Distilling Examples into Task Instructions: Enhanced In-Context Learning for Real-World B2B Conversations
Researchers have developed a new method to improve in-context learning (ICL) for classifying complex B2B conversations. Their approach, demonstrated on the new Call Playbook dataset, distills verbose examples into concise representations of classification criteria and task descriptions. This significantly reduces token usage by 99% and boosts macro-averaged AUC by up to 7% compared to traditional ICL, while maintaining robustness with increasing context length. AI
IMPACT This research offers a more efficient and transparent approach to applying NLP in specialized business contexts.