@UT_Hill_Lab

[@UT_Hill_Lab @UT_Hill_Lab]

@s OpenAI @AnthropicAI @GoogleDeepMind @GeminiApp — I hope next year I can tell students: "Yes, use these tools safely." Right now? Time-consuming verification required.
#AIinResearch #CitationHallucination #BiomedicalAI

[@UT_Hill_Lab @UT_Hill_Lab]

I use ChatGPT 5.2 in my biomedical writing — and it's been unexpectedly reliable at generating real, relevant citations.
Hearing lots of hype about other models. So I tested the same citation task across multiple LLMs (max thinking, web, $20/mo tier).
The variability was shocking

[@UT_Hill_Lab @UT_Hill_Lab]

Goal to push for fixes so we can trust models for high-stakes work (grants, manuscripts, reviews).

Needed:
Transparent retrieval traces
Hard constraints (PMID/DOI validation)
Native "citation needed" support (no forced filling)
Built-in claim-evidence alignment checks