Policy
118d agoLlama 3.1 Scores 43% On Nigerian Livestock Eval
A 420-question benchmark on Nigerian ethnoveterinary practices reveals that Llama 3.1 8B achieves only 43% accuracy. This gap highlights a lack of indigenous knowledge in training sets compared to Western data. The researcher will next test Claude Sonnet.
The Signal
These results expose critical regional blind spots in current model knowledge bases.