Skip to content

Conversation

jamesbraza
Copy link
Collaborator

When given the first page of A Perspective on Explanations of Molecular Prediction Models, our citation_prompt would infer:

Wellawatte, Geemi P., et al. "A Perspective on Explanations of Molecular Prediction Models." *Department of Chemistry, University of Rochester*, 2025, https://doi.org/10.xxxx. Accessed 25 Oct. 2025.

Note there are two LLM hallucinations:

  1. A doi.org URL: https://doi.org/10.xxxx
  2. Accessed date: Accessed 25 Oct. 2025

We don't extract the accessed date, so that's fine, but we should correct the DOI hallucination via prompting.

@jamesbraza jamesbraza self-assigned this Jul 25, 2025
@Copilot Copilot AI review requested due to automatic review settings July 25, 2025 21:16
@jamesbraza jamesbraza added the bug Something isn't working label Jul 25, 2025
@dosubot dosubot bot added the size:XS This PR changes 0-9 lines, ignoring generated files. label Jul 25, 2025
Copy link
Contributor

@Copilot Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull Request Overview

This PR addresses LLM hallucination issues in citation generation by adding explicit instructions to prevent fabricated DOI URLs. The change modifies the citation prompt to instruct the model not to invent DOI links when they cannot be found in the source material.

  • Adds explicit instruction to avoid fabricating DOI links in citations
  • Maintains existing functionality for date handling and MLA formatting

@dosubot dosubot bot added the lgtm This PR has been approved by a maintainer label Jul 25, 2025
@jamesbraza jamesbraza merged commit 579145e into main Jul 25, 2025
6 checks passed
@jamesbraza jamesbraza deleted the better-citation-prompt branch July 25, 2025 22:28
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug Something isn't working lgtm This PR has been approved by a maintainer size:XS This PR changes 0-9 lines, ignoring generated files.
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants