comicjk

"The way that Hugging Face defended itself was with an open source Chinese model, GLM, because they didn’t have access to the high cybersecurity models on the other side. They’re too dangerous."

This is not what happened. HuggingFace patched the vulnerabilities in the ordinary way, then used GLM for some of the forensic analysis afterwards. It's in the interest of HF to promote the GLM part of the story, because it emphasizes their agency in the situation as well as the open-source models they distribute. But GLM was a (useful) part of the postmortem, not of the defense.

My labor makes other people's lives better, so I would expect something that replaces my labor to do the same.

Watermarking will not flag something you wrote unless the AI rewrote significant chunks of it. AI watermarking works by exploiting the fact that lengthy phrases can be expressed in exponentially many ways, such that the selection of a single sequence from the exponential space is practically unique. For proofreading by contrast, if the AI is only changing isolated words in work that's otherwise yours, there are not enough exponentially branching options for the watermark to distinguish anything.

*Some might see a parallel with the old game Adventure, in which wording differences like "twisty little passages" and "little twisty passages" were used to build a maze of room descriptions, with the same meaning but still distinguishable to the attentive player.

Uranium-235 isn't useful for dirty bombs: it's too long-lived and not radioactive enough. Making dangerous levels of radiation requires short-lived isotopes. Some of these are the products of uranium fission, which is why spent uranium fuel is a radiation hazard. But isolating U-235 just to fission it to get radioactive isotopes like cesium-137 would not be the easiest way to get them.

I did this for the last college programming course I taught (2025) as an anti-LLM measure. Every class started with a very short closed-book auto-graded multiple-choice test on the reading. I was careful to specify which was the core reading needed to learn the material (on the quiz) and which was extra (not on the quiz - I think many professors assign this extra reading to everyone, thinking of their strongest students). I think this developed an expectation in the students that doing the reading was easier than finding a way not to.

Historically, this is the norm. The British Navy in WWI was a behemoth, but still couldn't force safe passage through the mines and cannons of the Dardanelles Strait against what was considered a third-rate power. "A ship's a fool to fight a fort."

I have bad news about reading the internet...

It's worth remembering that "fusion = abundant power" is a guess about technology that hasn't been finished yet. Fusion power might turn out like solar panels (easier to build than expected) or like nuclear fission reactors (harder to build than expected).

A Fire Upon The Deep is a fantastic novel for programmers to read, and I think the prequel A Deepness In The Sky is even better. There are some amazing old-school coding jokes in there, like that everyone thinks the universal time counter started at the first moon landing, but programmer archaeologists know it was really 15 megaseconds later.

People always need to be reminded, though. It seems to be in human nature to fear bad publicity, and the people who fear it less end up with disproportionate power as a result.

Not everything on Wikipedia is true, but the parts Elon Musk hates most are probably true.

This context is very important.

"Little by little, over-inflated results and breathless breakthroughs betray trust. They throwing dimes in a wishing well which people rapidly start to expect will never pay compound interest."

"Then, when one of those people is elected to parliament, or Congress, and start to cut the budget for the National Science Foundation, or declares that All Research Should Be In The National Interest (whatever that is), I wonder how much we reap what we have sown."

I gave my engineering students a CO2 removal design problem once, and at the end, asked why the theoretical efficiency had increased in the time since the textbook was written. The answer was that the concentration of CO2 in the atmosphere was higher.

It is tough to beat classical computers. They work really well, and a huge amount of time (including some of mine) has gone into developing fast algorithms for them to do things they're not naturally fast at, such as quantum chemistry.