4 ms·

Importantly, non-commercial. Almost all of Facebooks stuff used to be Apache, this new stance is really shitty of them and I hope limits adoption. Deigning to a

Importantly, non-commercial. Almost all of Facebooks stuff used to be Apache, this new stance is really shitty of them and I hope limits adoption. Deigning to allow others to play with models (and make improvements, give feedback, build an ecosystem) that only you can profit from is not good community behavior. I'd rather see them make it freemium or paid if that's their goal, this is the equivalent of a kid licking a cookie so the others can't eat it.

sangnoir 3y ago
> Almost all of Facebooks stuff used to be Apache, this new stance is really shitty of them and I hope limits adoption The AI research environment has changed from the earlier default-open publication - unlike it's competitors, FAIR is still releasing model weights instead of serving the models behind an API. > this is the equivalent of a kid licking a cookie so the others can't eat it. More like the other kid baking a cookie with the words "Free Cookie" on it so others can eat it if they are hungry, but can't sell it for money. It'd be foolish for FAIR to donate preconfigured homing-missiles to OpenAI and others via one-way tech transfer.
- version_five 3y ago
  It'd be foolish for FAIR to donate preconfigured homing-missiles to OpenAI and others via one-way tech transfer. No, they could GPL it, and I don't think they're worried about competition taking the models anyway, there's nothing particularly special about the weights or training data, just the compute. I think part of it is pressure from AI "safety" hangers-on who pretend that AI is dangerous so only those who don't want to abide by license terms should have unfettered access. The other commercial reasons are harder to understand. With pytorch they became the standard that everyone builds off of, they could do that with their recent AI, particularly LLaMA but they chose this silly route. Also, LLaMA has a more permissive license than this translation one, and is a more powerful model, so I don't really see the "homing missiles to open AI" angle.
  - taneq 3y ago
    True, LLaMA2 is more like "donating homing missiles to everyone except OpenAI, Google, and Apple."
  - sangnoir 3y ago
    > [...] I don't think they're worried about competition taking the models anyway, there's nothing particularly special about the weights or training data, just the compute If that is the case, then what do you suppose is the reason most researcher outfits stopped releasing model weights, or offer more restrictions when they do? Using the GPL won't prevent the larger AI competitors from using your model outputs from tuning their non-public models to consistently beat yours, but a non-commercial clause does. > Also, LLaMA has a more permissive license than this translation one, and is a more powerful model, so I don't really see the "homing missiles to open AI" angle. LLaMa lags ChatGPT 4, but SeamlessM4T is ahead of WhisperX in some ways
Zuiii 3y ago
I largely ignore the licenses on this stuff and use it for commercial purposes. If they want to sue me, let them. In fact, I dare them. Their models are not copyrightable and I would not waste this opportunity to establish this fact.

sangnoir 3y ago

> Almost all of Facebooks stuff used to be Apache, this new stance is really shitty of them and I hope limits adoption The AI research environment has changed from the earlier default-open publication - unlike it's competitors, FAIR is still releasing model weights instead of serving the models behind an API. > this is the equivalent of a kid licking a cookie so the others can't eat it. More like the other kid baking a cookie with the words "Free Cookie" on it so others can eat it if they are hungry, but can't sell it for money. It'd be foolish for FAIR to donate preconfigured homing-missiles to OpenAI and others via one-way tech transfer.

version_five 3y ago
It'd be foolish for FAIR to donate preconfigured homing-missiles to OpenAI and others via one-way tech transfer. No, they could GPL it, and I don't think they're worried about competition taking the models anyway, there's nothing particularly special about the weights or training data, just the compute. I think part of it is pressure from AI "safety" hangers-on who pretend that AI is dangerous so only those who don't want to abide by license terms should have unfettered access. The other commercial reasons are harder to understand. With pytorch they became the standard that everyone builds off of, they could do that with their recent AI, particularly LLaMA but they chose this silly route. Also, LLaMA has a more permissive license than this translation one, and is a more powerful model, so I don't really see the "homing missiles to open AI" angle.
- taneq 3y ago
  True, LLaMA2 is more like "donating homing missiles to everyone except OpenAI, Google, and Apple."
- sangnoir 3y ago
  > [...] I don't think they're worried about competition taking the models anyway, there's nothing particularly special about the weights or training data, just the compute If that is the case, then what do you suppose is the reason most researcher outfits stopped releasing model weights, or offer more restrictions when they do? Using the GPL won't prevent the larger AI competitors from using your model outputs from tuning their non-public models to consistently beat yours, but a non-commercial clause does. > Also, LLaMA has a more permissive license than this translation one, and is a more powerful model, so I don't really see the "homing missiles to open AI" angle. LLaMa lags ChatGPT 4, but SeamlessM4T is ahead of WhisperX in some ways

Zuiii 3y ago

I largely ignore the licenses on this stuff and use it for commercial purposes. If they want to sue me, let them. In fact, I dare them. Their models are not copyrightable and I would not waste this opportunity to establish this fact.