Aside: WTF are they using SSDs for?
LLM inference in the cloud is basically only done in VRAM. Rarely stale K/V cache is cached in RAM, but new attention architectures should minimize that. Large scale training, contrary to popular belief, is a pretty rare event most data centers and businesses are incapable of.
…So what do they do with so much flash storage!? Is it literally just FOMO server buying?
HDDs should be fine. No?
Well. I just FOMO ordered a SATA SSD. Thanks, OP.
Mostly because I got 2x64Gb sticks 2 months before they shot up to nearly 4X the price! And it doesn’t feel like this bump is going away soon.
Ehhhh, not in fall. Hurricanes sometimes curl up that way, along with a bunch of Saharan dust.
Irony is LG has their own open weights AI: Exaone 32B.
https://huggingface.co/LGAI-EXAONE
It’s… not terrible. Especially for a multilingual, locally runnable one. But they gave it a license from the depths of hell, that even forbids reverse engineering and basically claims all its outputs, so no one uses it.
https://huggingface.co/LGAI-EXAONE/EXAONE-4.0.1-32B/blob/main/LICENSE
Anyway, I find it darkly hilarious that they choose to snub their own research, and their Tenstorrent partnership, and shove copilot in instead. How much you wanna bet they namedrop OpenAI in their earnings report?
This is corporate enshittification at its purest.
It’ll be interesting to see how Discord enshittifies.
It’s the default destination for the niche-interest “cozy web,” and they could go down several paths.
For LLM hosting, ik_llama.cpp. You can really gigantic models at acceptable speeds with its hybrid CPU/GPU focus, at higher quality/speed than mainline llama.cpp, and it has several built in UIs.
LanguageTool, for self run grammar/spelling/style checking.
Keep an eye on Open Web UI. I’ve heard rumblings that it’s starting to enshittify.
FYI, A1111 is obsolete. The diffusers or comfy-based backends are way faster, richer, less buggy and support newer things.
I’d recommend UIs that support SVDQuant, in particular.
AFAIK there are also problems that Chinese companies have their own tool chain, and are releasing high level truly open source solutions for AI.
One interesting thing about the Chinese “AI Tigers” is the lack of Tech Bro evangelism.
They see their models as tools. Not black box magic oracles, not human replacements. And they train/structure/productize them and such.
But with AI you can use whatever tool is best value, and switch to the competition whenever you want.
Big Tech is making this really hard, though.
In the business world, there’s a lot of paranoia about using Chinese LLM weights. Which is totally bogus, but also understandably hard to explain.
And OpenAI and such are working overtime to lock customers in. See: iOS being ChatGPT-only; no “pick your own API.” Or Disney using Sora when they should really be rolling their own finetune.
Wuthering Waves
I have nothing against a beautiful animesque art style, but… does it have to be rail-thin women in schoolgirl skirts and stockings? Is this game the exception?
EDIT:
Watches Steam trailer.
Literally magic schoolgirl skirts and stockings. Complete with winking in action.
Yeah.
People harp on Onlyfans, but how sexualized and “softcore teasing” Insta and even TikTok are kinda creeps me out. They’re literally entry points to OF.
I have a parent who’s blissfully off of social media, and it was interesting to see their reaction to what they’re like now.
And functional! You’re looking at Dakar rally champions, and hatchbacks so unreasonably fast on dirt, their league was banned.
This is what actual off-roaders look like.
Exactly.
It’s not that I’m indifferent to OF; the idea of getting in such a relationship actively creeps me out. And I’m already pretty messed up.
BG3 turned out well, though?
I propose we trick our fellow Americans by making smol cars offroady enough to embarrass an F150:
Look at them! Who would want a rolling brick over that?
And the Ford Focus is already mostly there.
It’s literally style. Those pickup lifts often ruin durability and off-road capability.
Yeah, please no Starfield. I love space opera, I love Oblivion and Fallout, jank and all, but the Starfield I got through was so boring.
I’d rather them continue GOTG. I’m not a Marvel person, but the GOTG game was great.
Again, I don’t buy this. The training data isn’t actually that big, nor is training done on such a huge scale so frequently.