Prompt engineering, also known as ChatGPT-ing, is the process of effectively interacting with large language models (LLMs) like ChatGPT. One challenge is that text input has infinite possibilities, which can lead to harmful content or false information. Model-level alignment techniques aim to ensure that the model understands the difference between good and bad prompts. However, false positives and capability filters can hinder the effectiveness of these techniques. Prompt engineers play a crucial role in structuring prompts for LLMs, ensuring the desired output and compatibility with user interfaces. In most cases, apps should replace prompt boxes with buttons to provide a more constrained and robust user experience. By mapping specific utilities to buttons and menus, apps can optimize the use of LLMs based on common tasks like summarization, explanations, multiple perspectives, and contextual responses. The future of LLMs lies in interfaces that proactively anticipate user needs and minimize the cognitive load of prompt crafting. Delightful experiences with LLMs involve subverting user expectations and providing pre-cached questions and answers. The right interface can make LLMs powerful tools that augment human capabilities.
https://varunshenoy.substack.com/p/natural-language-is-an-unnatural