this post was submitted on 07 Sep 2026
469 points (97.8% liked)

Technology

87897 readers
3997 users here now

This is a most excellent place for technology news and articles.


Our Rules


  1. Follow the lemmy.world rules.
  2. Only tech related news or articles.
  3. Be excellent to each other!
  4. Mod approved content bots can post up to 10 articles per day.
  5. Threads asking for personal tech support may be deleted.
  6. Politics threads may be removed.
  7. No memes allowed as posts, OK to post as comments.
  8. Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
  9. Check for duplicates before posting, duplicates may be removed
  10. Accounts 7 days and younger will have their posts automatically removed.

Approved Bots


founded 3 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
[–] panda_abyss@lemmy.ca 387 points 19 hours ago (6 children)

Because they all use the illegal Colossus 2 data centre from SpaceX/XAi/fascism central and that data centre went down.

Google, Anthropic, and OpenAI all have contracts with them.

When that illegally running environmental disaster of data centre goes down, all those services all go over capacity and you get 502 rate limit errors.

[–] teslekova@lemmy.ml 2 points 54 minutes ago

Ah, so Colossus 2 is the target. Remember that, everyone.

[–] Zephyr@sh.itjust.works 1 points 6 hours ago

I'm not sure that would effect inference on an already trained model. Data sets are primarily used in training.

[–] pelespirit@sh.itjust.works 109 points 18 hours ago (2 children)

Does that mean Musk's businesses have access to everything people do on the systems that use their data centers?

[–] just_another_person@lemmy.world 136 points 18 hours ago (2 children)

Generally no, because most hosting companies would have something baked into SLA/SLO contracts, but all of this shit is done so illegally and shadily now, I wouldn't put a hard "no" on that possibility.

[–] devfuuu@lemmy.world 37 points 16 hours ago (1 children)

Nobody can verify that those "contracts" are actually being honored. We are all assuming the fascists will self police themselves out of good will.

[–] just_another_person@lemmy.world 23 points 15 hours ago (1 children)

Not true. I've been involved in litigation with both AWS and Google over unsecured comms that were defined as being TLS secured in SLA/SLO contracts and found not to be. Not that anything nefarious was happening necessarily, but the expectation is clear.

Whether these asshats even check for such requirements with a Musk run company right now 🤷

It COULD possibly be that they are logging every exchange happening at the network fabric between the service layers, but nobody knows unless they intentionally take steps to investigate or accidentally prove it.

[–] expr@programming.dev 2 points 6 hours ago (1 children)

Still means jack shit. Companies will lie through their teeth and violate any and all they can get away with.

They are not to be trusted.

[–] Cocodapuf@lemmy.world 2 points 4 hours ago* (last edited 4 hours ago)

You're missing the point, nobody is talking about trusting these companies.

You can tell if a connection is encrypted end to end or not. And if your paying for that service you can sue if you aren't receiving what your paying for.

In digital security nobody relies on trust if they can help it. And security matters if you want to keep a competitive edge on your competition, so even shitty companies care about that.

[–] skvlp@feddit.nl 51 points 18 hours ago (1 children)

I hope you’re right, but Elon don’t strike me as the guy whose most compliant to SLA, law, or anything else that might be an inconvenience to him.

[–] esc@piefed.social 13 points 17 hours ago (3 children)

He can't micromanage everything and regular management will try to comply.

[–] bedwyr@piefed.ca 17 points 16 hours ago

Executives are the biggest cheaters out there. They will only comply if it will hurt them if they don't, and it's won't here, as long as the protection money is produced they don't have to worry.

[–] shartgargle@lemmy.zip 5 points 14 hours ago

I highly doubt that he really knows that much, either.

[–] skvlp@feddit.nl 6 points 17 hours ago

I agree with that. But I think "the right" data scientists can infer too much from all those AI queries, and I think Elon can abuse that for his own gain.

[–] panda_abyss@lemmy.ca 22 points 18 hours ago (2 children)

Impossible to know.

There are systems for doing things cryptographically secure, but I don’t know much about that.

[–] 314xel@lemmy.world 33 points 18 hours ago (1 children)

Cryptography? Security? FROM A GENAI COMPANY?!

[–] Archer@lemmy.world 3 points 15 hours ago

Well yeah, for themselves

Homomorphic encryption? Much slower than "trust me, bro".

[–] QuantumSparkles@sh.itjust.works 18 points 18 hours ago (1 children)

And where is said data center? For research purposes

[–] bad1080@piefed.social 20 points 16 hours ago* (last edited 16 hours ago) (1 children)

https://epoch.ai/data/ai-data-centers/directory/colossus-2 says "Memphis, Tennessee, United States" but also provides satellite imagery

[–] BeatTakeshi@lemmy.world 20 points 13 hours ago (2 children)

The numbers of this madness are fucking scary, 1.5 GW of compute power, and huge arrays of cooler to keep it operating. It's basically a 1.5GW heater in the open. Oh and gas turbines to provide (part of) the electricity. Capitalism is burning this world down for a buck.

[–] NotASharkInAManSuit@lemmy.world 10 points 10 hours ago (1 children)

Today I learned that “AI” data centers use more energy than time travel.

[–] shrugs@piefed.social 8 points 9 hours ago

i bet they will also travel us to the apocalypse much faster

[–] filcuk@feddit.uk 1 points 6 hours ago (1 children)

We can't really produce enough heat through industry to affect the earth globally, if that's what you meant. It's insignificant in comparison to what the Sun provides.
However there is undeniably localised issues caused by these insane structures.

[–] teslekova@lemmy.ml 1 points 40 minutes ago

Probably the gas turbines producing C02 and waste methane.

[–] BarbecueCowboy@lemmy.dbzer0.com 18 points 19 hours ago (1 children)

It's kinda surprising,

I know specifically where one of the big ones hosts its models and its not there, but I guess they could have infrastructure in there.

[–] panda_abyss@lemmy.ca 20 points 18 hours ago (3 children)

They’re oversold though, especially prompt caching and the parameter count war

The US model is that they think more training compute and parameters will result in the winning model, while the Chinese are focusing on RL and parameters efficiency due to compute limits.

[–] teslekova@lemmy.ml 1 points 38 minutes ago

Considering which country is better at building power stations, that's a fascinating dichotomy.

The US going for brute force when the brute force is more available in China... Priceless irony.

[–] percent@infosec.pub 6 points 10 hours ago (1 children)

The efficiency of Chinese models really is impressive. I generated sooo much code yesterday with Qwen3.6 35B-A3B running on an RTX 5060 Ti 16GB (+ a little CPU offloading). It got the jobs done at ~50 tokens/sec.

(It's not super complex code, just some scripts that I would not have taken to time to write manually.)

I'd love to upgrade to something with more VRAM, but even my current card has doubled in price since I bought it last year 😬

[–] isVeryLoud@lemmy.ca 3 points 9 hours ago (1 children)
[–] percent@infosec.pub 4 points 7 hours ago* (last edited 7 hours ago) (2 children)

There's not really anything interesting to show. It's just a home server in a 13 year old desktop ATX case.

There's no desk, monitor, keyboard, or mouse... But also no cool server rack.

Function over form, and it sits in a spare bedroom out of sight.

EDIT: I found the receipt for the case. It's a Cougar Volant Black Steel mid tower, purchased in 2013. So my server just looks like this:

[–] teslekova@lemmy.ml 1 points 37 minutes ago

Gotta love the black slab.

[–] isVeryLoud@lemmy.ca 2 points 7 hours ago (2 children)

I meant your LLM stack lol. I just have an RX 6800 XT in my main Linux PC for inference, but it has to share VRAM with the DE. Maybe I'll set it up for remote development from my laptop instead to free up VRAM.

What are you using? vLLM? llama.cpp? Which params? How much CPU offloading? Do you use draft models? Is it a MoE model? Have you tried llama-swap? Which agentic front-end are you using? I presume you set it up to access it without SSH'ing into the machine, did you do anything special or is it just a raw unsecured open port on the machine to the LAN?

[–] Damage@feddit.it 2 points 5 hours ago (1 children)

it has to share VRAM with the DE. Maybe I’ll set it up for remote development from my laptop instead to free up VRAM.

eh, just systemctl isolate multi-user.target

[–] isVeryLoud@lemmy.ca 1 points 5 hours ago (1 children)

Correct, that's how I would do it, but then I need another machine to act as a head.

[–] Damage@feddit.it 1 points 4 hours ago

If your MB has onboard graphics, maybe you could mask the GPU and just pass it off to a container running the LLMs I guess

[–] percent@infosec.pub 5 points 6 hours ago (1 children)

Ohhh lol. Yeah it's Llama-swap, running llama.cpp for now, but might add vLLM to the llama-swap config to experiment with NVFP4.

I mainly use MoE models so I can get decent speed while using a 150-200k context window. My go-to model has been Qwen3.6 35B-A3B for a while. I tried Qwen3.8 27B, but it was too slow.

Gemma4 26B-A4B also runs nice and fast, but I generally get better results from Qwen3.6. I don't remember exactly how much CPU offloading is happening, but it's not much. As long as I can get like 40-50 tokens/sec, I'm usually satisfied enough.

For the coding harness, I've been running Pi in an Apple Container (sort of like Podman, but better isolation in a microvm). Though, I recently configured VS Code to use LLMs on my server, and it was actually pretty decent. Still need to explore a bit more, but so far VS Code's AI capabilities seem much better than they were a year ago (they seemed way behind, back then).

Also, I don't connect any harness directly to llama-swap. I have another container running Caddy, which acts as a gateway to AI providers. For other services (e.g. OpenRouter), the API key is injected in the Caddy container. I don't like having API keys or secrets anywhere where LLMs can read them. It's not so bad for my own self-hosted LLMs, but not cool to send secrets to a server owned by someone else.

[–] isVeryLoud@lemmy.ca 1 points 5 hours ago

How has tool use been for you? I struggled a lot with tool use with Gemma and Qwen, to the point where I needed to build a healing layer.

Regarding the coding harness, I was looking for something CLI-based or JetBrains-based, and I haven't had much luck getting my local llama.cpp models playing ball with OpenCode. They keep losing context and misusing tools.

I'm not too familiar with Apple containers as I'm running a full Linux stack, but I'll give Pi a try, seems interesting! Does it work for coding tasks or is it strictly an "orchestrator"?

[–] asbestos@lemmy.world 17 points 18 hours ago (1 children)

Reads like American vs European/Asian cars

[–] dgriffith@aussie.zone 8 points 15 hours ago* (last edited 15 hours ago)

Always ready to try brute force first. And then some other, less palatable options if that doesn't work, like slightly less brute force.

[–] AnalogAllamma@lemmy.world 5 points 19 hours ago (1 children)
[–] BeatTakeshi@lemmy.world 3 points 13 hours ago (1 children)
[–] AnalogAllamma@lemmy.world 2 points 12 hours ago

Loosely named after MicroSoft (penis).