I worry that we are in a brief interregnum during which web browsers can provide good online translation of news stories in other languages, from foreign media.

Sooner rather than later, the big tech folks backing Trump will realize they can corrupt the online translation LLMs to bias foreign news stories in favour of their own political agenda. (After all, isn't propaganda the ideal application for AI slop?)

@cstross There is still european legislation. Use european translators ;-)
@jackpearse How does that help with following the news from outside the English language filter bubble (which is dominated by right wing nationalist propaganda)?
@cstross I meant using european translator tools! In europe the legal system works the other way around. We first habe to proove that things conform to legislation. Then we are allowed to release. Deepl in example will probably never tune the meaning of texts.

@cstross

I think this is one area where the uselessness of LLMs is a bonus, I'm not sure they are reliable enough to be able to consistently bias stories in the way that you'd want.

@adaliabooks @cstross When you already have a half-decent translation, you can put another algorithm on top of it that makes simple subsittutions, insertions and deletions, eg. turns "driver kills 3 in a car accident" into "Black driver kills 3".

(how do I know they're black? I don't, and they're probably not, I just blindly inserted "black" because I could)

@jzillw @adaliabooks That's exactly what I'm afraid of.

Canary-in-a-coalmine flashpoints where this is likely to happen first will be news items about Russia/Ukraine and Israel/Palestine, but also possibly China/Taiwan—these are the obvious ones with lots of AI expertise to implement it and/or money to bankroll it, as well as a political agenda to move it along.

@cstross @jzillw @adaliabooks "100,000 Palestinians tragically died after carelessly getting in the way of defensive weapons"

Of course, if the AI is trained on the English language and the Arabic versions of the same story, the bias will be built in by default, unless you can find an English source that consistently portrays Ukraine and Palestine as both defenders against an invading force?

@cstross in an authoritarian world of strong men all translated content will be machine translated. This will function as information border with unwanted concepts, Infos etc by any side can be blocked instantly.
This keeps the walled garden clean.
And in the long run Learning languages will be forbidden.
@Zeugs @cstross
I'll just leave this here...

@Zeugs @cstross As a Pole who works (remotely) in a "German" team, together with two Brits, an Austrian[1], a Bavarian, and an Italian (and that's not even my most international" employer to date), I can't help but notice how that would hurt the entertainment industries. But of course that's what fascists always do: they outlaw fun.

[1] The funny thing about Austria and Bavaria is they have a different German there, arguably a separate though related language.

@jzillw @Zeugs Shorter Fascist manifesto: NO FUN ONLY RAGE! RARRR …

@cstross @jzillw @Zeugs There's the only word the Doom Slayer says in the entire of the Doom 2016 -> Doom Eternal+DLC storyline (in which many people read him as All Rage, All The Time: the game leaves around plenty of evidence that it's All Anticapitalist Rage, All The Time)...

...upon defeating his creator, who looks *suspiciously* like him (the game covers why), he is asked:

"Tell me... have you nothing to say to your creator... before you strike him down?"
"No"

@cstross @jzillw @Zeugs (Canonically he's *also* descended from the protagonist of the Wolfenstein games: his conception of capitalism very much includes fascism and nazism or the setup in those two games wouldn't make sense)
@cstross @Lazarou Isn’t it in 1984 where they a minute of hate?
@jzillw @cstross
I always wondered how Russia and China get their News censorship synchronized. But probably they just don't have to because the Chinese Russian cultural ties are not that strong. (But I could also be wrong.)
For example: Chinese people call the Russian conflict in Ukraine a war.
What will happen?

@Zeugs @cstross I would guess similarities stem from them having a similar agenda.

The thing about Russians not being allowed to call it a war is an idiosyncrasy of the Russian legal system (it's like setting a weight limit for cars to 20 tonnes, but making an exception for agricultural vehicles, and then insisting your new main battle tank is technically a tractor).

@Zeugs @cstross
NewSpeak embeds EngSoc in brains.

@Zeugs @cstross - "...Infos etc by any side can be blocked instantly. ..."

Your faith in the competence of AI (the intended evil purpose fully granted) is interesting.

@jmax @cstross
It's probably not a transformer Model. Those will be deprecated, like civil rights rights. It's more like an RegEx. Maybe sequence to sequence or something new.
it could also be working in combination with human censors(connected to some input/output neurons) keeping it from hallucinating, like the Human Input in ChatGPT does.

It will not be perfekt, but sufficient for the masses.

@jmax @Zeugs @cstross So long as people tend to believe the "AI" over human translators, it's enough for it to be incompetent.

And humans have messed up enough: consider "we will bury you", which wasn't a threat at all

@Zeugs @cstross Suddenly powerful members of the elite start disappearing in mysterious circumstances, leaving only blood spatter, and a single green feather
@Zeugs @cstross I believe you have created a new dystopian notion.
@johndberry @cstross
Sine the election there is no distopian anymore. Cyberpunk is now utopia 🙃

@cstross

Charlie, don't give these guys ideas. Remember "Five Days of the Condor."

🤣🤣🤣

@sysfrank I haven't seen it.

@cstross

That's probably because I've lost my mind. I should have said, "Three Days of the Condor." I should really have ChatGPT double check everything I write now that I've turned a certain age.

Apologies.

🤣🤣🤣

@cstross Why limit yourself to foreign news? On-the-fly AI “moderation" of domestic websites makes perfect sense. To protect the children, you understand.

Now that more than 40% of the most popular websites use the Cloudflare CDN, you even have the perfect chokepoint at which to carry out the transformation.

(accessing the original website directly without passing through the CDN is, of course, a DMCA violation)

@cstross News-oligarchs (and governments if they're still in the news-production business, BBC) have the resources to subvert the translation models. I'm not sure it's still true, but open source OCR tech is a decade behind closed source—If you want to do it well, you need to pay for it. If we're in that space for translation ML models, there are clear incentives to subvert (or pay to subvert) the for-profit models.
@cstross
Once propaganda begins it rarely stops. Once it becomes indistinguishable from the truth it begins to replace it. When I was in an Army psyops unit we studied how lies, repeated often enough, become the accepted "truth".
The average home computer is powerful enough and machine learning tech efficient enough now that translation by LLM can be done locally without depending on any online services, as Firefox already does.
@cstross People are using summarisers to read stuff that's in the same language and they're going to get corrupted as well.
@cstross
Teaching the progeny to read and write natively in three languages. Would do more, but we're humans.
@cstross The good news, if you can call it that, is that networks that do an adequate job of translation are a great deal smaller than the GPT-style LLMs, and so far cheaper to build (not *cheap*, but possibly doable these days with government grant money and the right source corpus).
@cstross To avoid this particular dystopia we need to support locally run models. The ollama project and similar let you run a growing number of models on your own hardware under operating systems you control. You can keep the snapshots of the pre-censorship language models and use those. How hard the browser manufacturer fight back against local translation service add-ons will be very telling.

@cstross
Translation in Firefox is done on-device using models built by the European-based Bergamot project plus some we've trained ourselves.

https://firefox-source-docs.mozilla.org/toolkit/components/translations/resources/01_overview.html

Overview of Translations — Firefox Source Docs documentation

@cstross oh yes, they will definitely put a lot of money into "reducing bias in AI output" 🤔

@cstross I feel as if I might immediately dismiss this possibility as too weird and wild, but for the fact that I got memetically primed to apprehend the risk of this sort of information warfare a long time ago. You see I read this mind-bending book called Glasshouse... 😃

All frivolity aside, a thing I love A LOT about reading science fiction is when some plot involving entirely-invented tech helps me understand, usually many years later, what's suddenly happening in the world around me.