That’s actually crazy.

For reference, Split Fiction’s development cost was ~$40 million USD over ~3-4 years, ~75 devs. That’s one of EA’s better games, and not cheap.

Yearly, $77 million could basically fund Battlefied development (with BF6 rumored to cost $400 million, maybe?).

It’s 11% of what they’re looking to cut!

Even if the rationale for the bonus is “Andrew Wilson put Battlefield back on track,” that’s still a good chunk out of the profits of ~500M of BF6 game sales so far.

It’s 1% of EA’s entire yearly revenue they last reported, $7.84B.

…Apathy?

TikTok is basically a time killer, short shots of dopamine. Such users aren’t thinking about it that hard, so it would stand to reason they aren’t considering the impact of the noise either.

I’d also posit it’s an “addiction,” akin to smokers who need to smoke somewhere they know they probably shouldn’t. Maybe they don’t have earbuds on them immediately, but they feel the pull of TikTok, so they pull their phone out and turn the audio on, anyway.

What’s more, it’s not recognized as an “addiction” by the public, so the noise doesn’t have the stigma of secondhand smoke, either.

deleted by creator

It’s ironic. Windows went on AI, not just visibly with Cortana, but under the hood too. For example, they spent a ton developing ONNX, even before “AI” was everywhere, and still push it as a format.

Same with the vendors. AMD NPUs (for instance) were only supported on Windows forever; I think they still are. Same with Intel’s.

…And yet Windows is basically useless for local AI, compared to linux.

gift cards to give to thirst traps.

What does he use specifically, though? Facebook, Instagram? OF?

On his phone, right?

I apologize if that’s a weird question, but I’m… scientifically curious. I know lots of folks are susceptible, but I’m interested in the systems that suck folks like your father in.

IMO it’s only a problem when the customer doesn’t understand that they’re buying an illusion.

+1. But I wonder how many know it’s an illusion.

I’m the wrong person to analyze this though, as I do not get why people engage in these services. Like, I get roleplaying, I’ve daydreamed or even written out long fantasies, but why would I (for example) pay for some girl to talk it out on the phone?

…Much less gawk at some influencer on Insta?

What’s the point?

I’m always shocked by how LLM illiterate professionals or even leading “science influencers” are.

…But I know why.

It’s on purpose

Literally everything about the most common (Big Tech) interfaces obfuscates everything about them, to the maximum degree, to present them as “sycophantic oracles” and maximize engagement. If that’s all one is exposed to, there is literally no way to get a cold ass bucket of water in the face by peeking under the hood.

It’s all very… anti science.

On a semi related note, see this absolutely unhinged but not wrong rant:

The Conspiracy Against High Temperature Sampling

https://gist.github.com/Hellisotherpeople/71ba712f9f899adcb08b94bce20d5397

It is related. Anyone, even Hank Green, will rely on AI more if they never get a chance to poke it and peel it apart.

Like, if I ware in charge of a university, I would mandate a course where everyone gets a untuned base model in raw completion mode, and gleefully watch the slow realization. “Wait… all “AI” is just this?”

Also:

I’d love to do case studies on the real people at the end of these things.

Like, who is scrolling Facebook and Insta to thirst over algorithmic posts, yet is somehow unaware of the generative explosion?

Do they know, and just don’t care? Do they think they could tell if an image is fake, or is it true unawareness?

What do they do in real life? What are their relationships like?

How does this system work.

I think it’s easy to assume they’re all basement dwellers or whatever, but one could stereotype Lemmy the same way, and that’s just not true.

EDIT: Now that I think about it, such studies must exist. I don’t want to fall into the “I didn’t even try to search for the answer” thing.

But all this has happened so fast. And clearly the studies aren’t being acted on…

I ran into a Discord where the sole topic was running hot girl “AI influencers” on Instagram. Literally hundreds of thousands of people thirsted over them, and the operaters made real money.

Their systems were pretty sophisticated. The whole process is automated; LLMs write the posts and do sexy DMs, imagegen and videogen models and such do the sexy images and videos, controlnet keeps the look consistent-enough between media. LLM agents press all the buttons. I think there may have even been OnlyFans stuff.

So a single operator could run dozens of bots.

…It’s not hard for these models, either. DMs and short Insta posts are stupidly simple language, and media of sexy people in clothes, doing insta-style poses, probably makes up most of the image/video model training corpus.

I think there was some engagement farming with other bots, of course, but it would snowball and turn organic. Real Instagram users were thirsting over these bots.

…It got me thinking.

Is it really that different from a parasocial relationship with a “real” influencer?

I mean, functionally, its exactly the same. The users can’t even tell the difference.

I posit that the whole “thirst” influencer system was broken and exploitive to everyone involved, except Facebook.

It was fucked up to begin with.

AI just made it more obvious to onlookers. It destroyed the pretense that any of it was “real.”

Gestures. This is what I’m talking about.

A large fraction lemmy.ml literally cannot participate in a thread without bringing up western imperialism. Like, it could be /c/superbowl picture of an owl, and somehow it would devolve into that.

Unfortunately condensation is, indeed, an intractable issue there.

Sometimes people put a heat pump in water cooling loops, where condensation along the pipe can be mitigated with insulation, but even that’s a little risky with a 5090 these days.

Enjoy your lovely greenery!

the old Encyclopedia Britannica.

No one uses this.

You’ve got dozens of alternative Wiki competitors with their own rarified communities of editors and contributors.

No one uses them.

You’ve got subject-specific Wikis for a thousand different fandoms, hobbies, and crafts.

Fandom has a grip on this space, unfortunately, and that’s under Wales too. Yes, I know there are community alternatives, but their traffic/SEO is small in comparison.

But the idea that tens of millions of curious people and subject-experts are just going to endure a vacuum? Come on.

They absolutely will.

They already have!

Look at the rest of the internet. Literally every space, every niche, every congregation of experts has rotted away, and these “curious people and subject-experts” are clingimg to big social media toxic spaces because it’s the default.

It’s a very English-language and Euro-History oriented source of truth for tens of millions of online nerds. If it gets subsumed or shut down or otherwise undermined by western-aligned governments, I might suggest that the problem will not be an absence of Wikipedia but the presence of a malignant surveillance-enabled censorship bureaucracy.

You’re speaking of an internet that, for the bulk of humanity, doesn’t exist anymore. It is all but gone. That’s what I’m saying.

Maybe a few million nerds would create new Wikipedias, but the billions of other people on their phones and tablets would almsot never see it.

And that is dangerous for society. Wikipedia’s sheer critical mass is a huge monkey wrench in “soft” censorship, and for all its faults, every bit Wikipeida is delegitimized is a fascist dream-come-true.

And I hear a lot of people talking like that popularity isn’t important. There’s this sort of “old internet” attiude like we can just archive it and keep to ourselves. What does everyone else matter?

Well. Without Wikipedia putting the braKES on loss of freedoms in our IRL lives, we may lose the ability to do that.

Well, unfortunately, there is no project like that with Wikipedia’s critical mass. Most attempts with any attention are rather horrific. See: Grokipedia.

If Wikipedia dies, nothing good will replace it.

I know that’s a bit defeatist, but I really think its “the best we got.” Nothing better is coming with the state the internet is in now.

And I just want to emphasize that maybe a small community could build something better. I know.

But in terms of “keeping a healthy knowledge base for civilization,” that kind of project doesn’t matter if no-one sees it. Wikipedia is valuable because its a reasonably sane default reference for billions of people, that isn’t some megacorp or government.

I think you missed my point.

I’m not worried about them running out of funding. I’m worried about their reputation in the public’s eye.

Duct it!

I have a 400W 3090 with zero case fans.

Its always cool, because I bought like $10 of weather stripping to “seal” its intakes against the edge of an SFF case. So it’s always sucking in ambient air:

You can also severely undervolt a 5090 to like 250-300W, with very little performance impact. Honestly its stock speeds are kind of crazy.

…But again, you aren’t gaining much over a 4090. A 4090 + DDR4 threadripper would be way faster than a 5090 + occulink, for many reasons. And TR CPUs are pretty reasonably priced compared to GPUs these days.

But even if you do go 5090, I’d highly recommend finding some way to shove it in the case and duct some air into its intakes. Its going to be way faster on a PCIe slot.

You could even get a riser and mount it somewhere else in the case, theoretically.

This is bad.

…But I’m very worried about the future of Wikipedia.

Questionable leadership or not, it is a lighthouse in the information dystopia of the 2020s, which literally helps hold society together.

The far right wants it gone, and are not shy about screaming it, because Wikipedia is not compatible with fascism.

If it becomes hated by the left too, and ignored by the masses because of Big Tech blacklists it, I worry it won’t have the political support to keep its prominence.

And to be clear, I’m not worried about it dying. Wikipedia will stay where it is for decades.

I’m worried about it ceasing to be popular.

It doesn’t function as a uncontrolled source of information if no one sees it. Modern fascist censorship works by simply shifting attention.

First of all, I mean zero offense with any purchase decision. A 5090 is very good.

…But if I were paying that kind of money, I’d probably get a 4090 and a new motherboard/CPU instead. Maybe a used DRR4 threadripper system.

Hybrid (CPU + GPU) inference is where it’s at these days. It opens up a whole world of huge MoE models, whereas on an 5090 you are stuck with Qwen 27B.

Having a fast CPU, with lots of RAM channels, with full PCIe bandwidth is much more important for that than having a 5090, where a 3090 or 4090 will get the job done.

Even if pure speed is your primary concern, you can tune a sparse 120B model (like Laguna) to run almost as fast as Qwen 27B on a 5090, and get at-least-good results.

It’s more finicky and involved, though. For sure.

Running an LLM on a 5090 is a task. Hybrid CPU + GPU inference is a hobby.

Be aware that you pay a significant performance penalty for going over occulink.

The hit modest for fully offloaded dense models (like Qwen 27B), but dramatic for hybrid inference of big MoEs.

Even my old 3090 got a noticeable performance gain going from a PCIe 3.0 x16 riser to a PCIe 4.0 one.

• It’s just under 300B, trained at FP4; absolutely the perfect size for servers with 128GB-192GB CPU RAM to spare.

• Its fast. I’m getting 17 tokens/sec on a single RTX 3090 GPU, all experts offloaded to RAM; for a 300B model this smart, that’s crazy fast.

• Its attention mechanism is cutting edge, good for long context without too much processing time.

• The benchmarks for coding/agenic usage are absolutely bonkers, within margin of error of frontier models or Deepseek Pro in some cases: https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731

• …Though I don’t put much stock in benchmarks. And I haven’t tested it enough to tell you if it lives up to that hype for specific use cases.

• Deepseek also publishes its base model. That means I can “unfry” the model with a merge if I have to.

I am afraid of the the model being “overfit” to coding and agenic stuff.

For reference, my previous favorite model was Xiaomi MiMo V2.5 310B. It benched well, but it also feels “smart” outside of benchmarks, like in knowledge of trivia without tool usage/internet access or comprehension of weird questions.