Search papers, labs, and topics across Lattice.
Affiliation:
3
0
4
5
Retrieval pre-training significantly boosts typed decision model performance for biomedical QA, especially when retaining retrieval priors, but its effectiveness hinges critically on the chosen decision head architecture and training objective.
Fine-tuned LLMs excel in generating evidence for biomedical claims, but PubMed retrieval can mislead on broader public health topics.
Narrative Anchoring reveals that language models can misdiagnose identical clinical facts based solely on their sociolinguistic presentation, leading to significant diagnostic variability.
