Ah still rolling out the old “stochastic parrot” nonsense I see.

It is a bunch of stochastic parrots. It just happens frequently that the words they are parroting were orginally written by a bunch of intelligent people which were knowledgeable in their fields.

Note this does makes the parrots intelligent - in the same way that a book written by Einstein to explain special relativity has any own intelligence. Einstein was intelligent, his words transport his intelligent ideas, but the book conveying them to other people (as, the printed pages with cardboard cover) is as dumb as a stone. You would not ask a piece of cardboard so solve a math problem, would you?

Reponding to another comment in [email protected]:

Writing code is itself a process of scientific exploration; you think about what will happen, and then you test it, from different angles, to confirm or falsify your assumptions.

What you confuse here is doing something that can benefit from applying logical thinking with doing science. For exanple, mathematical arithmetic is part of math and math is science. But summing numbers is not necessarily doing science. And if you roll, say, octal dice to see if the result happens to match an addition task, it is certainly not doing science, and no, the dice still can’t think logically and certainly don’t do math even if the result sometimes happens to be correct.

For the dynamic vs static typing debate, see the article by Dan Luu:

https://danluu.com/empirical-pl/

But this is not the central point of the above blog post. The central point of it is that, by the very nature of LLMs to produce statistically plausible output, self-experimenting with them subjects one to very strong psychological biases because of the Barnum effect and therefore it is, first, not even possible to assess their usefulness for programming by self-experimentation(!) , and second, it is even harmful because these effects lead to self-reinforcing and harmful beliefs.

And the quibbling about what “thinking” means is just showing that the arguments pro-AI has degraded into a debate about belief - the argument has become “but it seems to be thinking to me” even if it is technically not possible and also not in reality observed that LLMs apply logical rules, cannot derive logical facts, can not explain output by reasoning , are not aware about what they ‘know’ and don’t ‘know’, or can not optimize decisions for multiple complex and sometimes contradictory objectives (which is absolutely critical to any sane software architecture).

What would be needed here are objective controlled experiments whether developers equipped with LLMs can produce working and maintainable code any faster than ones not using them.

And the very likely result is that the code which they produce using LLMs is never better than the code they write themselves.

Writing code is itself a process of scientific exploration; you think about what will happen, and then you test it, from different angles, to confirm or falsify your assumptions.

What you confuse here is doing something that can benefit from applying logical thinking with doing science. For exanple, mathematical arithmetic is part of math and math is science. But summing numbers is not necessarily doing science. And if you roll, say, octal dice to see if the result happens to match an addition task, it is certainly not doing science, and no, the dice still can’t think logically and certainly don’t do math even if the result sometimes happens to be correct.

For the dynamic vs static typing debate, see the article by Dan Luu:

https://danluu.com/empirical-pl/

But this is not the central point of the above blog post. The central point of it is that, by the very nature of LKMs to produce statistically plausible output, self-experimenting with them subjects one to very strong psychological biases because of the Barnum effect and therefore it is, first, not even possible to assess their usefulness for programming by self-exoerimentation(!) , and second, it is even harmful because these effects lead to self-reinforcing and harmful beliefs.

And the quibbling about what “thinking” means is just showing that the arguments pro-AI has degraded into a debate about belief - the argument has become “but it seems to be thinking to me” even if it is technically not possible and also not in reality observed that LLMs apply logical rules, cannot derive logical facts, can not explain output by reasoning , are not aware about what they ‘know’ and don’t ‘know’, or can not optimize decisions to multiple complex and sometimes contradictory objectives (which is absolutely critical to sny sane software architecture).

What would be needed here are objective controlled experiments whether developers equipped with LLMs can produce working and maintainable code any faster than ones not using them.

And the very likely result is that the code which they produce using LLMs is never better than the code they write themselves.

Are you saying that it is not possible to use scientific methods to systematically and objectively compare programming tools and methods?

Of course it is possible, in the same way as it can be inbestigated whuch methods are most effective in teaching reading, or whether brushing teeth is good to prevent caries.

And the latter has been done for comparing for example statically vs dynamically typed languages. Only that the result there is so far that there is no conclusive advantage.

What called my attention is that assessments of AI are becoming polarized and somewhat a matter of belief.

Some people firmly believe LLMs are helpful. But programming is a logical task and LLMs can’t think - only generate statistically plausible patterns.

The author of the article explains that this creates the same psychological hazards like astrology or tarot cards, psychological traps that have been exploited by psychics for centuries - and even very intelligent people can fall prey to these.

Finally what should cause alarm is that on top that LLMs can’t think, but people behave as if they do, there is no objective scientifically sound examination whether AI models can create any working software faster. Given that there are multi-billion dollar investments, and there was more than enough time to carry through controlled experiments, this should raise loud alarm bells.

So what do you do with a file object?

You are right with this. But still, in Rust, a vector of u8 is different from a sequence of unicode characters. This would not work in Python3 either, while it’d work in Python2.

Thanks, I fixed it!

My experience (from using Linux since 1998) is that the best way to use Linux is to get compatible hardware (that is, unless you want to develop device drivers). And this doubly and triple for laptops and graphics cards. Refurbished business Thinkpads are a very good option.

What I find interesting is that move semantics silently add something to C++ that did not exist before: invalid objects.

Before, if you created an object, you could design it so that it kept all invariants until it was destroyed. I’d even argue that it is the true core of OOP that you get data structures with guaranteed invariants - a vector or hash map or binary heap never ceases to guarantee its invariants.

But now, you can construct complex objects and then move their data away with std::move() .

What happens with the invariants of these objects?

let mut bytes = vec![0u8; len as usize];
    buf.read_exact(&mut bytes)?;

// Sanitize control characters
let sanitized_bytes: Vec<u8> = bytes.into_iter()
    .filter(|&b| b >= 32 || b == 9 || b == 10 || b == 13) // Allow space, tab, newline, carriage return
    .collect();

This implicitly, and wrongly, swaps the interpretation of the input from UTF8 text to pure ASCII.

Did you ever note that when intelligent engineers talk about designs (or quite generally when intelligent people talk about consequential decisions they took), they talk about their goals, about the alternatives they had, about what they knew about the properties of these alternatives and how these evaluated with their goals, about which alternatives they chose in the end and how they addressed the inevitable difficulties they encountered?

For me, this is quite a very telling sign of intelligence in individuals. And truly good engineering organizations do collect and treasure that knowledge - it is path-dependent and you cannot quickly and fully reproduce it when it is lost. And more importantly, some fundamental reasons for your decisions and designs might change, and you might have to revise them. Good decisions also have a quality of stability which is that the route taken does not change dramatically when an external factor changes a little.

So and now compare that to when you let automatically plan a route through a dense, complex suburban train network, by using a routing app. The route you get will likely be the fastest one, with the implicit assumption that this is what you of course want - but any small hiccup or delay in the transport network can well make it the slowest option.

Cognitive Debt is where you forgo the thinking in order just to get the answers, but have no real idea of why the answers are what they are.”

So, how many users of Debian would even think about creating own packages?

I already have a hunch what went wrong: they were probably trying to package software that has no standard build system. This is painful because the standard tools, like GNU autotools for C programs, or cmake, or setuptools or its newer siblings for python, make sure that the right commands are used to build a package on whatever platform, and that, importantly, its components are installed into the right places. If they don’t use these, they will have a problem to build packages for any standard distribution.

Guix has support for all the mayor build systems (otherwise, it could not support building of 50000 packages).

So, what exactly were they trying to do?

Yes, Nix solves the same problem. The main difference is that the language used for package descriptions is less attractive to some developers compared to the language which Guix uses, which is Guile Scheme. Guile is very mature, well documented and has good performance.

I think that will give Guix an advantage in the long run, since for a successful disyribution, one needs a bunch of packages and for this, volunteers need to write package definitions and maintain them. Guix makes it easier to write definitions.

Clearly the strict focus on FLOSS will prevent some packages like NVidia drivers from appearing there. But on the other hand, this gives you a system which you will be able to completely compile from source in 10 years time.

Guix is really making fantastic progress and is a good alternative in the space between stable and fully FOSS distributions, likes Debian, and distributions which are more up-to-date, like Arch.

And one interesting thing is that the number of packages is now so large that one can frequently install additional more recent packages on a Debian systems, or ones that are not packaged by Debian.

For example, I run Debian stable as base system, Guix as extra package manager (and Arch in a VM for trying out latest software for programming).

The thing is now Guix often provides more recent packages tham Debian, like many Rust command line tools, where Debian is lagging a bit. There are many interesting ones, and most are recent because Rust is progressing so fast. Using Guix, I can install them without using the language package manager, regardless whether iy is written in Rust, Go, or Python 3.13.

Or, today I read an article about improvements in spaced repetition learning algorithms. It mentioned that the FLOSS software Anki provided it, and I became curious and wanted to have a look at Anki. Well, Debian has no “anki” package - and it is written, among other languages, im Python and Rust, so good luck getting it on Debian stable. But for Guix, I only had to do “guix install anki” and had it installed.

This works a tad slower than apt-get … but it still saves time compared to installing stuff and dependencies manually.

A big part of the changed software job market in the US is caused by the rise of interest rates, and in consequence a large part of high-risk venture capital money drying up. This was finsncing a lot of start-ups without any solid product or business model. And, this began very clearly before the AI hype.

The trope that AI is actually replacing jobs is a lie that AI companies want you to believe.

I don’t get that people constantly complain that the Guix project does not distributes or actively supports distribution of binary, propietary software. That is like complaining that Apple does not sells their Laptop with Linux, Microsoft does not sells Google’s Chromebooks, or that Amazon does not distribute free eBooks from project Gutenberg, ScienceHub or O’Reilly.

And users can of course use the nonguix channel to get their non-free firmware or whatever, but they should not complain and demand that volunteers of other projects do more unpaid work. Instead, they should donate money or volunteer do do it themselves.

But guess what? I think these complaints come to a good part from companies which want to sell their proprietary software. Valve and Steams show that a company can very well sell software for Linux, with mutual benefit, but not by freeloading on volunteer work.

And one more thing, Guix allows to do exactly what Flatpaks etc. promise: Any company, as well as any lonely coder, team of scientists, or small FLOSS project, can build their own packages founded on a stable Guix base system, with libraries and everything, binary or from source, and distribute it from their own website in a company channel - just like any Emacs user can distribute his own, self-written Emacs extensions from a Web page. And thanks to the portability of the Guix package manager, this software can be installed on any Linux system, resting on a fully reproducible base.

A million drunk monkeys on typewriters can write a work of Shakespeare once in a while!

But who wants to pay a 50$ theater ticket in the front seat to see a play written by monkeys?