I think one interesting lesson from this nice qualification here is that at the moment ML methods to learn mathematics may look trivial from a professional mathematician (ie the results are unoriginal or trivial) but perhaps the target audience of this method may be for non professional mathematicians or students training to be mathematicians. I could still see this ML tool as a way to automate the work of some more “trivial” (from the POV of an expert) mathematics, although not the work of professional mathematicians.

The knowledge gap in mathematics between professional mathematicians and non professionals is vast, and this tool could narrow the gap.

I would bet the majority of readers of nature would not be able to point out that the outputs of the ML tool were trivial. So there is need to narrow this gap.