> How would anyone think releasing this was a good idea? Why would it not be? A model built on any consistent large data set could be interesting I’d expect.

> How would anyone think releasing this was a good idea? Why would it not be? A model built on any consistent large data set could be interesting I’d expect. But just because the data set has some semantic property, that doesn’t mean that a model built from it will necessarily generate text that had the same property (in this case, “being scientifically correct”). The real question is why would anyone think that?