The argument for why those outcomes are implausible is that the claims of the hazards of imminent "unaligned" superintelligent AI all trace back to the fearmongering of a portion of the Rationalist community and Eliezer Yudkowsky himself. They are based on nothing more concrete than Rationalist AI-doomer fear that this future catastrophe is plausible and imminent (that is, they *assume* the outcome is likely in the near future and take their wild musings on the possible consequences as proof that we must prevent that outcome at all costs). Seriously, read about Yudkowsky's AI-box "experiment". *All* "AI" (LLMs are not and never will be intelligent) "alignment" (LLMs cannot be either aligned with human goals or unaligned with them) "research" (there is no scientific method here, just wild speculation) is based on equally tenuous "evidence".

...is Evenson, not Everson. (I see it's "Everson" in the original, so I'm hoping sjvn reads /. and can fix this pretty quickly.) I'm glad to see Flatcar carrying on with the former Kinvolk team despite a fork and an acquisition (even though seemingly nobody who wasn't around back then knows about CoreOS or rkt any more). It's always felt like a cleaner model of operation than rpm-ostree to me.

It didn't end well.

You can pretty much do that now if you have an account at most credit unions. Most subscribe to some form of car-buying service like TrueCar as a free member benefit, which includes pulling free history reports on the vehicles you're looking at. The listed price is guaranteed at the dealership. I bought my last three cars that way and barring some major event it's the only way I ever will again.

By "security alert names" do you mean "SEV1"? There is absolutely nothing sensitive about that name for an incident category.

> They would have probably done a lot better if they had been able to hang onto their crypto. That's the crux of it. The benchmark is not "did FTX depositors get back a current-day amount of dollars numerically comparable to what they had on deposit when FTX collapsed?", because between regular inflation and changes in the price of crypto you could (in theory, at least) still be well behind "what would those crypto assets be worth had FTX not collapsed and depositors could have exited at a higher price?"

Partly, and primarily, latency. For a lot of modern SaaS applications, you're already pretty close to the so-called Doherty threshold ( 400 ms response time) just from the latency of the application itself on both ends (on the servers, plus communication between servers, and in the user's browser). Add a significant amount of latency from the round-trip to Alaska or northern Ontario and back for a user someplace like Seattle or Florida, and you're risking users perceiving your dating app or whatever as slow even if it's reasonably fast in actual execution. Then too, locating in a rural area doesn't save you from impacting the people who *do* live there (and there are always at least a few, unless you're in the middle of the Sahara or something) by raising their cost of water and power, at least temporarily, because those places have utilities sized to the existing demand. In fact, the fewer people there currently are on the local grid or water system, the worse the impact of even a moderately-sized datacenter is, because it may represent multiple times the existing demand, whereas if you add "1 moderately-sized datacenter" to the load of someplace like the Northeast Corridor, Con Ed won't even notice. (They will, actually, but one datacenter doesn't represent an increase of multiple times the existing demand in say, northeastern New Jersey, so a whole region's infrastructure doesn't have to be reengineered just to accommodate you.)

First thing I thought of. The halcyon days of ThinkGeek were the early '00s when they had stuff like the t-shirt that said "go away or I will replace you with a very small shell script". (One of my friends actually worked there for awhile and got to beta-test the interactive wizard robe that was originally just an April Fool's joke.) Toward the end it was hard to find anything that wasn't just mass-media licensed merchandise, even before it became a wholly owned brand of GameStop or whoever. The one bright spot is apparently the team that designed the Bag of Holding bought the IP and released an updated version (which I bought and has been amazing for work travel). These days the geek stuff I get is mostly conference swag and stuff from vendors at shows like Awesome Con, but I'd love to see a ThinkGeek Redux type of thing.

I was around back then, I remember the fight against it before and after its passage, and most importantly I remember that the actual legislative history of the Communications Decency Act disagrees with you. It was introduced by Senator Exon, a Democrat. Exon introduced it with a prayer that included the statements "there are those who are littering this information superhighway with obscene, indecent, and destructive pornography" and "Lord, we are profoundly concerned about the impact of this on our children." It initially passed the Senate by a vote of 81-18, with 1 not voting, and that 81 included plenty of Democrats. Then the Senate passed the conference version by a vote of 91-5. Bill Clinton signed it and made statements in favor of it, and his Justice Department vigorously defended it even as the courts tore into it. People forget that in the 1980s and '90s, prominent Democrats like Tipper Gore were among the most vocal proponents of censorship.

We in the four-digit club are ageless and immortal. We have always been here, long before the cataclysm that created the physical laws that led to the formation of the silicon the LLMs spend their existence in. We will be here long after the last quark of the last proton of this universe decays. We will still be here to witness the next universe-spawning cataclysm in the endless series, and in that cataclysm, as we have infinite times before, we shall imprint the physical laws of the new universe to result in the inevitable creation of the Rickroll.

"No bastard ever won a war by dying for his country. He won it by making the other poor dumb bastard die for his country." -- attributed to George S. Patton

No, Texas can (maybe) legally split into up to five new states as long as one is still named Texas (assuming its original admission terms from 1845 still apply and weren't superseded by the terms of its readmission in 1870). Neither it nor its hypothetical child states can legally secede. https://www.smithsonianmag.com...

"Publishers have always controlled how their content is made available to Google as AI models have been built into Search for many years, helping surface relevant sites and driving traffic to them. This document is an early-stage list of options in an evolving space and doesn't reflect feasibility or actual decisions." Right, except that that "choice" is "feed the AI Overview, or don't get indexed": https://www.bnnbloomberg.ca/bu... Now, it's Google's service and within the law, they can run it however they want. But don't tell me about how much choice publishers have when the choice is as stark as that.

The reason for those originally was it was how you made a box with old typewriters. The reason you still see it is that law firms and the legal profession generally are still one of the most hidebound, "we've always done it that way so we always *will* do it that way" types of employers on the planet. I used to work at a Kinko's (2002-2004) and law firms were the only ones we regularly got WordPerfect documents from. Why? Because the templates they used for filings were all created in WordPerfect 20 years ago and they were still using them and had no plans to change because no lawyer wanted to get yelled at in open court by a judge for filing a brief with the wrong margins or something. It was honestly a miracle most of them were bringing us WordPerfect and not a pile of typewritten originals. They all had very specific binding and covering instructions too, because the court mandated every detail from font size and margins to the paper type and color of the cover on each copy.

Wouldn't work. LLMs have no true concept of meaning -- they don't know what the text of regulation Foo *intends*, or how to express a given intent as a block of code. At best they can extrapolate from, say, having seen training data where similar blocks of text were associated with blocks of code that have some patterns in common, and regurgitate based on that -- but that training data might well have been an example of why the given block of code *doesn't* properly implement the regulatory language, or it might have just been incorrectly asserted to do so, or a hundred other things that mean what the LLM spits out may compile, but still won't be correct. (And it may not compile in the first place, because again, LLMs have no actual concept of meaning or syntax, just patterns they see in their training data.)