Add post: When AI Doctors See What Isn't There (visual grounding in medical VLMs)

[anas-zafar]

Summary

• Adds a new research post by Anas Zafar, Leema Krishna Murali, and Ashish Vashist from the Cohere Labs Open Science Community
• Post covers the paper "Beyond Accuracy: Evaluating Visual Grounding in Multimodal Medical Reasoning" (arXiv:2603.03437), published at ICLR's CAO workshop
• Introduces counterfactual evaluation metrics (VRS, IS, HVRR) showing that RLVR fine-tuning improves accuracy while degrading visual grounding

Files added

• _posts/2026-06-19-visual-grounding-medical-vlms.md — the post
• assets/bibliography/2026-06-19-visual-grounding-medical-vlms.bib — 5 BibTeX references
• assets/img/2026-06-19-visual-grounding-medical-vlms/modality-skeptic-paradox.png — figure 1
• assets/img/2026-06-19-visual-grounding-medical-vlms/results-tables.png — figure 2 (results tables)
• assets/img/authors/anas-zafar.png — author headshot

Test plan

• Run docker compose up and verify the post renders at http://localhost:8080
• Check author card and headshot appear correctly
• Check both figures render with captions
• Check citations render inline via <d-cite> and populate the appendix bibliography
• Verify animated hero header displays

[github-actions]

PR preview (updates on each push to this branch):

https://cohere-labs-community.github.io/blog/pr-preview/pr-8/