Tag
Maxime Rivest shares a trained model that classifies any abstract against any given taxonomy, matching Qwen 27B and Jev-level superhuman accuracy at an order of magnitude lower cost and latency, while reflecting on the specialization vs. generalization tradeoff.