Apertus: The AI Model That Actually Respects Your Privacy

Switzerland just dropped Arpetus, the AI model that actually respects your privacy. Now, I'm stoked about this. I love this. It's an open- source model that really is just I mean, if you don't know anything about Switzerland, they've got all the banks that all the multi-billionaires store all their money in. So, they know how to deal with, you know, privacy. They're pretty true to their word here. So, I believe them when they say it. It's an open-source multilingual large language model. It's privacy focused. And what I love about it is that it's trained across thousand languages. Now, we spoke a few weeks ago about uh you know what languages to train on and this and that and whatever. So, that's quite cool. 40% of the train data is non-English including under representative languages such as Swiss, German, and Roman. Of course, it is because it's all pretty much the same. That's insane. 40%. Yeah. So, it's going to be really 8 billion and 70 billion parameters. So, seems reasonable. I mean like I wonder if the 8 billion one is something that can that could probably run on something really lightweight quite quite a fast model and I love that it's European I mean Switzerland in Europe Shenen I don't know I think it's kind of it's in this funny spot where it's like it is and it isn't the respect and allow shen activity but they're not actually part of shenen I think or something like that they do their own thing they do their own thing theoretically you could run any model on server and just keep the data to yourself or have a company like you tomorrow you could raise a billion dollars and just rent some warehouse and theoretically run any open-source model there and respect your users privacy. So it's it's essentially the choice of the provider. It's not about the open source model. It's who provides who's running the AI model. It's up to them to respect your privacy or not. I guess so. this is this is a whole world or when I hear open source I just think well it's not I run the model it's not being sent off anywhere um and whatever how I'd read that am I am I if I was going to be you know uh if I was going to use your data and I was going to wrap a product in this open source tool what am I doing am I intercepting your questions and the answers and then just taking that and storing it somewhere in some sort of logs I honestly don't know like I could wrap this and call it any I could call it something else but secretly under the hood be using this model and no one can do anything about that's saying that there's probably needs to be you need to look at licensing maybe you have to say we're using arbitus or appetitus I don't know how you say it but but they are making their develop their entire development process public publicly accessible so they are properly going all in on like look we're so we're open about what we do you can trust us trust and arpetus is Latin for open I think I read so Okay, that's There you go. You learned you learned Latin today. That is that's what you come here for. You learn Latin. This is part of a larger conversation on my show, Command AI, which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter.