.

The debt will increase fast because the BBB creates large tax cuts for wealthy individuals, and to the middle class to some extent. Or rather, it permanently renews the "temporary" tax cuts with which Trump originally inflated the debt in his first term (these cuts lasted throughout Biden's term too). Since the tax cuts reduce the federal budget by a much, much greater amount than the spending reductions to Medicaid, food stamps, and clean energy, the debt is heading moonward. This was Trump's campaign promise; people voted for it.

The difference this time around is that (1) interest rates are much higher now than before Covid and (2) the debt is so large (and congress so dysfunctional) that America's credit rating has been downgraded. Both factors increase the amount that the government must pay investors to convince them to buy U.S. treasury bonds.

If its expiring provisions were made permanent, that total would grow to more than $800 billion of additional interest costs -- resulting in interest payments on the debt approaching $2 trillion annually. - Committee for a Responsible Federal Budget

Reportedly, total tax revenue is $4.92 trillion, so this should mean that over 40% of your taxes should go straight to interest payments in the future. But since U.S. politics is thoroughly broken, what will happen instead is that the interest payments will be made by creating and selling even more debt (IOW even more borrowing). This in turn will inflate the money supply, causing severe inflation.

It's 96 Gbps (Gb/s), not 96 GB/s. 96 gigabits per second is 12 GB/s.

If it were up to me, I would start by introducing a simple lossless or near-lossless compression scheme inspired by QOI before doubling the data rate yet again. (It would require variable framerates, but there's nothing technically difficult about that, I think? The only thing holding back VFR in the past was NVIDIA making this incredibly basic feature proprietary in order to charge licensing fees, but eventually they did relent.)

I'm also bitter about NVIDIA and the HDMI people killing off the market for stereoscopic 3D displays. The HDMI people, for their part, only seemed to care about movies, so they originally designed a 1080p 3D format that reduced the framerate to 24 fps while also using high-latency side-by-side packed frames. You could get 60fps at 720p but still with high latency, requiring both frames to be fully rendered before transmitting the first one, and of course if you had a shutter-glasses display, only one frame could be displayed at a time so requiring both to be fully transmitted pointlessly increased latency. Their decisions were very dumb because making 3D movies and TV shows is very laborious, whereas generating stereoscopic video game content is very easy for game developers and can often be done automatically by video drivers.

NVIDIA, meanwhile, insisted on only supporting their own expensive proprietary 3D glasses system ($150 + $150 per pair of glasses) which guaranteed that the market for stereoscopic 3D would be very small. After a few years NVIDIA got bored because they had few customers, stopped maintaining their 3D drivers and, in newer GTX and RTX models, removed support entirely, and that's why I returned my new GTX and bought a power-hungry AMD space heater instead. (Not that AMD is any better, I was just boycotting NVIDIA for bricking my hardware, but I heard there's a third-party stereoscopic solution for AMD.)

To me the most interesting thing about this whole story is how CEO Stockton Rush didn't worry about the dangers, and even went down on the dives himself, including the one that killed him.

I find it errily similar to the downfall of SBF of FTX (and also eerily similar). Both men seemingly took huge risks because they lacked a sense of danger. In both cases they might've had a successful business if they hadn't been so reckless, though this seems more clear of SBF who could easily be a billionaire today, instead of an inmate, if hadn't done all that fraud.

Anthropic was at least partly supportive of the AI safety bill that was vetoed by governor Newsom. But Anthropic first and foremost considers itself a business, so it's racing against the competition in the hope of earning more money than said competition. Those of us concerned with AGI's catastrophic risk are well aware of the dynamic where people can believe AGI is dangerous and race toward it anyway because it is also a golden goose that lays golden eggs. The people racing in this way probably think the risks are overstated by the "safetyists", yet can still believe the risks are substantial as they run the race.

A lot of people act weird when AI companies cite risks of AI like "it might kill us all" or "it might take all our jobs", treating this like it's somehow a way of hyping up or fundraising for AI?

You know the cigarette companies did their best to deny the link to lung cancer, right? It's not in companies' interest to say their product could have bad effects. The reason AI company people say these things is because they believe them. Many of them have a background in thinking about AI safety and AI alignment, whether because they were involved with OpenAI, or because they were involved with EA (effective altruism), or because they saw some AI Notkilleveryoneism Memes, or because they used their head. And a common mode of thinking is that "if we don't build it, someone else will. We enjoy AI research, so we might as well be the ones to build AGI."

Some people do treat the expected job losses as a benefit to human freedom, but I think any intelligent person could see that the benefit requires that people still have some form of income, such as UBI, which, uh, Republicans won't allow.

We're not talking about AIs simply reading the internet. You remember this recent /. article?

'Torrenting From a Corporate Laptop Doesn't Feel Right'

[...] Last month, Meta admitted to torrenting a controversial large dataset known as LibGen, which includes tens of millions of pirated books. But details around the torrenting were murky until yesterday, when Meta's unredacted emails were made public for the first time. The new evidence showed that Meta torrented "at least 81.7 terabytes of data across multiple shadow libraries through the site Anna's Archive, including at least 35.7 terabytes of data from Z-Library and LibGen," the authors' court filing said. And "Meta also previously torrented 80.6 terabytes of data from LibGen."

"The magnitude of Meta's unlawful torrenting scheme is astonishing," the authors' filing alleged, insisting that "vastly smaller acts of data piracy -- just .008 percent of the amount of copyrighted works Meta pirated -- have resulted in Judges referring the conduct to the US Attorneys' office for criminal investigation."

Hasn't the RIAA sued people for millions just for downloading few songs? And when's the last time you pirated even one single terabyte of books, hm? I'm not sure how this, then, could be okay.

Don't kid yourself, all the big models are trained this way. New models don't learn every human language and every field of study up to college level in three weeks flat by being smarter than us. Someday that'll happen, but for now the industry is based on extreme piracy. If they were just trained on the internet, they would be as stupid as the internet.

But if Trump fired Shira Perlmutter over this, I'm sure the next leader of the Copyright Office will have a more "Might Is Right" perspective on it. Now if you'll excuse me, I have to get back to work on my new "disinforsloption" tool designed to help confuse the public into thinking real images are actually AI-generated.

It boggles the mind that people confuse a statement saying "Republicans bad" with a statement saying "Democrats good". No, I think the mismanagement of California by Democrats has been terrible. But mismanagement by the Trump GOP has been even worse. Don't let the two-party system melt your brain, okay?

The party of John McCain and Mitt Romney is well and truly dead. Following the crank realignment, when you see smart, caring independents nowadays -- Scott Alexander (ACX), Nate Silver, Eliezer Yudkowsky, me -- they're always aligned with Democrats. The Ukraine vlogger Professor Gerdes always identifies as Reagan Republican, but I don't believe for a second he voted for Trump.

If you think 1.8 million is implausible for a single dark web site, then like most people you're probably not familiar with social dark matter.

Note that there should be a difference between "CSAM" and "underage porn", but activists have pushed for the second to be labeled as the first. I would guess that there are millions of teens doing various perfectly legal activities *ahem* every single day, but if they should make a record of it with their cellphone then they are "producing CSAM", and if they intentionally distribute it, then they are "victims" under activist logic. Maybe that's not important here, given that it's called "KidFlix" not "TeenFlix", but in general, if the law says that teenage masturbation is legally the same type of crime as a 3-year-old with a 43-year-old, there's no incentive for dark web markets to favor one type of content over the other. We can assume some customers wanted the first type of content, and others wanted the other.

in the video it turns out that Jobst did, in fact, find conclusive evidence that Elon did in fact come in second place in the first professional Quake tournament. Which is all Elon ever claimed.

The recent controversy about Musk wasn't about his Quake play, but yes, that video focuses mainly on the Quake angle. That's not all Elon ever claimed, though. Elon is shown in the video repeatedly saying things like:

"I was you know maybe one of the best Quake players in the world, actually won money for what I think was the first paid tournament in the US."

"I think like we made a few thousand dollars or something."

Jobst found it strange that top players at the time didn't remember any such tournament, nor could he find out anything about it by googling. But he uncovered that actually a couple of people did remember what happened. It was a tournament held on MPlayer, as Jobst explains:

MPlayer was a brand new paid online platform that the best Quake players didn't use. MPlayer did try to promote their event but the promotion was so poor that by the time news of the tournament reached beyond MPlayer to places like NextGen and Blues News it was already passed the registration deadline. None of the top players knew this tournament was being held before the registration cut off, which is why you just don't find it mentioned anywhere. Even still, the fact the tournament was run on MPlayer at all meant that most players simply wouldn't have been able to compete anyway due to their dialup connections.

[Later] To then go on to claim because of this second place victory that he was one of the best Quake players in the world is honestly pretty insulting.

Also, Musk's team won $750 for second place, not the "few thousand" he remembers, but I see how a billionaire might forget such details.

You could've us some time by linking to the summary, where Jobst also points out in regard to the recent controversy that Musk said:

After so many deaths, I'm carefully building up a @pathofexile hardcore character. Might make it into the top 1000 ladder in the next few days.

and that "it's obvious to anyone with half a brain that streaming the character is implicitly taking credit for it. Unless you state otherwise, people are going to assume that the only person playing your character is you." So there's that.

And in relation to Diablo 4 Musk claimed to be "in the top 20 in the world" on Joe Rogan, but Jobst says the "top 20" claim relates to "pit times in Diablo 4" in a mode where you "enter the dungeon fully equipped" so "the most important factor isn't your skill, it's your gear, it's the specific build you choose to use which involves many factors that have nothing to do with skill." And Elon admitted to letting someone else play his character. When asked "Have you level boosted (had someone else play your accounts) and/or purchased gear/resources for PoE2 and Diablo 4?", Musk said "It's impossible to beat the players in Asia if you dont, as they do!"

And therefore, you think, AGI cannot exist.

So many people want to define intelligence to include humans and exclude machines--highly ironic given how "smart" the average human is. But some people actually are really smart, really inventive, and really trying to work out how to replicate intelligence like their own in a machine. And when, someday, those machines outsmart us (or "use brute force computation", as you will say), calling them stupid won't actually help you.

I remind you that OpenAI's "core values" begin by saying:

AGI focus: We are committed to building safe, beneficial AGI that will have a massive positive impact on humanity's future. Anything that doesnâ(TM)t help with that is out of scope.

We're not stupid. We know AGI doesn't exist. But several companies are in a race to create it. Take OpenAI: Its leader Sam Altman managed to get the entire safety-oriented nonprofit board fired, after the board tried to fire him because they didn't trust him for reasons that are still not public. Now OpenAI is trying to delete its safety-oriented nonprofit part and go full for-profit.

Some people don't think real AGI is possible--you know, the kind of AGI that could learn to drive a car by practicing, much as a human would... or take a course to learn something new and then teach that same course... or use deepfakes to run for office while pretending to be human... or orchestrate a coup. But how sure are you? Because I would draw your attention to how fast AIs are today.

Sure, ChatGPT is less than 2.5 years old, but I don't mean how fast they are built, but how fast they run. I've seen LLMs running on new AI chips that let them write an entire page of text in one second. I've seen hi-res photorealistic images generated in under five seconds. Right now they're just imitation machines, yes. But they're f**king fast imitation machines. They can recite more facts (and bullshit) than ten Ken Jennings, and probably write more in a month than all humans on Earth can write in a year. My daughter is two, and shows no sign of comprehending the word "why". My one-year-old still can't say "up". Meanwhile, a new AI that speaks 30 languages can be created in three weeks, then copied onto a million concurrent instances housed inside a single data center. So what if you're wrong--what if my concerns aren't stupid? What if AGI is built and thinks vastly faster than we do? What if it isn't completely safe and completely benevolent?

All these LLMs saturating the world grew out of a neural net architecture invented less than eight years ago. The difference between then and now is the billions, and billions, and billions of dollars of investment money funneled into this field today with the goal of making AGI as fast as possible.

So what's stupid? Are all these investors stupid to think that AGI can exist at all? Is it stupid to think this is all happening too fast and that not all corporations involved in this race are responsible? A while ago I tried to paint a picture of how and why this could end badly. And I have an unfinished manuscript for a much longer story about this.

For what it's worth I am up thousands of dollars on Polymarket, a site where people bet on the future. But I can't really predict what will happen once real AGI arrives--and not for lack of trying. Total utopia and total extinction seem about equally likely to me, and there are many other potential outcomes besides. I am not dumb enough to think AGI will never be invented. But I do want to delay it long enough for my children to grow up and experience what life was like before, in that time when humans still ruled the world. That's what a ban will do. A ban will not prevent the invention of AGI. It just delays it awhile. I would settle for that.

When I get an incoming voice call, I want a new "Answer with anti-scam bot" icon. Oh, and and for those who run a business, "answer with AI secretary". (Would Android or iOS allow an app to do this?)

Trump said the case represents "an unprecedented, novel, and difficult tension between free-speech rights on one side, and foreign policy and national security concerns on the other."

No he didn't.

Have you really not heard how Trump speaks?