Large Language Models (LLMs) like ChatGPT are powerful, but their performance hinges on the instructions they receive. Crafting effective prompts can be time-consuming and tricky. What if we could automate this process and unlock even greater LLM potential? New research introduces ACING, an innovative approach that uses reinforcement learning to optimize instructions automatically. Imagine an AI agent learning to write the perfect prompts, navigating the complex landscape of language to find the instructions that yield the best results. ACING frames this challenge as a continuous-action reinforcement learning problem, treating instruction optimization like an agent exploring a vast, unknown territory. It leverages an actor-critic model, where the 'actor' proposes new instructions and the 'critic' evaluates their effectiveness. This dynamic duo learns from the LLM's responses, refining the instructions iteratively. The results are impressive. Tested on 30 diverse tasks, ACING consistently outperforms existing methods, boosting performance by a median of 10 percentage points. Remarkably, it even surpasses human-written prompts, achieving up to a 39 percentage point improvement in some cases. This means ACING can discover instructions that are not only more effective but also potentially more creative and insightful than those crafted by humans. While ACING primarily focuses on instruction optimization, its implications are far-reaching. This research opens doors to more efficient LLM usage, reducing the time and cost associated with manual prompt engineering. It also hints at the possibility of LLMs learning to communicate more effectively with us, paving the way for more intuitive and seamless human-AI interaction. However, challenges remain. ACING relies on a white-box LLM to translate continuous prompts into discrete ones, a dependency that future research could aim to eliminate. Overfitting to validation data is another concern, especially with smaller datasets. Despite these limitations, ACING represents a significant step towards unlocking the full potential of LLMs, promising a future where AI can truly understand and respond to our needs.