Artificial Intelligence Research Reveals Efficiency Gains in Natural Language Processing

Strategic Insights -

Researchers at Queen's University Belfast have discovered a framework that significantly improves the efficiency of optimizing prompts for classification tasks in large language models. According to the study, the PO2G (prompt optimization with two gradients) framework demonstrates up to 89% accuracy after just three iterations, outperforming a comparable framework, ProTeGi, which requires six iterations to achieve a similar level of accuracy. This breakthrough has important implications for the development of artificial intelligence, particularly in the context of natural language processing.

Key Takeaways:

  • The PO2G framework was developed to improve the efficiency of optimizing prompts for classification tasks in large language models.
  • The framework demonstrated an accuracy of almost 89% after just three iterations, outperforming a comparable framework, ProTeGi, which required six iterations to achieve a similar level of accuracy.
  • The PO2G framework was evaluated on a benchmark of nine NLP tasks, including three tasks from the original ProTeGi study and six non-domain-specific tasks.
  • The framework also demonstrated effectiveness in the legal-domain classification tasks, providing broader insights into the efficiency and effectiveness of prompt optimization frameworks for classification across diverse NLP scenarios.
  • The research was supported by Invest NI and the European Regional Development Fund.
  • The study involved a team of researchers from Queen's University Belfast, led by Anthony Jethro Lieander, and co-authored by Hui Wang and Karen Rafferty.

Statistics:

  • The PO2G framework achieved an accuracy of almost 89% after just three iterations.
  • ProTeGi required six iterations to achieve a comparable level of accuracy.
  • The framework was evaluated on a benchmark of nine NLP tasks, including three tasks from the original ProTeGi study and six non-domain-specific tasks.
  • The framework demonstrated effectiveness in the legal-domain classification tasks, achieving a similar level of accuracy to ProTeGi without the need for manual prompt optimization.
  • The study was published in the journal AI, Volume 6, Issue 8, 2025.

Sources:

  • "Prompt Optimization with Two Gradients for Classification in Large Language Models." AI, 2025, 6(8):182. doi: https://doi.org/10.3390/ai6080182
  • NewsRx. New Artificial Intelligence Data Have Been Reported by Researchers at Queen's University Belfast (Prompt Optimization with Two Gradients for Classification in Large Language Models). Robotics & Machine Learning. September 8, 2025; p 185.