#Fedle #634 2/6

🟩🟨⬛⬛⬛⬛
🟩🟩🟩🟩🟩🟩

https://fedle.fedilab.app/share/14995

@patrickhadfield

This tried to be journalism but is written like marketing for the LLM vendors.

LLM’s do not ‘go rogue’. They do not ‘ignore instructions’. They produce a sequence of tokens that, according to their training data, have a high probability of following the ones that they user provided. That’s it.

‘Agentic’ systems take the stream of output tokens and feed that into some other system. That other system should be expected to do any of the things that it has the credentials to do, because it is effectively being driven by a random number generator. It’s a biased random number generator that is more likely to generate numbers that make it do things that you want the other systems to do than anything else.

Talking about it ‘ignoring instructions’ implies that it understands instructions. This is misleading.

Talking about it ‘going rogue’ implies that it has a set of well-defined behaviours and can somehow leave that space.

And these are exactly the kinds of misleading statements that LLM vendors want people to believe because that’s the core of their sales narrative.

@John_Livingston 🤣

Still early to say but Florian seems to be finding the way forward. Still inexplicable why we have Gakpo on the right.

#LFC

@corpsmoderne
(Finkelkraut ? 🤞)

@Tattooed_Mummy it's like "can nuclear be good?"

And while sure, nuclear medicine can be really useful to get amazing visualisation, that would be impossible to otherwise obtain.
Nuclear is really bombs, and a plague on areas with nuclear plants.

So on the whole, and since you cannot have the benefits without the large problems, I would say no to the whole lot (nuclear and AI).

GLM-5.3-Flash: Frontier Intelligence, Flash Cost (e chip cinesi) https://blog.quintarelli.it/2026/08/glm-5-3-flash-frontier-intelligence-flash-cost-e-chip-cinesi/

"Especially important is the warning to avoid conversations with the demon. We may ask what is relevant but anything beyond that is dangerous. He is a liar. The demon is a liar. He will lie to confuse us. But he will also mix lies with the truth to attack us. The attack is psychological, Damien, and powerful. So don't listen to him. Remember that — do not listen."

- Father Merrin providing user instructions for ChatGPT in the film The Exorcist, 1973

RE: https://eldritch.cafe/@temptoetiam/117178397020937317

« Lui affirme avoir recueilli son consentement » et plus loin l’intéressé : « Je suis venue chasser les femmes, les proies (…) C’est mieux qu’elle voie pas qu’elle est filmée » 😰🤮
Au gnouf tous ces pervers, bordayl !

TinyMUSH

A TinyMUD-derived softcode server whose current line modernizes the long-running TinyMUSH implementation.

Current releases use #LMDB and provide tools for legacy GDBM databases

https://www.mudverse.com/mud-engines

@corpsmoderne
C'est p'tetre un immortel qui viens de se faire décapiter ?

@matt @TheBreadmonkey I think the Seagulls would quibble about the word "Lose", they may consider it a rightful donation.

No hay suficientes cabritillas en MastodonLand
😍🤩🥰

@emenel @zzt that is such a bizarre inconsistency. I'm seeing people on Twitter accuse others of being "bigoted" against genAI, while happily insulting, pigeonholing and accusing those detractors of being NPCs. Like, what do they actually care abou?

Ah, I get the Meta $17B thing now. Big tech doesn't like EU regulations because they are, well, from the EU and have to be followed once implemented. They much prefer to pay a bit of money to be allowed to write the rules themselves. Because that's what the US deal effectively is. Meta pays $1.2B per year in exchange for being left alone and come up with some soft rules that they get to implement and control themselves.

I hesitate to say this bluntly, because I know a whole lot of folks on here are actively involved in technical development of one sort or another, and derive both pleasure and a significant part of their identity from that fact. There’s something I think we all need to get clear on, though, and it’s something I myself had to learn the hard way: no tool will liberate us. The achievement of justice cannot be automated, or otherwise outsourced to a technical apparatus of any sort.

You work on tools because that’s what you’re good at, because that’s the way your brain works, because that’s what you enjoy or can get compensated for doing. That’s all as may be. But please, please do not make the mistake of thinking that the tool – any tool, of any sort whatever – will produce “progress.” There is and can be no technical fix for the social-psychological, existential or ontological predicaments that confront us. In the end there is only us, staring back at us from whatever mirror we’ve built.

Disappointed that #Debian has chosen what seems like the worst possible option in its vote on LLMs.

No acknowledgement of the harms involved in LLM usage; instead, a claim that they ‘substantially improve productivity’. (It claims not to be an endorsement but makes only pro-LLM statements.)

Vague platitudes about using LLMs ‘responsibly’ when it’s become very clear by now that LLM users can’t be trusted to be responsible without constant supervision. And no requirement for disclosure of LLM usage, even.

It didn’t seem likely that the strongest anti-LLM proposals would pass, but it would have been nice to have at least some restrictions put into place.

@neil the only "responsible" use of plagiarism generators is no use