My thought as well. Anglin's going hard on "the government is setting up an AI police grid and conservatives are cheering it on". While a fair concern, it seems to me they were going to do that anyway, and in the meantime, they're destroying the trannies. I'll take that deal, as you say, for the time being.
I was talking about this a few months ago, I think it's worse than just enforcement. Traditional forms of propaganda are broadcast at people and it's one-way: you can't afford an entire guy to wander around telling people what to do so you broadcast something that you hope has broad appeal. But if you've got enough disk space (and, for what they encode, tensors are remarkably compact) and enough juice to push vectors around, you can watch people individually and tailor your message to each person, make it appear next to as many avatars as you want. It's how phishing campaigns and spam work already.
> I'll take that deal, as you say, for the time being.
Khrushchev said that most people compared the Allied casualties to the Axis casualties to see who was winning, but Stalin added them together to see if he was winning.
@p@judgedread I mean, more basic, it doesn't appear I can stop what the government is doing, so might as well be happy they're fucking over antifa and trannies in the process. They're gonna do their schtick whether I support it or not.
@DailyStormerDigest@judgedread Well, sure, if you model them as an unprincipled chaotic entity (basically the only way to model them), this time it's raining on the other parade. I wasn't really disagreeing.
@p@judgedread Nor me with you. Just wasn't sure how else to respond.
I do agree the extent to which we can all be Skinner Boxed with fake AI generated news is as hard to grasp as it is frightening. That was actually the first thing I thought of when there was the apparently false reports that the video showing blood gushing out of Kirk's neck was an AI fake: "here is the first instance of all of us being Skinner Boxed". It obviously wasn't the case, to the extent we accept the official story at face value.
Best outcome: democratization of AI so we all have it as a weapon. Control of cyberspace is the most important front in the war ahead.
> I do agree the extent to which we can all be Skinner Boxed with fake AI generated news is as hard to grasp as it is frightening.
I mean, why wouldn't they? I can't wait for, 30 years from now when it is somehow revealed that this has occurred, everyone agrees that the government has to promise to stop doing that and the government says they'll stop doing it and nothing changes.
Troll farms are prohibitively expensive; an LLM isn't, and it writes better than a minimum wage temp and it will analyze some Karen's entire social media history without getting bored. So for the compute equivalent of a $15 Pi (which will be idle most of the time), you can have a machine follow an entire person around, day and night, and fabricate some whispers in their ear.
> That was actually the first thing I thought of when there was the apparently false reports that the video showing blood gushing out of Kirk's neck was an AI fake: "here is the first instance of all of us being Skinner Boxed".
I think it's a reasonable expectation to have at this point.
There was some series of hand-wringing articles ~20 years ago about millennials being too cynical and advertisers being mystified because people were trusting word of mouth from friends about which yogurt was best instead of a 15-second short film where three diverse women all agree that Yoplait is too delicious to be healthy but it somehow is both delicious and good for you and is guaranteed to contain less than an LD50 of corn syrup. Then the Data Science decided what to put into people's mailboxes and consumption was saved forever.
And if they can't herd everyone on earth into the same platform and people keep wandering out to places like fedi (if you've read the "Hydra on the Web" paper that the German government put out, that is their lament: https://media.freespeechextremist.com/rvl/full/faa13e7f66b7ff9e97634aaf04f6fd4faab95cf3931080fd03b0d98d76c7df2a ), they can do this. I don't know how many spare cycles all those DoE machines have but Microsoft and Oracle love federal contracts, and they're most of the privately held clusters on https://www.top500.org/lists/top500/list/2025/06/ . You know, back of the envelope, right, I think a lot of organizations on the high side of medium-size could afford the computes. I don't know if you're familiar with the academics scraping fedi but enough time on the UU cluster to process a small place like fedi fits in individual professors' grant budgets, apparently.
> Best outcome: democratization of AI so we all have it as a weapon.
Yeah, for the first time in a very long time, the bottleneck is hardware availability. The datasets and a lot of models are open, but the sophisticated ones want a lot of juice, and even if the "regulations" grow legs, they won't in China, so the Chinese models kind of set a floor on availability.
> Control of cyberspace is the most important front in the war ahead.
@p@DailyStormerDigest@judgedread > for the compute equivalent of a $15 pi Which specific LLM(s) did you run? I kinda want to mess around with those too but every Person On The Internet told me I needed 5 bajillion GBs of VRAM or some nonsense.
@rfgee@DailyStormerDigest@judgedread Yeah, the Internet's full of shit as usual. Take this with a grain of salt, like I'm not making a large amount of serious use.
Get the really low-parameter ones, qwen3, deepseek-r1, llama3, they're all pretty serviceable on a small ARM. Really small machines, you can get away with a lot using smollm for simple stuff. They do okay at fuzzy natural language stuff; I played with code generation and even the fancy ones seem terrible at that, but moderately useful for answering some pain-in-the-ass questions like "What's the library that does $x in language $y called? There has to be something for that in $y."
I read some of that paper and it infuriates me, the only time they use "free software" correctly is to demonize it.
>This report concludes by looking ahead, including the challenges and opportunities involved in regulating use of free software by the far right
>The fact that instances are not managed by large companies, but by individuals or groups at their own expense and with the help of free software, also has implications for their regulation
>The Dutch Blender Foundation, which hosts an open-source 3D program" Blender is also FREE SOFTWARE! REEEEEEEEEEEE
>Pleroma, which is also a part of the Fediverse, found that less than 5% of the accounts on blocked instances were sharing “toxic” content. The blocking of instances may be a reaction to this minority, but it also punishes the majority who are not spreading any offensive content. Therefore, this study proposes the following for tempering these side effects: 1) Creating blocklists for topics such as hate speech which are continuously curated by the community;
>2) New guidelines that moderate on an account-by-account basis Only good suggestion in the paper I believe
There is also a bit on free software and open source which rather correctly differentiate them. I find the bit to be neutral.
@p@DailyStormerDigest@judgedread Thanks! I didn't realize ollama was this painless to get running. Turns out my mediocre laptop can run all of the small ones and even gpt-oss 20b acceptably on the cpu too.
>newsmast-scrapers--mapping-fediverse-communities.pdf ffs what's with these people and their inability to paste graphs in a legible form into their almost default Word documents.
> Troll farms are prohibitively expensive; an LLM isn't
Fresh water isn't enough scarce yet, but give it time and hopefully people will start burning datacenters to the ground when they find out they waste as much water as some small countries while their sons and daughters are starving.