It’s not particularly surprising - I assume a lot of this is motivated by the fact that people building language models are training them on data they don’t have the rights to, but that doesn’t mean we should encourage it, or come close to defining it as open 🙃