awesome, congrats! happy to see this out :)

Thanks! I'm not questioning the method, I'm just not clear on why you use topo nets instead of regular cnns. Your topo nets consist of re-arranged cnn units anyway, if I understand correctly? So you could apply the exact same method with cnns? Maybe I just need to dive into the weeds of the paper :)

Looks great! Did you try the same using CNNs? I don't get what the Topographic nature of the models buys you in this setup?

Very exciting, I like the idea of ai sovereignty for switzerland. Do you know if there will be any senior long term job openings at apertus in the future?

Thanks for the reply! I get all of this, but it feels like the background conditions will end up playing a more important role than reality monitoring. I like the RM idea, but it seems more like one piece of a complex puzzle than a standalone explanation. Still I think it's a good first step.

(But I agree: excellent paper!)