m0xEE @m0xee@librem.one

in a hypothetical where we could figure out a non-ecocidal machine learning language model:

there's one ability that this technology has that's genuinely novel, and it's human language syntax. you can't trust the semantics or paradigmatics; unlike humans, these machines have no world model and no consciousness, so they're incapable of understanding and their output can't ever be trusted. however, human syntax is a much harder problem than most people give credit for, and it is impressive that you can hold a conversation with these elizas and the syntax is, to a large degree, turing-passable.

what is a practical use for a chatterbox that speaks nonsense but in linguistic syntax? one thing that comes to mind is language learning. now, current LLMs are of limited usefulness for this use because their conversation isn't just nonsensical, it's also exceptionally bland and the opposite of engaging. (besides, you know, the fact that you have to harm and exploit thousands to even use them). tools like character.ai seem to be an attempt at improving that, but due to limitations of the datasets and of corporatespeak limiters, every character sounds the same.

if we had a functional linguistic model that could be trained and run in a standard laptop with no more input than a few books, maybe we could produce chatterboxes actually able to sound like characters in a novel. and due to people's attachment to these characters, maybe talking to them would feel engaging enough that you would want to do it, even despite the absence of a world model or persistent context. then the unsolvable hallucination problem wouldn't matter; all that you need is to produce interactions that are 1) syntactically correct, 2) intelligible and 3) engaging, and you'd have a language acquisition app that (by contrast with flashcards, SRS etc.) would be actually useful.

but given the current requirements to produce and use an LLM, the most economical and practical language interaction device remains "another human".

@Hyolobrika @munir
Your reply set my mind off wandering and this is what I came to: you know what vacuum cleaner is called in Russian? Пылесос! Its literal translation is "dust sucker" — and it's one of the rare occasions when Russian is very on point: this thing isn't about cleaning and has little to do with vacuum — it just sucks!
In our childhood the authorities (mom) have often made us clean our rooms and made us interact with the machine for that. Some of us grew up and adopter the behaviour — doing that to their own children, turning that into a tradition. For how long are we going to live under the boot heel of the machine? We have to fight back!
But there is more! Ever wondered why the so-called vacuum cleaners were the first to adopt the so-called AI? Some say it's a mere coincidence or even technology well applied, but I don't think it is! Oppression used to be just forced labour, but in the XXI century it's surveillance! And the machine follows that to the letter! Now they spy on us and report to the corporate overlords: https://www.technologyreview.com/2022/12/19/1065306/roomba-irobot-robot-vacuums-artificial-intelligence-training-data-privacy/
What is the so-called "vacuum cleaner" called in your language? I want brave people from all around the world to start speaking out against our pneumatic oppressors that became a symbol of tyranny! We have to take our communities back!

If you're looking to host your very own single-user/a-few-users #fediverse instance, you cannot go wrong with #snac. It is simple to install on Ubuntu and works very well with some really solid clients. It is also written in C, so it is fast, with few dependencies. Great work @grunfink@comam.es, you've got a new monthly supporter on Ko-fi!

#snac2 #ActivityPub

https://codeberg.org/grunfink/snac2

If you're doing a lot of work in C/C++/Rust consider using sccache to cache compilations. It's easy to set up and will save you a lot of time and a huge amount of power.

https://github.com/mozilla/sccache

As an example this is just a couple of hours of work on Firefox, it saved me from re-running the compiler 90% of the time.

#CPlusPlus #RustLang

Oh wow, Maxim Kononenko, aka Mr. Parker, has reportedly died. A talented humorist who once wrote a satirical blog all about Putin, he was also a rabidly patriotic, chauvinist z-monster who penned propaganda essays for the last decade and change. https://t.me/rian_ru/245509

Nowa Huta in Kraków.

Built from the ground-up as a separate city to settle workers employed in the newly built steelworks (it's even in the name, which literally means 'the new steelworks'). Designed as a poster city for Soviet propagandists. Huge, industrial, pompous. These days it's one of the most populous parts of Kraków, even though it completely didn't feel that way during my long weekend visit.

I barely scratched the surface, as severe planning mistakes were made on my end. Plus I don't like taking cliche postcard photos in harsh light. But I managed to see three different faces of this place and that counts I guess.

We start with the socrealistic face. The one that dates back to 1950s and truly feels different than whatever my peanut Warsaw brain assumed Kraków to be.

#photography #cracow #poland #kraków #polska #architecture #travel

Мы все родом из детства. В нем всегда была сила и юмор. Теперь же мы стали очень грустными.

bbs.geminispace.org was launched exactly one year ago today. I wrote up a little retrospective to honor the occasion and review what has been happening on this Gemini-based bulletin board over the past year:

gemini://bbs.geminispace.org/u/skyjake/16845

Musk claimed that Signal has unpatched security vulnerabilities. As a cryptography expert I’m not aware of any: instead Signal has top-notch state-of-the-art encryption which has been adopted by its competitors it’s so good.

Telegram isn’t end-to-end encrypted by default. It has no group chat encryption whatsoever. It uses a bizarre, badly designed “MTProto” protocol about which dozens of cryptography papers have documented myriad unfixed flaws.

one month ago

Soviet actor Georgi Burkov in his memoirs first released in 1998 (found by Maxim Mirovich):

> The Bolsheviks need the atomic bomb to stay in power, i.e. against their own people. It won't be long before they start blackmailing the whole world, so that the whole world begs us, the people, not to make a revolution and to tolerate these ghouls and support them

Very much to the point both then and now.

Rest in power, Steve Albini
https://www.theguardian.com/music/article/2024/may/08/steve-albini-us-alt-rock-musician-and-producer-dies-aged-61