←back to thread

Meta's open AI hardware vision

(engineering.fb.com)
240 points GavCo | 5 comments | | HN request time: 0.754s | source
Show context
Gee101 ◴[] No.41851793[source]
Zuckerberg and Facebook gets a lot of hate but at least they invest a lot into engineering and open source.
replies(6): >>41852156 #>>41852273 #>>41852671 #>>41852972 #>>41853469 #>>41855820 #
1. diggan ◴[] No.41852972[source]
> and open source

I'm kind of split about this. Yes, Facebook done a lot of great Open Source in the past, and I'm sure they'll do more great Open Source in the future.

But it's really hard to see them in a positive light when they keep misleading people about Llama, and publish blog posts that say how important Open Source is etc etc, then refuse to actually release Llama as Open Source, refuse to elaborate on why they see it as Open Source while no one else does it and refuse to take a step back and understand how the FOSS community feels when they actively mislead people like this.

replies(2): >>41854421 #>>41858500 #
2. lolinder ◴[] No.41854421[source]
What a lot of people complain about with Llama is the fact that the weights are open but not the training data and training code. That feels like a red herring to me—code is data and data is code, and we shouldn't require someone to be developing entirely in the open in order for the output to be open source.

The weights are the "preferred form of the work for making modifications to it", to quote the GPL. The rest is just the infrastructure used to produce the work.

Where "open source" is misleading with Llama is that it's restricted to companies under a certain size and has restrictions for what you can and can't do with it. That kind of restriction undermines the freedoms promised by the phrase "open source", and it's concerning to me that people have gotten so fixating on weights vs data when there's a big gap in the freedoms offered on the weights.

replies(1): >>41858168 #
3. diggan ◴[] No.41858168[source]
> The weights are the "preferred form of the work for making modifications to it", to quote the GPL. The rest is just the infrastructure used to produce the work.

I disagree with this, if you want to actually have a fundamental impact on the model, the real work and innovation goes into how the training is done, and the architecture of the model. That's the "source", and what Meta et al is currently trying to protect and keep private.

Guess why they're so adamant at keeping the training code secret?

Besides that, I agree wholeheartedly with you :)

4. a2128 ◴[] No.41858500[source]
To provide an example, the recent Llama3.2 release includes a clause in their acceptable use policy that says any individual or business located in EU has no rights to use their multimodal models. This is discrimination against persons/groups which violates most open source definitions. They even went as far as getting Huggingface to implement georestrictions so that EU users would get an error message and be unable to download the weights.

From my understanding this comes from a feud with EU privacy law because Facebook wants to train models on EU users data, but GDPR makes that complicated (see their letter at euneedsai.com). So they made this license change to "punish" the EU.

replies(1): >>41859100 #
5. diggan ◴[] No.41859100[source]
Not to mention that they are discriminating against fields of endeavors, and also requiring anyone that uses Llama in any shape or form to display "Built with Llama" prominently.

As far as I know, no one does that (https://ollama.com as an example of a platform that breaks Llama's terms and conditions) and Meta isn't enforcing it. But who knows, they might do so in the future.

Not sure how any sane person can claim Llama is Open Source after realizing these things.