choeger

It's a long-used term for improvements in manufacturing due to increased scale.

Code generation should be relatively straightforward with language model. Start with a problem description and let the LLM generate the code. It should be fairly simple to ask for a proof of correctness for the generated code, then verify that proof (for instance by creating property-based tests), and finally ask for a human-readable summary of the proof. Pass this to human code review and you have a pretty good code generation machine. Failure at any step can simply lead to another iteration.

The thing is: SE doesn't work like this anymore. People aren't used to formalized requirements and take 99% just for granted. (Which web developer really takes network errors into account nowadays?).

With "AI" we might have the tools we need to really solve the software crisis.

It should be obvious that the linear no-threshold model of radiation danger is so weak that other findings are possible. After all, the human body has some repair mechanisms. Vitamin D is just the most well-known positive effect of sunlight.

Considering that when the launch order came, MAD had probably already failed: Would you have obeyed the order and killed a few million people?

From the outside: Diversification is always great. Build a whole ecosystem of small-scale rocket manufacturers instead of one big monopoly. That will foster competition and innovation.

From the investor: SpaceX might fail. Even if there Falcons are pretty much unbeatable now, you don't know what's going to happen with Starship. And even the Falcons could conceivably be grounded for years after some hypothetical flaw is found. More likely: With the price reductions made by SpaceX, the market will grow and there will be more than enough clients.

From the inside: Because it's a fun challenge and literally rocket science, of course.

The thing is: C++ ain't a language. It's two, if not three languages mashed together. You start with the C preprocessor, then you have the template language and only then you find classes, objects, pointers, etc.

That's like learning Rust but having to learn POSIX sh, and some badly documented Haskell dialect on top of it just to get going.

I have a hunch that by "rendering" HTML and SVG, they don't mean what I think the word means.

Does this library really render HTML and SVG in the sense that it creates a bitmap from the document?

Great comment. Just one thought: Language, unlike weather, is meta-circular. All we know about specific words or sentences is again encoded in words and sentences. So the embedding encodes a subset of human knowledge.

Hence, a LLM is predicting not only language but language with some sort of meaning.

Funny thing is that we already live in a world where theoretically no one should be accountable for what "their" devices are doing. How the eff am I supposed to know which messages my phone is sending? From hardware vendors over operating system vendors to messaging app vendors and potentially even the mobile network provider, there's a shitton of people that could be messing around with my phone already. The price for a targeted attack might still be significant, but it's not prohibitive for many scenarios.

So why would the content even be interesting for any law-enforcement agency?

This. I wonder why everyone in the security apparatus acts as if they will never be affected by these kind of requests. I can only imagine two possible answers:

1. They trust in legal exceptions for themselves (look at EU chat control exceptions), which implies a thinking closer to the Soviet Union than a real republic. 2. They think they're already compromised by "the Chinese" or any other state actor and just feel incompetent themselves.

Wait a second. Does that mean every DNS request on a Fritz Box network is send to fritz.box first?

I query google.com and get google.com.fritz.box every fucking time? Shit.

Came here for that comment.

The thing is: Growth is the driver for increased efficiency and that is the driver for any kind of progress.

Progress does not just happen out of thin air when a bold mind in their ivory tower has a great insight. Progress requires (wo-)manpower. And that power in turn needs to be freed from more mundane tasks.

In other words: We, as a society, don't want Apple to become ever bigger and make ever bigger profits. We want them to continue making MacBooks and to need fewer and fewer people to make them. Because we want these people to cure cancer or drive spaceships to the outer system or develop artificial meat.

Unfortunately, this simply won't happen if we take out the unsatisfiable desire for more, more, more.

So the attacker put the payload into the tests folder of the compression library and injected it upon build via some arcane build script? Kudos. That's quite a smart way to hide your backdoor.

Nevertheless, they got caught. So big respect for the analysis. Amazing work.

That being said, what does this attack tell you about the future of software security? Free software could easily be compromised by having a highly paid team create such a payload and then convince some maintainer, one way or another, to deploy it.

How do we back off from that cliff?

In a eagerly evaluating functional language you could always wrap the arguments in a thunk:

So if your built-in functions were implemented in Python, you'd use:

  def if_(cond, t, e):
    if cond:
      return t()
    else:
      return e()

"creativity", my ass. 99% of people aren't hired to be creative. They're hired to be quick and efficient.

After over three years WFH now, I know that my performance has improved. Anyone asking me to move around town every day will simply have to pay for that additional task now.

> The team fine-tuned the network by training it on thousands of experimentally determined structures of antibodies attached to their targets

So that should explain the relatively low yield.

To avoid the word "AI", the researchers took a thought-to-be universal model and ran a parameter estimation for a very complex 3D (bio-)chemical interaction model.

Of course this approach will work with enough samples and a model that is universal enough to express the problem solution. But they simply had way fewer samples than, say, a large language model.

I think this approach won't carry that far because of this lack of training data. It might give us some exciting results, but as long as manufacturing of antibodies and testing them takes a (comparatively) long time and huge effort, the model will be much less efficient than a classical algorithmic simulation. But of course, there's a chance that it provides insights into improving such "boring" models.

Good post. Good analysis. One problem:

If turing-completeness is necessary, why should there even exist a build system independent from a programming language now?

Isn't the obvious solution to script your language's default build system in precisely the language you're trying to build from?

So when you build Rig, build it in C++ and make C++ your scripting ... Ok, that left a bad taste in my mouth.

Seriously, though. Build systems as the author describes them are going to compete against the project managers that come with languages (Cargo, Go). The only real use case for Rig would be legacy languages that don't come with language-central packaging repositories, i.e., Fortran, C++, C.

While I applaud the author for the attempt to improve this ecosystem, I really hope it will become less relevant some day.

Important note: Right-wing propaganda usually is built upon a small (sometimes tiny) piece of truth.

Examples?

January 6th was no bloodbath in the literal sense. There simply wasn't that much blood. It was a deeply undemocratic, treacherous act of insurrection, but compared to other events in history, the term "bloodbath" is indeed inaccurate.

Google definitely has a woke, left-wing bias: Case in point, their latest AI model generating the dark-skinned Wehrmacht soldiers. Does it mean the search results are biased, too, though? We simply won't know.

And of course, there is an anti-Trump bias in "the media" (which at this point means anything that's not pushing out non-stop right-wing propaganda). A well-deserved one at that.

Now here's an application for "AI": have a ultrasound device the size of a medical tricorder from Star Trek in every household with children. If child gets one of the dozens of inevitable infections, use the device to check for the standard complications (otitis media, pneumonia ...) and allow a physician to replace 95% or more of the visits with a remote diagnostic.

Monomorphization.

For every generic function f, rustc will generate as many instances as there are type instances (shape instances? Does the Compiler distinguish between different kinds of references that all get compiled to pointers?).

This feature has a cost. Compare to OCaml's uniform object representation that enables comparatively blazing compilation performance but pays a prize in performance and weird FFI restrictions (integers with a tag bit).

Btw. It's misleading to say "it's the backend" when the frontend is responsible for creating so much work for it.

I think these numbers aren't comparable because they're normalized by employees, not employable population.

If you compare a traditional mom-stays-at-home family with a more typical German family they both might work 50-60 hours a week. The difference is that in the former case it's a male burning the midnight fuel in an office or working a side job and in the latter it's two parents working a 20-40 or 30-30 split or so.

So I think the difference would look much differently if you'd account for that fact.

If one considers how desperately Microsoft tries to squeeze additional revenue from Windows (forced cloud account, ads, browser war 2.0) one has to wonder at what point they'll simply shut down OS (i.e., kernel and plumbing-level user-space tools) development completely.

I mean yes, it has been a proud piece of Microsoft but the OG NT kernel team probably left by now and from a management perspective it doesn't offer any revenue to develop an OS kernel.

Otoh, how big is their OS engineering team? Drivers are usually developed elsewhere. So maybe 100 folks or so? With a relatively low quota of non-devs, I assume? So Microsoft would potentially save what, tens of millions of dollars per year? Maybe they just continue this as some kind of tradition department.

Came here with exactly this on my mind. Thanks for confirming my suspicion.

That being said, having specific requirements for the environment of your integration tests is not necessarily bad IMO. It's just a question of checking these requirement and reporting any mismatches.

It could be a consequence of racism. Racism lead to white people dominating the United States, which lead to images of white people dominating the internet, which lead to images of white people dominating the training data for facial recognition algorithms.

But it could also just reflect the composition of the population of the countries that dominate the internet.

I am not aware of anything that works today, but I think that there's room for shopping agents. Say you need a new USB Stick or a pair of shoes. Something between $10 and $1000 that you simply have to buy ASAP but doesn't warrant spending one or more evenings on research. A language model could sift through the descriptions and comments and try to eliminate trash and even outright fraud.

But then again, it's just another search engine, essentially. So for how long would it stay useful before it accepts payments to promote certain offers?

Sulfur is probably not the best agent to do it, but we know that it works from the unintentional ship fuel experiment.

In any case, if 2023 wasn't an exception, I'd rather have someone start a program this year. It can always be stopped or scaled down. And I am confident that renewable energy, electrification, and battery deployment will continue unabated anyways.

A hacker news post to a stack exchange question copied from reddit? At some point we might simply fuse these networks into one, don't we?

We have 500Wh/kg batteries by now?

And they think there'll be 800Wh/kg in 2030? Wasn't that well beyond what's needed for medium-range electric flight?

Is that even possible, chemically speaking?

Very good questions.

Here are a few pointers:

Podman runs rootless and avoids (some of) the problems with permissions this way.

It's possible to mount the source directory (when running a container) over the place you copy it to (when building it) so you can start a container once and rebuild and test inside it while you edit outside of it.

I think that containers are a good reason to make a technical distinction between unit tests and integration tests. The former should work outside the container to facilitate quick development whereas the latter can rely on the environment the container provides. That setup saves a lot of headache for configuring paths and dependencies.

Finally, i find it very important that building the software and executing the unit tests should be possible outside the container. This way you can always use your local setup, maybe after some tweaking. This tweaking is the (small) price everyone has to pay every now and then. That way the build environment doesn't run stale. Imagine developing software with a frozen tool stack packed into a container ten years ago. Because that's what happens when everyone just uses the image.