tlb

Filters catch particulates, but not gases. Heated ABS releases chemicals like 1,3-Butadiene, a small molecule (4 carbons), gaseous at room temperature, and mildly gasoline-smelling. It's bad for you in high concentrations.

Bales of straw are a cheap, large, soft object you could always find nearby in the pre-motor car days. I can't think of a better object to require.

Yes, if every government was reliably incorruptible, they could also work together to build a global human verification network.

I predict that Worldcoin will get it done first, and will be more dependable than most countries. But it could turn out otherwise. In the end, services that need humanity verification will have multiple provider options and the market will decide.

There are two levels of this. If I get some other European language, I can generally figure out which is their word for English and it's not a big deal to switch. But if it gives me a script I don't know, like Bengali or something, it's a problem.

Perhaps every "choose language" menu should include English and Chinese in non-localized form, as an escape hatch, since almost every web users can recognize enough of them to navigate a menu to find their actual language.

Yes, but the quietness outweighs all other considerations.

The 128- and 256-core ARM server chips (like from Ampere) are pushing server performance in interesting ways. They're economically viable now for trivially parallelizable things like web servers, but possibly game-changing if your problem can put that many general-purpose cores to work.

The thing is, there aren't that many HPC applications for that level of parallelism that aren't better served by GPUs.

That's about the right level of ceremony to request an uninitialized variable.

It has a 10 second 0-50 mph section, which for the kinds of cars we're talking about (those that people might buy aftermarket performance ECUs for) isn't close to full throttle.

Emissions are tested on a standardized driving cycle that doesn't include full-throttle acceleration, so emissions at full throttle aren't counted. Which is reasonable: high-performance cars spend only a tiny fraction of their time at full throttle so it doesn't contribute much to overall emissions.

The best way to lock out tuners would be for the stock ECU to provide as much power as possible.

Although with turbocharged engines, you can turn the power up to the point where longevity goes way down. Still, you could give people a power limit knob they could turn at their own risk to have access to the whole Pareto frontier of performance and longevity. That'd leave no room for tuners.

It's also not very hard for scammers to get a SIM card for your number, unless you're using a carrier that specializes in not allowing SIM swapping attacks.

I wonder what the companies requiring 2FA think about uncompleted 2FA bounces. Deterred fraudster? Short attention span? SMS sucks?

Being initialized to zero is at least repeatable, so if you forget to initialize something you'll notice it immediately in testing. The worst part about uninitialized variables is that they frequently are zero and things seem to work until you change something else that previously happened to use the same memory.

LLM coding assistants are a partial solution. Recently I typed

  vec4 rgb2hsv(vec4 rbg)

and a few tab-completes later it had filled in the body of the code with a correct color conversion routine. So that saved me searching for and pulling in some big-ass color library.

Most of lodash.js can be avoided with LLMs too. Lodash's loops are easier to remember than Javascript's syntax, but if your LLM just writes the foo.forEach((value, key) => {...}) for you, you can skip the syntactic sugar library.

Worse than the small amount is the length of time it took. If you accept the accusations at face value, China has been spying on a large fraction of EU citizens for 4 years and can keep doing it for another 6 months (after which they probably won't actually stop) for the eventual fine of a few dollars per victim. The US isn't moving any faster, and most other countries aren't moving at all.

So the end result is that potentially hostile countries can run vast spying operations for a long time with no major consequences. As long as they do it with funny videos with annoying soundtracks.

> the serialised version of the data is usually bigger than its in-memory representation

I don’t think this is common. Perhaps for arrays of floats serialized as JSON or something. But I can’t think of a case where binary serialization is bigger. Data types like maps are necessarily larger in memory to support fast lookup and mutability.

How do you deal with instructions that iterate through variable amounts of memory, like concatenating strings? Are such instructions interruptible?

Perhaps they don't need to be interruptible if there's no virtual memory.

How does it allocate memory? Malloc and free are pretty complex to do in hardware.

Robot walking. But really, a programming language and tools to make it easy to solve hard nonlinear control problems like walking. There are videos & source code for 2-legged walking robots (in simulation) at https://throbol.com/post/how-to-walk, and videos & source code for 4-legged walking robots (on real hardware) at https://throbol.com/

They are about as ethical as non-monopoly businesses. Which is to say that some are bad and most are OK.

I guess some people think all business is unethical, but for those people monopolies don't change the sign.

By itself, becoming a monopoly can be legal & ethical.

But once you have a monopoly, you're not allowed to do certain things like buying competitors for the purpose of crushing them.

It wasn't the intended purpose. It turned out that way because the Detroit lobbyists were smarter and more motivated than the government policy people, and they bamboozled them.

It's more that /etc/hosts and /etc/passwd are good for testing because they always exist with predictable contents on almost every system. If you inject "cat /etc/passwd" to various URLs you can grep for "root:" to see if it worked.

So it's really blocking doorknob-twisting scripts.

You don't need an external trust store. Just have IT add and verify the relevant contacts, being careful not to include any journalists with similar names.

DOD's IT department should be in charge of making sure everyone in the employee directory has their credentials properly installed on everyone's phone. You don't need to fork it, just administer it properly.

> Fix for range-based for loops

Oh man, having different compilers in c++20 mode handle things differently is going to cause more grief, not less.

Reminder: Prior to c++23 the following is broken:

  vector<int> const &identity(vector<int> const &a) { return a; }

  for (auto a : identity(vector<int>{1,2})) { ... }

That's because the lifetime of the vector isn't extended through the life of the for loop. That is, the vector is destructed right after identity returns, and the for loop ends up trying to iterate through a vector that's been destructed.

But now gcc in c++20 with -frange-for-ext-temps mode will extend the lifetime of the vector and the above code will work, and people will write code like that, and it'll break mysteriously on other c++20 compilers. The usual way it breaks is that the for loop does nothing because in destructing the vector it sets the begin and end pointers to null, so it's a subtle kind of breakage.

BTW clang with -Wall doesn't complain about the above broken code.

Signal (or something like it) probably should supported by the DOD for high-level communication among defense & administration people. Under the traditional rules, certain topics can only be discussed in SCIFs, which are a huge pain in the butt to arrange a meeting in, especially as politicians travel a lot. So military topics end up disconnected from politics.

(Maybe the military likes being disconnected from politics, but that's not the setup that political philosophers recommend to preserve democracy.)

If the DOD managed dedicated phones with no apps except Signal, that might be better than whatever they do between SCIFs.

There are plenty of articles online (and surely in OpenAI's training set) on this topic, like https://earthsky.org/space/planet-nine-orbit-map/.

Answer quality is a fair test of regurgitation and whether it's trained on serious articles or the Daily Mail clickbait rewrite. But it's not a good test of reasoning.

It's far too soon to call any AI-specific ideas tarpits. Nothing newly made possible by LLMs or generative image models has been tried long enough to give up on. There are no startup bones to dig up yet.

"Are you not impressed?"

Empires should spend part of their budget on impressing. The English lad who spent whole days schlepping a few bushels of turnips to the market would be thinking, "I should join the Empire and transport lions".

* 714 ps