Ah yes, for dense problems like that I wouldn't expect it to work well. The example graphs in the submission were mostly quite sparse, hence why I thought of LLMs. But perhaps that was just for illustrative purposes.