Large Language Models (LLMs) excel at many things, from writing stories to translating languages. But ask them to solve a simple real-world problem like calculating how many seats you need for a party, and they often stumble. Why? LLMs are trained on text, not tasks. They struggle to extract the specific inputs and steps needed to solve practical problems. A new research project called TITAN aims to change that. TITAN enhances LLMs by adding two key reasoning stages: input extraction and step extraction. First, using a technique called 'step-back prompting,' TITAN helps the LLM identify the essential pieces of information within the problem, like the number of guests and how many friends each guest is bringing. Second, 'chain-of-thought prompting' guides the LLM to break down the problem into logical steps, like calculating the total number of people coming to the party. By combining these extracted inputs and steps, TITAN generates a script that the LLM can execute to produce the correct answer. This approach has been shown to significantly improve the accuracy of LLMs on a variety of tasks, including mathematical reasoning, finding patterns in text, and true/false questions. What's particularly exciting about TITAN is its use of zero-shot learning. Unlike previous methods, TITAN doesn’t need to be trained on specific examples for each task. This makes it a more general and adaptable solution for a wider range of real-world problems. While TITAN shows great promise, it's not a silver bullet. Like all LLMs, it still faces challenges in understanding complex scenarios and can occasionally miss subtle details. However, TITAN represents a significant leap forward in empowering LLMs to tackle everyday tasks, opening up new possibilities for AI to assist us in our daily lives.