I don’t think this is really all that surprising if you understand how diffusers actually work. They don’t just draw contours and shit. They apply sequential transformations across a matrix of pixels in an iterative fashion, the transformation varies each iteration based on prompt embeddings and the surrounding pixels of each respective transformation, typically starting from a seed of arbitrary static. A lot of images share overwhelmingly identical techniques, with a tiny sliver being responsible for the truly creative deltas. Diffusers effectively learn techniques so it tracks that dropping singular training data here and there doesn’t change much.

I have prosecuted plenty of trademarks. The logos and words are separate trademarks. For the word mark alone, Supergiant and Supermassive absolutely are likely to raise an initial LOC rejection in the same exact class of goods and services. It boils down to whether the similar words are likely to leave a confusingly similar market impression upon consumers. I legit think here it would be hard to argue around with a typical examining attorney–but the trademark office has a ton more lattitude in grants than the patent office, so it isn’t out of the question it could’ve been successfully argued or even granted initially.

It boils down to likelihood of confusion. I think there would be a legitimate concern here, tbh. But they could have come to an agreement when one registered their mark after the other one had already.

Yes, but they don’t make news or to the top of lemmy for very long ever. Paizo is an example of a relatively established company that welcomed unionization. I’m sure there are others.

Yes, as the other commenter said, in most jurisdictions (i think China is the only exception?), generative AI outputs (this covers more than LLMs, like diffusers and other modern image generators) are not eligible for copyright. However, while a human made arrangement of the outputs gets copyright protection, in the US at least, it is a very narrow scope of protection, literally only a direct copy/paste of large swaths, and doesn’t extend to the individual AI generated components.

That isn’t really what that article says. It says that in edge cases, remote workers (mostly in the Philippines) give additional contextual inputs to the autopilot. There are some outstanding issues with waymos, especially around getting stuck idling or infinite loops at construction sites, but really just being remotely operated is not one of them.

Thia is to settle their infringement from when they literally torrented giant chunks of their training corpus. It has nothing to do with any sort of infringement theory based on encoding learning from material but just old fashioned liability for flagrant piracy.