DeepMind's latest development, Optimization by PROmpting (OPRO), allows large language models (LLMs) like ChatGPT to optimize their own prompts and improve their accuracy. While conventional prompt engineering methods are reliant on a human's understanding of language, OPRO employs LLMs' inherent ability to process language and detect patterns that are not apparent to humans. OPRO begins with a 'meta-prompt', a natural language description of a task paired with example problems and solutions. The LLM continues to generate and refine solutions based on the meta-prompt, with the process repeating until no more improved solutions can be found. DeepMind conducted tests of OPRO on mathematical problems like linear regression and the travelling salesman problem. The method showed impressive outcomes and demonstrated its real potential for optimizing the use of LLMs like ChatGPT and PaLM. OPRO guides LLMs to optimize their prompts, discovering the prompt that delivers the most accurate results for a specific task. It can, for example, solve math word problems by generating sets of different optimization prompts that maximize accuracy of responses. From the way a problem is configured to the optimal prompt, it can improve the performance of LLMs significantly. Despite the absence of a released code for OPRO by DeepMind, its intuitive nature allows custom implementations in a short period of time. In fact, step-by-step guides exist that demonstrate how to use OPRO to enhance an LLM's performance for a specific task. This technique represents a significant stride in utilizing the full potential of large language models and allows users to tailor LLM's prompts to increase accuracy in performance.