## Optimizing LLM Prompts: A New Approach Large Language Models (LLMs) often generate relevant content but struggle to adhere to specific formal constraints. This leads to conceptually correct but procedurally flawed outputs. A new study proposes a multi-agentic workflow to address this issue, decoupling the optimization of the primary task description from the constraints. ## Multi-Agentic Workflow and Compliance Scores The approach relies on using quantitative scores as feedback to iteratively rewrite and improve prompts. This allows refining both the task description and the constraints, leading to an overall improvement in the model's deliveries. The evaluation of the method demonstrated that the revised prompts produce significantly higher compliance scores with models such as Llama 3.1 8B and Mixtral-8x 7B. ## Implications and Future Developments This new approach could have a significant impact on the development and use of LLMs, improving their ability to follow complex instructions and produce accurate results that comply with specific requirements. Further research could focus on applying this workflow to other models and tasks, also exploring integration with reinforcement learning techniques.