Use LLMs to Explain Research Papers and Learning Materials

[InAnYan]

A common student workflow today is to use large language models (LLMs) to help understand academic papers or other learning materials. When a passage is unclear, students often copy parts of the paper into an AI tool and ask for an explanation. With JabRef’s integrated PDF viewer, this workflow could be supported directly inside the application.

At a high level, the idea sounds straightforward: take some part of a paper and ask an LLM to explain it. In practice, however, several non-trivial challenges arise.

• Integration with the PDF viewer: The feature should feel natural within JabRef’s existing PDF viewing experience. However, designing a good UI for selecting what exactly should be explained (text, equations, figures, or arbitrary regions) is itself a complex problem and needs careful thought.

• How content is sent to the LLM: Academic papers—especially in STEM fields—often contain a lot of mathematical notation. OCR or PDF text layers may not represent formulas correctly, since mathematical expressions are inherently two-dimensional. In such cases, sending extracted text can lose essential structure or meaning. For math-heavy content, sending images (or combining text with images) may be necessary to preserve correctness.

• Preserving definitions and symbol meaning: It is often not sufficient to send only a small excerpt of a paper to the LLM. A given paragraph may rely on symbols, variables, or definitions introduced earlier. If that prior context is missing, the LLM may incorrectly reinterpret symbols based on its general knowledge rather than the paper’s actual definitions.

• Support for follow-up questions: Explanations should not be one-off. Users need to ask follow-up questions, clarify misunderstandings, and iteratively build understanding. Maintaining conversational context is essential and arguably more important than generating a single explanation.