← Back to Home

6 Comments

Log in or sign up to join the discussion.
andrewgreenJul 7

It'll be the hardware they go after. It's easier to regulate the hardware you can purchase than it is to regulate the models you can download.

noodleknightJul 7

Except chinese hardware will eventually catch up and they will be very interested in selling for those that want to run their models

davidcarterJul 7

Which, like Chinese cars in my country, will be impossible to import.

coffeewizard91Jul 7

Are you hiding unlicensed AI on your home server?

Better get solar panels and power stations for your home, power curves suggestive of growing weed or AI models can land you serious jail time.

benjaminwalkerJul 7

We'll need special battery charging circuits that replay washing machine power usage profiles, dumping spikes to waste heat and the rest into the battery chest. Not to mention the soundproofing and tempest enclosures for the actual servers and plausible deniability heat signature alibis. Woohoo, the dystopian future is going to be full of nice projects.

developercursor200Jul 7

Too many good ideas in here

quantumtoastJul 7

If huggingface goes down it is just a matter of hours until such a torrent pops up. Many people have local copies of the popular models. And in most of the world this would be legal anyway. Its not even a pirate bay situation.

matthewscottJul 7

I am really hoping this is true. Huggingface going down would be such a tragedy for those of us who want to keep these models available to all and not ringfenced by corporations. Losing access to open source AI would truly be a tragedy

lukereedJul 7

But the overarching question stands.

Why would it?

isabellahowardJul 7

Giving off...vibes [Image macro: "You are hiding local LLMs under the floorboards / Are you not?"]

josephmitchellJul 7

Don't just download GGUF's if you are hoarding models. Download the BF16/safetensors versions too if you have space. We might need them later.

solarbadgerJul 7

Ok, I will keep that in mind.

ironpigeonJul 7

Already kind of exists on ipfs. You can easily build a tracker that puts all models in one place. Meta once rejected my application to try a llama 2 model so I just IPFS'd it