ENHANCING DIVERSITY AND NOVELTY IN TEXT GENERATION VIA MULTI-VIEW EMBEDDINGS

Large Language Models (LLMs) demonstrate remarkable proficiency in generating accurate and fluent text. However, they often struggle with diversity and novelty, leading to repetitive or overly deterministic responses. These limitations stem from constraints in training data, including gaps in specific knowledge domains, outdated information, and an over-reliance on textual sources. Such shortcomings reduce their effectiveness in tasks requiring creativity, multi-perspective reasoning, and exploratory thinking. To address this challenge, we introduce multi-view embeddings, a novel approach that enriches input prompts with diverse perspectives derived from both textual and visual sources. By incorporating additional contextual information, this method enhances the variety and creativity of generated outputs. Importantly, our approach is model-agnostic, requiring no architectural modifications and being compatible with both open-source and proprietary LLMs.Furthermore, we propose a comprehensive evaluation framework that simultaneously measures diversity, novelty, and correctness—a first-of-its-kind methodology for assessing these three crucial aspects of LLM-generated content. We evaluate our method and framework on over 469,000 generated outputs from various well-known LLMs, demonstrating significant improvements in output diversity and novelty while maintaining quality and relevance. Our approach provides a scalable and practical solution for enhancing LLM performance across a wide range of applications, including brainstorming, creative writing, and multiple-choice question generation.