Onderzoek toont: AI-modellen verwerken 16 keer meer informatie zonder kwaliteitsverlies

Researchers have demonstrated a practical method for compressing language model inputs by up to 16 times while maintaining accuracy in real-world applications. The breakthrough addresses a major challenge where accumulated tokens from documents and conversation history consume excessive memory and computing resources during extended agent operations.