And you believe the other open source models are a signal for ethics?
Don't have a dog in this fight, haven't done enough research to proclaim any LLM provider as ethical but I pretty much know the reason Meta has an open source model isn't because they're good guys.
That's probably why you don't get it, then. Facebook was the primary contributor behind Pytorch, which basically set the stage for early GPT implementations.
For all the issues you might have with Meta's social media, Facebook AI Research Labs have an excellent reputation in the industry and contributed greatly to where we are now. Same goes for Google Brain/DeepMind despite their Google's advertisement monopoly; things aren't ethically black-and-white.
A hired assassin can have an excellent reputation too. What does that have to do with ethics?
Say I'm your neighbor and I make a move on your wife, your wife tells you this. Now I'm hosting a BBQ which is free for all to come, everyone in the neighborhood cheers for me. A neighbor praises me for helping him fix his car.
Someone asks you if you're coming to the BBQ, you say to him nah.. you don't like me. They go, 'WHAT? jack_pp? He rescues dogs and helped fix my roof! How can you not like him?'
Open weights fulfill a lot of functional the properties of open source, even if not all of them. Consider the classic CIA triad - confidentiality, integrity, and availability. You can achieve all of these to a much greater degree with locally-run open weight models than you can with cloud inference providers.
We may not have the full logic introspection capabilities, the ease of modification (though you can still do some, like fine-tuning), and reproducibility that full source code offers, but open weight models bear more than a passing resemblance to the spirit of open source, even though they're not completely true to form.
Fair enough but I still prefer people would be more concrete and really call it "open weight" or similar.
With fully open source software (say under GPL3), you can theoretically change anything & you are also quite sure about the provenience of the thing.
With an open weights model you can run it, that is good - but the amount of stuff you can change is limited. It is also a big black box that could possibly hide some surprises from who ever created it that could be possibly triggered later by input.
And lastly, you don't really know what the open weight model was trained on, which can again reflect on its output, not to mention potential liabilities later on if the authors were really care free about their training set.
I use Gemma3 27b [1] daily for document analysis and image classification. While I wouldn't call it a threat it's a very useful multimodal model that'll run even on modest machines.
And you believe the other open source models are a signal for ethics?
Don't have a dog in this fight, haven't done enough research to proclaim any LLM provider as ethical but I pretty much know the reason Meta has an open source model isn't because they're good guys.
> Don't have a dog in this fight,
That's probably why you don't get it, then. Facebook was the primary contributor behind Pytorch, which basically set the stage for early GPT implementations.
For all the issues you might have with Meta's social media, Facebook AI Research Labs have an excellent reputation in the industry and contributed greatly to where we are now. Same goes for Google Brain/DeepMind despite their Google's advertisement monopoly; things aren't ethically black-and-white.
A hired assassin can have an excellent reputation too. What does that have to do with ethics?
Say I'm your neighbor and I make a move on your wife, your wife tells you this. Now I'm hosting a BBQ which is free for all to come, everyone in the neighborhood cheers for me. A neighbor praises me for helping him fix his car.
Someone asks you if you're coming to the BBQ, you say to him nah.. you don't like me. They go, 'WHAT? jack_pp? He rescues dogs and helped fix my roof! How can you not like him?'
3 replies →
The strongest signal for ethics is whether the product or company has "open" in its name.
Can those be even called open source if you can't rebuild if from the source yourself?
Even if you can rebuild it, it isn’t necessarily “open source” (see: commons clause).
As far as these model releases, I believe the term is “open weights”.
Open weights fulfill a lot of functional the properties of open source, even if not all of them. Consider the classic CIA triad - confidentiality, integrity, and availability. You can achieve all of these to a much greater degree with locally-run open weight models than you can with cloud inference providers.
We may not have the full logic introspection capabilities, the ease of modification (though you can still do some, like fine-tuning), and reproducibility that full source code offers, but open weight models bear more than a passing resemblance to the spirit of open source, even though they're not completely true to form.
Fair enough but I still prefer people would be more concrete and really call it "open weight" or similar.
With fully open source software (say under GPL3), you can theoretically change anything & you are also quite sure about the provenience of the thing.
With an open weights model you can run it, that is good - but the amount of stuff you can change is limited. It is also a big black box that could possibly hide some surprises from who ever created it that could be possibly triggered later by input.
And lastly, you don't really know what the open weight model was trained on, which can again reflect on its output, not to mention potential liabilities later on if the authors were really care free about their training set.
Are any of the models they've released useful or threats to their main models?
I use Gemma3 27b [1] daily for document analysis and image classification. While I wouldn't call it a threat it's a very useful multimodal model that'll run even on modest machines.
[1] - https://huggingface.co/google/gemma-3-27b-it
Gemma and GPT-OSS are both useful. Neither are threats to their frontier models though.
They are, at the same time I considered their model more specialized than everyone trying to make a general purpose model.
I would only use it for certain things, and I guess others are finding that useful too.