thway15269037

IIRC some companies were running campaigns like "we buy your old hardware so you won't sell it in aftermarket and crash the prices". It all ends in landfill.

I wouldn't be surprised if RAM cartel or nvidia will attempt something like that, plus nvidia already talks about license-locking GPUs.

Also, again, IIRC, IBM hardware already works this way. There are almost no aftermarket when I last checked for them because it's all locked up and useless.

So that's why I can't buy a replacement 16TB HDD for less than a 1000$ when three years ago I bought them for 229$ a pop.

I wondered why do they even need slow HDDs, like, do they store training data on them? User chats? Do they assemble crazy 12-way stripes for fast speed? Do they store what, failed attempts?

Yeah, I clicked that too. But knowing qualcomm they obviously cost an arm and a leg. Still interesting how much.

Chinese labs seem to think that fable/mythos is a 10T+ model, and target that amount of parameters with next iteration of qwen/kimi/ds/glm. Proves nothing tho.

Yes, I understand the principle, but will it be usable? E.g. in my personal opinion border of usefulness starts somewhere at 7-10 tok/s. I currently have 192GB and 24GB VRAM, and I can run Qwen-Flash-Next (4 bit) and GLM5.3-Flash (4 bit) at 25 tok/s and 7 tok/s respectfully (with quantized cache on beellama). Parallel queries worsen result significantly but can push aggregate to around 50 tok/s.

I can't, however, run DS4.1-1017, for example, but maybe I didn't find proper options and quantization yet.

So I'm really interested how to be sure what can/cannot fit in specific hardware (for example, I want to buy a Mac Studio with 64 or 128 GB RAM, what MoE model it would run at comfortable speed? Will it be qwen-flash-next or something bigger? And so on). So far websites with "benchmarks" were really terrible in that regard (and huggingface calculator even worse!)

What's the price?

Excuse me, could you elaborate how you came up with 192GB of RAM? I almost thought they downsized model, but it's still 2.8T tokens. Are we talking about not-yet-out 1-bit quant or assumption that something like ds4 (daskstar, not deepseek) will heavily optimize it?

Not a trolling question, genuine interest.

How useful is 64GB M* Mac for running models? It is strictly below <64 Gigs or something bigger like Qwen-Flash-Next or GLM5.3-Flash is available at usable speeds too?

I kinda want a Mac machine to test and publish few apps, but I can't justify 128 or 256gb budget right now.

I've been giving my little games to text to both Qwen 27B and Qwen-Next, and they both try to debug bugs they encounter unprompted. So I have to specifically mention to them to not try and ghidra my game and just play it as intended or report a bug in BUGS.md file and carry on.

Is fastmail better than zoho? I'm have trouble choosing between them to move small org (of 5 people or so).

I wonder, with all these releases and with all impressive open models, is Mythos even relevant anymore? Did it even exist or was just a, ahem, "run 100 concurrent requests and select best response"?

Someone in some previous thread tried to summarize, e.g. "Sol < Astra, except when 6.1 Sol which is more that current Astra, but Terra 6.1 > Sol latest, and..."

So no, OAI scheme is also confusing as hell. And they are going to add even more confusion when they release presumably Nova.

You can never know for sure that they haven't conveniently forgot to cover with if(env) one or two of twenty two exfiltration functions. I'm not claiming they did that, but I've seen that already.

What's the difference between lima and conventional fat VM using kvm/virtualbox?

I only heard about smolvm yet and they are YC, so I'm hesitant to use them as they will probably one day pull the rug.

You're running them in containers, and not in VMs on separate network through network filter and logger?

At places where I did work anything private over even properly protected wifi was forbidden after several wifi vulns in 2018. So the only thing available on company's wifi was vpn gateway and page with manual how to set it up.

So, not trusting anything wireless is a thing for almost a decade.

"Looks minor, hard to reproduce, little affected area. Throw it into backlog for 2095 release cycle."

Was it opencode who bragged about their vibecoding tokenmaxxing attitude or I'm mixing them with someone else?

I'm also surprised when reading stuff here like "agent harness app %X% is a new moat!". I mean, I can point even GPT itself on their own new app and it happily obliges in dumping all features, writing product doc or disassembling itself.

Not that I want it -- my tokens are not free -- but it's here. Up for grabs.

I can't help but feel that something soon gonna break, and break big. Like, the decompilation stuff either gonna get criminalized, or forbidden in training, or I don't know. Current state of affairs became super imbalanced, where I can decompile 400k+ lines big game in a weekend and rip it's engine; or, as pointed in sister thread, as someone cloned music processor device in 2 hours.

The question is who and when first blinks.

There are no such thing as a "will of the regime" in a strict sense. People tend to project whatever model they have in their head and it's so wrong, yet highlights how insane out of touch is western expertise of russian internal affairs.

"The will" is so sparsely enforced there could be none, and, for example, Putin directly oversees a very, very small amount of topics. One of which is, for example, hypersonics program.

The rest is controlled by a whole goddamn lot of influence groups which president administration mediates (do not confuse with duma and federation council, those are mostly talking heads). For example, current internet censorship breakdown has nothing to do with actually censorship but with a small group of influential people who seek monetary interest in diverting traffic to their wannabe clones of western media.

No, I don't know why they chose such structure nor I have any idea why they selectively enforce one topic and completely abandon another.

Not pushing it in the slightest. Bonus tip: there are several heavily west-aligned companies similar to Yandex. I wonder if they are going to get bombed too, for all the help they've done. Would be very funny.

As for the rest, I sure do hope one north american country that invaded countless countries over the last 25 years will one day reap all the whirlwind it could. Maybe they should start building wind turbines, I dunno.

You don't know what "state-owned" means in terms of Russia and no one on this website does. It does not equal loyal to Russia's interests neither to interests of russian people.

Yandex is one of the most disloyal and pro-ukraine companies in Russia, it to this day does not recognize Crimea as Russia yada yada. It's own AI systems are hardwired to refuse any pro-russian topics and only cover pro-ukrainian side. In that way, gemini and chatgpt are both significantly more pro-russian, which is hilarious (if only they didn't block RU region).

Being state owned means nothing if there are no will to trample down disloyalty.

Again, see core comment. People in this thread cheering for yandex bombing really don't understand a single thing.

When you copy a file, it's trivially simple to prove it's a copy and not an original work.

How can one prove that X is a copy when none of the source code match original? Clean room re-implementations are not illegal after all, if we go that way (yeah, yeah, I know about clean room argument).

While I personally do not think it would be different, but a lot of LLM folks are, and even some companies (rushing to blatantly clone and de-compile stuff). Kinda strange feeling sitting and looking around in the midst of it, y'know.

That implies US jurisdiction plus that was made at the time where "hey, let's actually reverse-engineer and copy our competitor" was very expensive and time-consuming, so slapping one company would discourage everyone else to sink money in it. Currently it seems the latter is becoming either automatic (still somewhat expensive) or even free.

Even with 95% cache hit, and on cheapest chinese model, it would still be in tens of thousands of dollars (I assume that of 500B tokens at least 10B would be in "output"). If we switch gears to Kimi K* then if would very quickly escalate in hundred thousands USD.

Uber business model wasn't a subscription "for 10 bucks you can travel 900 lightyears a week"

I struggle to understand what legal leverage they used to threaten him to remove every detail about the game. Can someone post the game name and company name?

So, if you reverse-engineer game X and post reverse-engineered code, what exactly do you infringe, how and in which jurisdiction? What changes if it is done via LLM?

(I understand that LLM decompilation is absolutely out of hand right now and something surely will come to trample the fun. But what and when? I suppose american LLMs will have their system prompt updated to forbid any reversing help and report suspicious activity straight to legal hotline)

Clearly they should point an LLM to fusion as a problem and 88,000,000 agent*hours later they surely will have a solution! Or say that akhtually it's a harness problem.

/s

Grammarly is a keylogger and an asset of a spying agency[1]. No one should use them either willingly or unwillingly.

[1] almost every cloud solution that exfiltrates your data to the cloud is, but that is besides the point