This academic article explores whether smaller, compact Large Language Models (LLMs) can serve as cost-efficient alternatives to larger LLMs for meeting summarization in real-world industrial settings. The authors conducted several experiments comparing the performance of fine-tuned compact LLMs (like FLAN-T5) with zero-shot larger LLMs (such as LLaMA-2, GPT-3.5, PaLM-2). The findings indicate that most smaller LLMs, even after fine-tuning, fail to outperform larger zero-shot LLMs in meeting summarization datasets. However, FLAN-T5, a compact LLM, performed on par or even better than many larger LLMs, highlighting its potential as a cost-effective solution for real-world applications.
Publication date: 2 Feb 2024
Project Page: Not Provided
Paper: https://arxiv.org/pdf/2402.00841
