Rodrigo Agerri

Fine-tuning smaller LLMs on in-domain data can rival the performance of larger, proprietary models when used as multilingual judges, challenging the assumption that bigger is always better for text evaluation.

Turns out, the best external knowledge source for multilingual medical QA depends on whether you're working with a high- or low-resource language, and blindly adding PubMed might not be the answer.

LLMs struggle with physical commonsense reasoning in low-resource languages like Basque, especially when presented with dialectal variations, revealing a significant limitation in their real-world applicability.