In a "clean room" style implementation, as well. We've looked at the outputs, wrote a spec, and built from there.

In that sense, I guess at least LLMs are not infringeing on the kind of patents that would get the owners of AI labs hit by lightning.