If thats using pre-ukraine ratios that’s believable but this war has been extremely lethal for both sides. Instead of cassevac in hours it can sometimes take weeks so a survivable injury in Afghanistan becomes a death sentence in Ukraine. Look at that one dude who was stuck in his position wounded for months. They used a drone to cassevac him because rescue kept getting attacked and killed and the russian drones wouldn’t let him leave.

Punji sticks. Me use poop and sharpness. Give big lizard tiny microbe. Big infection. Me eat meat and watch it slowly die painful horrible death weak from no food or water.

Take them apart and smash any black chip. Although if they still have part codes you can search the code up to find what it is.

Well if you’re that paranoid set up encryption per disk so it hardware encrypts on the platter. A bad sector is just gibberish then. Wiping the encryption key effectively wipes the disk, and you can zero it out after as well. If you also don’t trust that use encryption at rest from a software side as well on top of everything.

I have electric baseboard heaters so i like to heat my home with my server in the winter. I dont use it enough in the summer to warrant the power draw, but i’m looking forward to doing so this winter.

I’m hoping the AI bubble pops soon… i’d really like to pick up a decent accelerator card for AI for my desktop. My current server doesn’t have the PCIe slots left (all filled with U.2 drives on PCIe adapters or HBAs) so it would need to go in my desktop… i thought 64GB was a lot when i built this 3 years ago but now i’m wishing I put 128GB in it…

My server has 238GB though (ordered a 256GB kit and one stick was dead. Caught that on the 4th test cycle with memtest86+ pro which I 100% paid for.) Since I use it for not only ZFS but also all of my random containers and game server VMs.

Maybe try netbird?

https://docs.netbird.io/

Strata is a good runner for that. It also can run Qwen3.8-flash-next which is also great

.ssh/config and tab autocomplete are how I manage. I do the squinting once if I can help it

Its ok, text is also really ambiguous so I don’t blame you

Yeah I know I was agreeing with you

Stm32 has made micro controllers with neural processors for years now and you don’t need LLMs to do a lot of useful stuff for a spyware company. A small model can run even in really constrained hardware if you train it specifically for it. Afaik the STM32 ones are for machine vision processing on limited hardware where power is a concern. I’m sure you can do a lot more with a TV with constant power.

Plus studies on skill decay when using AI. Also the linux maintainer onboarding pipeline—which was already critically endangered—is now no longer building understanding through writing code, just monitoring it’s generation.

I personally think you need to write a language by hand without AI doing it for you to learn in detail. It’s like trying to learn how to do math or how to write mandarin by just supervising an AI doing it… you need to do it yourself. Supervising is fine if you’re already fluent in mandarin and have been writing it for years and years. If you’re just learning it’s less than useless. If anything its directly harmful to learning.

Same with speaking a language yourself without crutching your native language as the best method to learn to speak it (immersion)

Edit: The long term externalities of AI—like pretty much everything nowadays—is going to be made worse by short term thinking. Quarterly profits and yearly results take precedence over long term skill development. Even OSS like linux will eat its seed corn out of expediency because of a lack of maintainers and new candidates being onboarded into the ivory tower mailing lists…

Yeah depends on the tracker some just credit you for seeding at all which helps people like you with bad upload.

Search lists in qbit are how most people manage.

I personally use torrentleech which is a private tracker but has basically everything and super permissive rules. Very very easy to maintain ratio and to also get invited and started. I have 4.8TB of upload credit. Searching in qbittorrent or manually across public sites took too long for me when i usually want to find a higher res or higher bitrate version of something. I also always saturate my download speed even with only a few seeds since they’re almost always seedboxes.

I’m in the same boat and i rent the place. Both bedrooms are on the same breaker and I’m upgrading my GPU (215W to 415W) and might get a Mi50 as a second card in my desktop for local LLMs.

At full power on both machines i’m already close to tripping the breaker. Witj these new card(s) i don’t think it’ll fly. My server uses around 700W at max, 450W idle if it’s nice and cool in the spare room. Mostly due to loooots of spinning rust. I’m a datahoarder and have a full 18 disk 3.5" JBOD, and 24/25 drives in my 2.5" JBOD.

I’ll figure something out I guess…

Like wire tapping enron 3 months pre meltdown

This is where things seem to be headed. Local models can be pretrained and crammed full of your work’s information, and opencode and other harnesses can pick the right model for the task ahead of time. I could be wrong but i’ve been predicting a drop in frontier model credit usage especially if they hike token prices again. People will only use it if they have to especially if you can find a way to task a local model with a lot of the grunt work.

Here’s a video on it

Qwen3.8-flash-next is making waves since its near frontier or frontier (according to some i havent used it) that can be run on an ancient 8GB video card (12GB or more is recommended. 16GB is best.) with around 32GB of RAM using Strata.

Strata

It’s a mixture of experts MoE model, so while it’s around 66GB (depending on which quantization you decide to run) you’re only really using about 6B parameters per token processed, meaning the VRAM requirements are a lot lower.

I personally don’t see the value proposition in paying $4/1M tokens when i can generate around 30tok/s on 500W of power. I pay ¢13/kWh so to generate 1M output tokens takes me 9.25 hours and costs $0.60.

I’m also running an ancient video card and have a new one in the mail. If i got 200 tok/s even at higher power use i’m still a fraction of the cost of paying for API access and I want that GPU for other things regardless.

I feel lile medium and large businesses have a strong incentive to rent or buy inference using open weight models for a fraction the cost of paying OpenAI or Anthropic. Especially if you can pre load them with everything about your org to make them better suited for internal use.

The vast majority of uses can use a open weight locally run model. I think you would only want to pay for frontier models for really complex tasks with lots of steps, and even then there are usually ways to be smart about how you task agents with work to get better output and use less money.

Also the financial ponzi schemes break all at once pissing off old people who vote in mecha-hitler to enslave everyone under 40 so their house prices and pensions stay high.

The only innovations made are new and unique ways to screw taxpayers and the general public