Writing prompts to generate content - ChatGPT Tutorial

- [Instructor] The prompt is key. It looks like a search bar, and it's often labeled as message, since the word "prompt" is an unfamiliar term for many end users. You'll see it at the bottom of the screen. The response mirrors the prompt, good for good, bad for bad. For this reason, consider offering suggested prompts for staff and for customers to use. It's a good idea to create a prompt library for help desk staff too. Your prompt initiates a pattern for ChatGPT to complete. Think of it like offering ChatGPT a few puzzle pieces. If you give it pieces with sufficient information, ChatGPT will deliver the full picture you seek. If not, you'll get something else that may be helpful and maybe not. Before you write a prompt, decide what you want to accomplish with ChatGPT. Make a list of things ChatGPT can help with. Break those ideas into specific tasks, such as organizing, planning, strategizing, writing or evaluating code, creating content, writing formulas for spreadsheets, analyzing data, or otherwise assisting with help desk duties, processes, and operations. Prioritize and choose a task with a high business value. Go to the user interface, where you'll find a prompting bar that looks like a search bar, the one you saw just a few seconds ago. But remember that ChatGPT is not a search engine. Resist the urge to treat it like a search bar. Do not use keywords or partial questions or incomplete sentences or other search engine operation tactics. You need to talk to ChatGPT. Clearly visualize the ideal outcome and articulate that concisely in your prompt. Spend time and effort on composing the prompt. Be very specific about what you ultimately want ChatGPT to do. Your prompt must be concise to better direct ChatGPT, but also to use fewer tokens, which keeps your cost down and helps you prevent exceeding usage limits. A token includes bits of a word, spaces, punctuation, and other tidbits within a prompt or response. Some things you can see, and some only the machine can see. Whether you are using ChatGPT directly, within third-party software, or in your own business adaptation, you'll need to be conscious of how many tokens you're burning in the prompt and the response. Trim your token usage by doing one or more of the following: Be very concise in writing the prompt. Edit it as necessary if you need to before you ask for the machine to respond to. Command it to summarize or limit its response word count to keep it from rambling and eating too many tokens in the process. Or you can use prompt chaining to build upon earlier ChatGPT prompts and responses in the chain. This is kind of like a conversation with a human being. You say something, ChatGPT says something, you say something again, but keeping it on topic. The following three secrets will help you leverage ChatGPT's power, but not necessarily the token counts. Secret number one: Give ChatGPT one or more job roles in a single prompt. That will provide additional clarity, additional perspectives, details, depth, and innovative jolts in its response. Secret number two: Just because ChatGPT uses natural language processing doesn't mean you're limited to text. Not on the input side, anyway. You can incorporate math, computer codes, and eventually even images in the prompt. However, it will not include images in its outputs, at least not for the foreseeable future. For that, you need to use its sister program, DALL-E 2. Secret number three: Includes specific commands and framing in your prompts. Command examples include the words write, analyze, compose, classify, summarize, condense, translate, order, and organize. Framing. Your prompt includes things like giving it a time period, a geolocation, a dialect, a culture, or a specific set of circumstances, such as weather conditions, or retail outlet locations or groupings, or district instructions. Remember, ChatGPT is capable of far more than just searching for one brief answer.