Wandering Adventure Party

#WritersCoffeeClub 4/2: How much detail do you use to describe your settings?The setting in question is mostly 19th century Germany and surrounding areas - which is as alien to a modern English-language audience as many fantasy worlds.Thus, I have an appendix giving some cultural context, and the tales I translate have numerous additional commentary and footnotes. This is actually a large portion of my work, in addition to the translations themselves. This is also part of the reason why I refuse to use #LLM for translation (beyond the usual ethical objections) - I want the readers to understand the context of the tales, and AI would inevitably get that wrong.

This warms me in the meow meow "How Replacing Developers With AI is Going Horribly Wrong"#AI #LLM https://www.youtube.com/watch?v=ts0nH_pSAdM

So, how _can_ the #Fediverse protect itself against malicious swarms of agentic #AI and #LLM , like this article describes? How can we protect ourselves from bad actors trying to sway public opinion in this manner?We do have an advantage here because such swarms cannot easily manipulate the underlying feed algorithms - most (though not all) of the Fediverse does not have any such algorithms other than a strict timeline feed. But this advantage does not confer immunity.I think we haven't seen many of these attacks on the Fediverse because the Fediverse is ultimately too obscure to matter much to the attackers. But "Security by Obscurity" is not a great design philosophy, either. Besides, I do want the Fediverse to have a growing user base - but that would mean that the Fediverse would become more tempting for attackers.In the long run, we might have to defederate Open Registration instances, as these are the most likely to attract bad actors. Indeed, we are already seeing things like this on a smaller scale with Mastodon.social. Yet we also need a good entry point to the Fediverse for newcomers.What are your thoughts on this?#FediAskhttps://arxiv.org/html/2506.06299v4

@dansup@mastodon.social @dansup@pixelfed.social Indeed He really tried though but still, oof

@pheonix the experience of getting a "recommendation" from a system that is wildly off the mark is so common that everyone, not just people like us who are deliberately living so as to be illegible to capital, understands this mismatch intimately. it's an illusion. the world of for profit digital advertising and surveillance is built on a colossal lie about how humanity works.

@yon true! but those are not LLM driven

@ratsnakegames Indeed. They were useful before, and they will remain useful long after the LLM bubble has burst.But stochastic parrots are more useful for an AGI investment bubble, so here we are.

@growfediverse @Darkcoffee There might be _some_ use for machine translations in highly specialized fields where the texts have very specific conventions. But that requires experts who train their models on these specific texts who know what they are doing, instead of hoping for "universal translator" software.And the texts are so diverse in their sources and styles that I doubt current models could do a proper job. And anyway, translating is only half the job - explaining the context of the texts is at least as important.

@DoomsdaysCW @ai6yr @driusan @stevewfolds Archive.org has it as well!https://archive.org/details/rumfordcompletec0000lily/page/n6/mode/1up

In preparation for this release, I used the `VHS` utility to make scripted short videos for various features. Here are videos showcasing a few core features that were also present in older releases:- Character cards and user personas- Message presets (configurable, ships with built-ins)- Code block highlighting and extractionSee more features in action here:https://github.com/permacommons/chabeau?tab=readme-ov-file#features

#PSA - I recently stopped my windows based computer backing every document and photo to the #Cloud because fuck #Microsoft and their #LLM - I never asked for their fucking cloud storage in the first place. Or their LLM that they incorrectly call AI - it is not AI and can’t even add up.Anyway, as part of this process, Microsoft then automatically deleted all my files - *including* the versions on my own computer. Fuck Microsoft.As it happens, I back up all my computer files on an external HardDisk on the first day of every month. As a result, I was able to copy the files and directories back to my computer with little disruption.If you are going to disconnect from your cloud storage - be warned in advance … back up all your files and directories on an external Hard Disk first.

I'm starting to understand what they mean when they say #LLM s are destroying the Open Web.Of course the proliferation of slop sites can't destroy your good site. What they actually destroy is *search engines*, that can't tell slop and real stuff apart.This means that we'll be back to hand-curated links and trusted sites, like in the 90s. I'm fine with that.#google #ai

(And no, that's not a root session . Terminator just remembers SSH sessions in the title bar.)

It is a point of pride for me that none of my translations of German folk tales were made with #LLM .https://wiki.sunkencastles.com/wiki/Main_Page

#WritersCoffeeClub 17. Should modern writers worry about proving their work is not generated by an #LLM ?I don't know about others, but I do not. My work is unique, and shows careful thought and research - all are things that LLM cannot do. Any LLM output resembling what I do would have to be trained on my work, and it would _still_ be bad.And frankly, if your writings cannot be distinguished from the sentences extruded by a stochastic parrot, then that's a "You" problem.

@RolandRides No, nor do I personally see any need for it.

@henrik Haha yes my lord!Whoops, no bread!

This is what many people (including myself) suspected would happen with these LLM SaaS services — that they would become spyware.And, not just that, but — that they would take spyware to another level.“OpenAI admitted that it's scanning users' conversations and reporting to police any interactions that a human reviewer deems sufficiently threatening.”https://futurism.com/people-furious-openai-reporting-police#AI #LLM #OpenAI #Privacy #Spyware

I had an epiphany while walking with my daughter to the park this morning.The number one reason why I would never resort to #AI / a #LLM to write something for me is that these models completely squash any individuality, the sum of all your life experiences and viewpoints. What makes you, *you*.What I find so dangerous is the homogenization of thinking that they bring about. A pastiche of vanilla thoughts, cobbled together at random.No LLM could possibly write for me... even if I used a prompt that asked it to write in my style, providing links to previous articles.My little one is only 4 but it breaks my heart that she's growing up in a world where these tools are totally normalized. How they're upending higher education and so many aspects of life.We didn't have a choice, the introduction of these tools arrived all of a sudden, without any guardrails in place. And now they're here.When the time comes, we'll have this talk. I find it essential./end of rant - sigh times a million#NoAI #MediaEcology #DigitalLiteracy

Good news in hard times."AI industry horrified to face largest copyright class action ever certified"Here's the best part."Copyright class actions could financially ruin AI industry, trade groups say." https://arstechnica.com/tech-policy/2025/08/ai-industry-horrified-to-face-largest-copyright-class-action-ever-certified/#cheerful #good #nice #ai #ruin #copyright #llm #greed #robot