You are correct, I wrote that in a rush and mixed up examples in my head. I don’t think you need to trust the LLMs for this kind of thing to be very useful. Th

You are correct, I wrote that in a rush and mixed up examples in my head. I don’t think you need to trust the LLMs for this kind of thing to be very useful. The LLM could generate the KG with every node labelled as “autogenerated.” When you use the graph for research, you are still going to read the papers you are interested in so you can then update the relevant citation node with the label “human checked.” If a research group uses the same graph over time, the nodes will gradually become “trustworthy” (ie verified by humans). Maybe even get reviewers to update a papers graph during review and publish that for other groups to add to their graphs.