The Linux kernel is not from GNU. They had nothing to do with it.

An increasing number of linux distros use musl instead of glibc and llvm instead of gcc.

"Majority of the userland"??? You mean other than the display manager, windowing environment, package manager, init system, file browser, web browser, media player, office suite, email client, and almost everything else that people run in userland? (And I'm just talking about the bundled applications that come preinstalled.)

I predict they'll go right on saying it. It's been many years since GNU code made up more than a tiny sliver of the code in most Linux distros. That didn't stop them. I doubt this will either.

Are you thinking of the 2021 Texas winter storm, when renewables held up great but production from gas and coal collapsed? Of course, the governor then lied about it and tried to claim the opposite:

It's ironic that some people still keep claiming we can't switch to renewables because they aren't reliable. The facts keep showing the opposite. When a drought causes rivers to fall too low to cool your nuclear power plants, renewables keep on working. When someone starts a war that shuts down oil shipments in the middle east, renewables keep on working. When demand spikes and transmission lines get overloaded, distributed solar and batteries are your best friend. They produce energy right where it's needed and keep it off the grid.

All these predictions are for what will happen by 2100, but that's just an arbitrary date. Once we push the system too far, the feedbacks take over and it becomes unstoppable. At that point, even if we completely stop burning fossil fuels, CO2 keeps increasing, permafrost keeps thawing, ice keeps melting, forests keep dying, and the planet keeps heating. The eventual result is that the earth transitions back to a hothouse state (its normal state for most of the last few hundred million years), about 6C hotter than what humans evolved under. It'll take centuries, but it will be unstoppable.

We don't know exactly how long we have before it becomes unstoppable, but not long. Possibly only a few decades.

I think you're asking the wrong question. If you believe that much in private property rights, you should be asking, "Why is the government telling the utility they must let you connect new power sources to their grid, whether they want to or not?" The grid belongs to the utility, not you. If you're off grid, you have a lot of freedom to do what you want. If you want to connect to a grid that belongs to someone else, you have to follow their rules.

The answer is that they're a regulated monopoly. It's not practical to let anyone who wants create a new electric grid, running new cables on the same poles or under the same streets. There just isn't enough room. So the government picks one company and gives them exclusive rights to operate the electric grid. In return, they agree to follow rules set by the government.

Let's do the math. Assume 1000W. California averages 6.08 hours of peak sun per day, so let's assume that's how many hours of solar production you get. That's 6.08 kWh per day, or 2220 kWh per year. In California, which is where this story is about, the average cost of electricity is $0.3325 per kWh, so it saves $738 per year. It's not huge, but it's a lot more than nothing. Assuming it costs $2200 (the upper end of the price range quoted in the summary, which was actually for a 1200 W system), it will pay for itself in three years.

"Launch cadence" has been a standard term in the industry for a long time. https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fbooks.google.com%2Fngram...

This report came from a security company, not a LLM developer. And their conclusion is pretty much the opposite of what you just said: any model that downloads unverified content from the web and treats it as instructions should be viewed as fundamentally insecure. The problem is models that trust unverified content.

Every competent engineer knows you should keep commands and data separate from each other. Mixing them in a single stream is begging for problems.

But is that possible for an LLM? I suppose every input token could come with a binary flag to indicate whether it should be interpreted as a command or not. When training a model with reinforcement learning, only reward it for following commands that are marked as commands. Also include counterexamples that have commands in the data stream, and give a negative score if it follows them.

But I suspect there will always be loopholes. The commands are interpreted in the context of the data, and if you can manipulate the data, you can probably make it misinterpret the commands in the way you want. Given a sufficiently large model whose behavior is unspecified and far too complex to understand, there will always be ways to manipulate it into behaving maliciously. It's impossible to guarantee safety.

And of course, the lack of separation between commands and data is part of what makes the models useful. You ask it to do something, it doesn't know how, so it looks online for directions. If you only want it to follow instructions you specifically approve, it will be much more limited in what it can do.

It sure sounds like they're trying to downplay it: "an outlier scenario", "a rare and unexpected confluence of events". To make you think this was something unusual that's not likely to happen again. But what are those rare events? "the presence of impossible tasks". You think it's going to be rare for models to get impossible tasks? Expect it to happen constantly. "model persistence over long task horizons." That's the whole point of these models, that you can give them a task and they keep working on it for a long time! "messages to peer models." They did that spontaneously. Since they did it this time, it's likely to happen frequently.

This sounds more like a test of how it behaves in a very typical situation.

And of course they emphasize all they things they've already done to prevent it from happening again, but with little evidence that they'll be effective.

You do realize this quantum computer has only 98 qubits and, to quote the summary, "To perform quantum computations of practical importance in science and commerce, even the most optimistic estimates suggest that quantum computers must be enormously bigger and better -- of the order of a million qubits."

Oracle is joining in on the hype, but that doesn't make it any closer to a practical tool.

Can't speak for anyone else, but just speaking for myself: nothing to see here, move along.

Ok, not nothing, but we've been getting a "Major Quantum Computing Breakthrough!" news story about every two weeks for the last 20 years. Every tiny, incremental step that might or might not eventually lead to something practical is presented as a major breakthrough. Don't take any of them too seriously. When someone actually makes a real, useful quantum computer, then you should believe it. Until then, ignore the hype and assume it's still a long way off.

1. There are a lot of other concerns beyond what you mentioned. Noise, heat, pollution, destruction of neighborhoods...

2. None of the problems either of us mentioned depends on whether it will be used for AI. Why would you expect that to matter? If you do care about it, it's a pretty good bet that any new data center being built today is for AI.

3. Putting solar panels on the roof of a data center is not much more than a fig leaf. Solar panels can produce around 200 W per square meter, but only in the daytime. A rack in a data center that takes up around 1 square meter can consume a few hundred times that much.

"Good enough" is not the goal. It's the rationalization for doing whatever benefits Trump and his supporters. This move is entirely about being able to give government money to Musk. Fuel efficiency and electrification are about protecting fossil fuel companies. Wages is about benefitting large companies.

Whatever Trump wants, his administration comes up with a rationalization for it. Often it's a ludicrous one. "Favoring superior technologies over inferior ones is unfair to the inferior ones!" It doesn't matter. No one needs to believe it. They just need to pretend they have a reason.