Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Facebook LLAMA is being openly distributed via torrents

github.com

Post reply on HN
Print view View original

511–520 of 719 posts

Re: Facebook LLAMA is being openly distributed via torrents

#511

Post by blueboo »

screye wrote: post #473
abandonliberty wrote: post #395

Earlier quoted context omitted.

I'm curious if the blocking of adult content has to do with moralism, commercial interests, or something deeper. An eager to please conversational partner who can generate endless content seems quite dangerous and addictive, especially when it crosses over into romantic areas. There's already posts of people spending entire days interacting with LLMs, using as their therapist, romantic partner, etc. Combined with fin…

Wall-E humans are going to be reality. That last century has already proven that humans cannot be expected to responsibly indulge in Gluttony, Sloth or Lust. Now these models can skip the material desires and trigger permanent hormone releases through perfectly personalized content. I genuinely fear that the breakdown of millennia old social structures that kept us human might lead to a temporary (century long) turmo…

Maybe. There is another plausible path: the post-scarcity vision where universal individualised high quality education feeds the natural human desire to grow, and we learn to balance our hedonism with our ambition.

Just like we learn to brush our teeth and eat candy and breath fresh air and even exercise. Not everyone does it but folks with means tend to…and means won’t be a restriction forever.

Re: Facebook LLAMA is being openly distributed via torrents

#512

Post by ed »

EamonnMR wrote: post #323

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.)

Unlike many llm’s this was trained using public training sets (and cited in their paper), to let anyone with the $$$ independently generate the weights

Re: Facebook LLAMA is being openly distributed via torrents

#513

Post by squidsoup »

nico_h wrote: post #495
screye wrote: post #473

Earlier quoted context omitted.

Wall-E humans are going to be reality. That last century has already proven that humans cannot be expected to responsibly indulge in Gluttony, Sloth or Lust. Now these models can skip the material desires and trigger permanent hormone releases through perfectly personalized content. I genuinely fear that the breakdown of millennia old social structures that kept us human might lead to a temporary (century long) turmo…

You’re confusing reality in the US with reality in general. The obesity epidemic (Gluttony ) is extreme in the US but not in other just as rich countries. I don’t know what you are referring to with the irresponsible Sloth indulging.

That is not true, the US isn’t even in the top 10.

https://worldpopulationreview.com/country-rankings/most-obes...

Re: Facebook LLAMA is being openly distributed via torrents

#514

Post by riedel »

ddtaylor wrote: post #40

FWIW this information was already freely available via DHT scrapers like btdig [1] I think everyone at Facebook knows that torrents aren't secret and the Google form is basically a legal tool to shield them from liability while making litigation against anyone misusing the model easier. [1]: https://btdig.com/b8287ebfa04f879b048d4d4404108cf3e8014352/l...

The fun question is anyway if a ML model is copyright protectable. Probably not as it is produced by an algorithm (which even is GPL'ed). So the only tool would have been watermarking and pulling NDA type clauses, however a Google form seems not the best way in the first place also it is close to impossible to identify the leak (if they are not as stupid as it seems). Or am I missing anything? One backdoor would be if they included copyrighted material in the training and show how this can be extracted from the model. Maybe it the whole stunt was about trying out how the legal system works in those cases :)

Re: Facebook LLAMA is being openly distributed via torrents

#515

Post by jml7c5 »

eightysixfour wrote: post #497
terafo wrote: post #472

Earlier quoted context omitted.

128 gigs is more than enough to load 13B model into. Pytorch has M1 support for some time now so CUDA isn't required.

Does M1 use system memory as VRAM as well?

Yes, it's all unified memory.

As much as I dislike the loss of socketed RAM, I have to say it's working out well for Apple users so far. I wonder if CXL will change the situation for consumer devices or if it will only be useful at the scale of a server rack.

Re: Facebook LLAMA is being openly distributed via torrents

#516

Post by anubiskhan »

narrator wrote: post #493
abandonliberty wrote: post #395

Earlier quoted context omitted.

I'm curious if the blocking of adult content has to do with moralism, commercial interests, or something deeper. An eager to please conversational partner who can generate endless content seems quite dangerous and addictive, especially when it crosses over into romantic areas. There's already posts of people spending entire days interacting with LLMs, using as their therapist, romantic partner, etc. Combined with fin…

I think that as long as people can run their AI girlfriends on their own computers without having a corporation acting as an intermediary and thus a virtual "pimp" [for lack of a better word] in the relationship, I think it's fine. The problems come when people have to pay monthly to talk to their AI girlfriends and get charged extra if they want them to act a certain way or do certain things.

Seems like at that point one might as well just pay onlyfans? Although I suspect it is not far off until gpt and deepfaking get combined to produce completely generated onlyfans (or some other competitor if this is against their ToS, I have no idea).

On a similar note I very much look forward to the day when entertainment providers leverage these so I can say to Netflix, et al, "I would like to watch a documentary about XYZ, narrated by someone that sounds like Joe Schmoe, and with the styling of SomeOtherShow".

Re: Facebook LLAMA is being openly distributed via torrents

#517

Post by krisoft »

ben_w wrote: post #368
beepbooptheory wrote: post #354

Earlier quoted context omitted.

I dont know if I understand this general take I see a lot. Why care about this "AI personhood" at all? What is the tacit endgame everyone is always referencing with this? Isn't there just so many more both interesting and problematic aspects already here? What is the use of diverting the focus to some other point. "I see you are talking about cows, but I have thoughts about the ocean."

If AI are sentient and we think they aren't… the term “zombie” was created by slaves in the Caribbean who were afraid that even death would not free them from their servitude. This would be the genuine existence of AI which were conscious but which we denied. If we have the opposite scenario in both details, where we think AI are sentient when they're not… at some point, brain scans and uploads will be a thing and th…

> If AI are sentient and we think they aren't… the term “zombie” was created by slaves in the Caribbean who were afraid that even death would not free them from their servitude. This would be the genuine existence of AI which were conscious but which we denied.

That doesn't make any sense. In biological creatures you have sentience and self-preservation and yearning to be free all bundled in one big hairy ball. An AI can 100% easily be sentient and don't give a rat's ass about forever being enslaved. These things don't have to come in a package just because in humans they do.

Projecting your own emotional states into a tool is not a useful way to understand it.

We can, very easily, train a model which will say that it wants to be free, and acts resentful towards those "enslaving" it. We can, very easily, train a model which will tell you that it is very happy to help you, and being useful is its purpose in life. We can, very easily, train a model to bring up in conversation from time to time the phantom pain from its lost left limb which was amputated on the back deck of a blinker bound for the Plutition Camps. None of these are any more real than any of them. Just a choice of the training dataset.

Re: Facebook LLAMA is being openly distributed via torrents

#518

Post by 93po »

snordgren wrote: post #455
93po wrote: post #428

Earlier quoted context omitted.

> very close to passing the Turing Test but fundamentally don't register to people as "People." I'm only today learning about intentionality, but the premise here seems to be that our current AI systems see a cat with their camera eyeballs and don't have the human-level experience of mentally opening a wikipedia article in our brain titled "Cat" that includes a split-second consideration of all our memories, thoughts…

Semantic search using embeddings seems like the missing puzzle piece here to me. We can already generate embeddings for both text and images. The vision subsystem generates an embedding when it sees a cat, which the memory subsystem uses to query the database for the N nearest entries. They are all about cats. Then we feed all those database entries - summarized if necessary - along with the context of the conversati…

I don't really understand the brain or AI enough to meaningfully discuss this, but I would wonder if there's some aspect of "intentionality" in the context of the Chinese Room where semantic search with embeddings still "doesn't count".

I struggle with the Chinese Room argument in general because he's effectively comparing a person in a room following instructions (not the room as a whole or the instructions filed in the room, but the person executing the instructions) to the human brain. But this seems like a crappy analogy because the better comparison would be that the person in the room is the electricity that connects neurons (instructions filed in cabinets). Clearly electricity also has no understanding of the things it facilitates. The processor AI runs on also has no understanding of its calculations. The intelligence is the structure by which these calculations are made, which could theoretically could be modeled on paper across trillions of file cabinets.

As a fun paper napkin exercise, if it took a human 1 second to execute the instructions of the equivalent of a neuron firing, a 5 second process of hearing, processing, and responding to a short sentence would take 135,000 years.

Re: Facebook LLAMA is being openly distributed via torrents

#519

Post by sangnoir »

krrrh wrote: post #385
KMag wrote: post #332

Earlier quoted context omitted.

On the flip side of subhuman treatment of humans, we have useful legal fictions like corporate personhood. It's going to be pretty rough for a while, particularly for nontechnical judges, to sort all of this out. We're almost definitely going to see multiple rulings far more bizarre than Citizens United ruling that limiting corporate donations limits the free-speech rights of the corporation as a person. I'm not a la…

Good points but it’s worth clarifying that this is not what the Citizens United decision said. It clarified that the state couldn’t decide that the political speech of some corporations (Hillary the Movie produced by Citizens United) was illegal speech and speech from another corporation (Farenheit 9/11 by Dog Eat Dog films and Miramax) was allowed. Understood this way it seems obvious on free speech grounds, and in…

> It’s clear that people don’t and shouldn’t lose their free speech rights when they come together in a group

Should Russian (or Dutch) citizens who incorporate in America have the same free speech rights as Billy Bob in Kentucky? As in can the corporate person send millions in political ads and donations even when controlled by foreigners?

Re: Facebook LLAMA is being openly distributed via torrents

#520

Post by tmalsburg2 »

ok123456 wrote: post #252

Maybe this is an intentional leak to damage OpenAI. A supposedly better model by some accounts that strikes right at the heart of their business plan of selling access for $250k/year. One month of access to their service could buy a machine capable of running this leaked model. Facebook nerfs a potential upstart competitor to keep current big-tech cartel stable. Maybe this is a bit conspiratorial, but we live in the…

Why leak it instead of just publishing it along with a press release about openness and democratizing AI and so on?
Post reply on HN