Facebook LLAMA is being openly distributed via torrents • Hacker News Board

Re: Facebook LLAMA is being openly distributed via torrents

#321

Post by passion__desire » Fri, Mar 03, 2023, 3:41 PM UTC

Laaas wrote: post #232

DharmaPolice wrote: post #228

Earlier quoted context omitted.

It became something of a meme but there are huge numbers of guys out there that would pay good money for Joi from Blade Runner 2049. https://bladerunner.fandom.com/wiki/Joi

I think what Blade Runner 2049 got wrong was the way they depicted having sex with the Joi instance. I assume in 2049 we'll either have Neuralink available to enter a virtual world (à la VRChat) where we can do it more realistically, or we'll have the ability to buy full sexbots and put the Joi instance in them. We'll likely also have virutal brothels using AI along the same lines.

No need to create environment when the neurons can be stimulated directly. This scene from Pacific Rim Uprising movie freaked me out.

Dr. Newt (Charlie Day) heads home after a tough day at the office to his wife 'Alice'. Turns out, 'Alice' just happens to be a massive Kaiju brain in a tank. [0]

[0] https://www.youtube.com/watch?v=mIDTUYSIkcs

Re: Facebook LLAMA is being openly distributed via torrents

#322

Post by speedylight » Fri, Mar 03, 2023, 3:42 PM UTC

Legally speaking is it a good idea to download these models this way?

Re: Facebook LLAMA is being openly distributed via torrents

#323

Post by EamonnMR » Fri, Mar 03, 2023, 3:42 PM UTC

Gonna be interesting to see if Facebook tries to tell people they can't use this because it's stolen (when it was presumably built using data taken without permission.)

Re: Facebook LLAMA is being openly distributed via torrents

#324

Post by ImprobableTruth » Fri, Mar 03, 2023, 3:42 PM UTC

madmod wrote: post #235

Would there be some way to “launder” the model to make it plausibly viable for commercial use? Train a new model with the weights of this model with some kind of noise added to make it hard to tell what it is based on?

Distillation would be the ideal way (especially because it also has efficiency gains), but as far as I know distillation for LLMs is kinda unproven.

Honestly though, even if you just finetune it, which you will want anyway for any serious commercial application, it's essentially impossible to determine the origin.

Re: Facebook LLAMA is being openly distributed via torrents

#325

Post by onetokeoverthe » Fri, Mar 03, 2023, 3:43 PM UTC

Good thing ive had decades of real relationships and sex.

Knew the net would probably squash print and privacy the first minute i logged into aol.

Who knew it would breed a generation of robot loving losers?

Re: Facebook LLAMA is being openly distributed via torrents

#326

Post by CGamesPlay » Fri, Mar 03, 2023, 3:43 PM UTC

datadeft wrote: post #244

Taek wrote: post #164

Earlier quoted context omitted.

Electricity costs are basically irrelevant because the cards are so expensive. A100 cards consume 250w each, with datacenter overheads we will call it 1000 kilowatts for all 2048 cards. 23 days is 552 hours, or 552,000 kilowatt hours total. Most dataceneters are between 7 and 10 cents per kilowatt hour for electricity. Some are below 4. At 10 cents, that's $53,000 in electricity costs, which is nothing next to $30 mi…

> Electricity costs are basically irrelevant because the cards are so expensive. You mean in terms of money. I think this is exactly the problem that we have in CS, nobody really cares about CO2.

No, I'm willing to bet the CO2 cost of the cards is also way higher than the electricity. Those things are built on the global supply chain, with materials potentially making multiple thousands of kms journeys between each step.

Re: Facebook LLAMA is being openly distributed via torrents

#327

Post by EamonnMR » Fri, Mar 03, 2023, 3:44 PM UTC

speedylight wrote: post #322

Legally speaking is it a good idea to download these models this way?

My compliance brain says no, but the fact that models get trained with data they obtain without explicit permission makes says that finders keepers would be the relevant case law.

Re: Facebook LLAMA is being openly distributed via torrents

#328

Post by Filligree » Fri, Mar 03, 2023, 3:47 PM UTC

EamonnMR wrote: post #327

speedylight wrote: post #322

Legally speaking is it a good idea to download these models this way?

My compliance brain says no, but the fact that models get trained with data they obtain without explicit permission makes says that finders keepers would be the relevant case law.

It's not clear that the model weights can be copyrighted. But of course, I wouldn't want to be the test case.

Re: Facebook LLAMA is being openly distributed via torrents

#329

Post by hn_20591249 » Fri, Mar 03, 2023, 3:47 PM UTC

linearalgebra45 wrote: post #123

Hypothetically, what would the consequences be if I ran this on my university's computing cluster?

Seems a valid use of resources if you have a way to vaguely associate it to some academic side-project, just don't start monetizing the output and beware the wrath of stressed out PhDs if you use too much capacity.

Re: Facebook LLAMA is being openly distributed via torrents

#330

Post by EamonnMR » Fri, Mar 03, 2023, 3:47 PM UTC

Beaver117 wrote: post #20

Funny. iirc some of the big tech (I think it was Google?) use torrents internally to deploy very large images to servers. Piracy is not the only use case!

Used it to download linux distro images back when the size of an install CD was huge.

Good times.