Why not just call the page Chat Control 2.0?

I mean, look at all the geniuses saying "I'll just use a VPN" in response to the latest ID for age verification. A week later, the law was amended to also involve VPNs.

Since decades.

GPT5-Pro is the real deal.

Gemini if I need creative insights and a pleasant talk, but this comes at the cost of more mistakes (and it's hella stubborn).

Hopefully Gemini 3.0 will fix this.

Except for the ransomware thing, or phishing mail writing, most of the uses listed there seems legit to me and a strong reason to pay for AI.

One of these is exactly preparing with mock interviews which is something I myself do a lot, or having step by step instructions to implement things for my personal projects that are not even public facing and that I can't be arsed to learn because it's not my job.

Long life to Local LLMs I guess

Someone is mentally ill, and can use AI. Doesn't mean AI is the problem. A mentally ill person can also use a car. Let's ban cars?

If the latter, they'll be the ones committing crimes at this point. Not everyone, but some of them. Every large enough subset of humans contains people that will commit crimes at some point and I don't want them to have full power and accessibility on my entire life without a warrant.

As the author admits, there's lots of good interests in keeping Tor a thing, and nothing good can come out from removing it.

Such people sharing illicit material should be investigated and prosecuted (with severe and appropriate court orders) through the ordinary means, but the author doesn't seem to have a good solution on hand, other than showing provocative questions like "why are we funding these Tor guys".

There's no single solution that they can implement TOR-side without breaking the whole point of the project. They won't implement any backdoor or similar bullshit, no matter what. There's no point in TOR if they do that. May as well disband the project.

TL:DR There's many ways to go about it.

Quick start?

Clone llama.cpp repo or download the .exe or main linux binary from the "Releases" on Github, on the right. If you care about security, do this in a virtual machine (unless you plan to only use unquantised safetensors).

Example syntax: ./llama.cpp/main -i -ins --color -c 0 --split-mode layer --keep -1 --top-p 40 --top-k 0.9 --min-p 0.02 --temp 2.0 --repeat_penalty 1.1 -n -1 --multiline-input -ngl 3 -m mixtral-8x7b-instruct-v0.1.Q8_0.gguf

In this example, I'm running Mixtral at quantisation Q8, with 3 layers offloaded to the GPU, for about 45GB RAM usage and 7GB VRAM (GPU) usage. To make sense of quants, this is the general rule: you pick the largest quant you can run with your RAM.

If you go look for TheBloke models, they all have a handy model card stating how much RAM each quantisation uses.

I tend to use GGUF versions, which run on CPU but can have some layers offloaded on GPU.

I definitely recommend reading the https://github.com/ggerganov/llama.cpp documentation.

Speaking for Einstein@home, for example, all their research is public and you also get a mention in their paper if a result was in your workunit.

It's quite interesting to see how it all began.

The peak of this project is largely behind us, although I was a proud member about 10 years ago.