Skip to Main Content

Generative Artificial Intelligence (GenAI): GenAI Prompts

Text generated by XIPU AI (GPT 4)

XIPU AI:

Writing good prompts for chatbots is a combination of clear, precise instruction, and understanding the bot's capabilities. Here are some tips:

  1. Be Specific: The more specific you are with your prompt, the better the bot can provide you with the answer you're looking for. For instance, instead of asking "What's the weather?", ask "What's the weather today in New York City?".

  2. Use Full Sentences: While chatbots can understand and respond to short prompts or keywords, full sentences typically provide more context and can lead to more accurate responses.

  3. Request One Thing at a Time: Bots can get confused with too many requests at once. So instead of saying “Find me a Mexican restaurant in Los Angeles that’s open now and has vegetarian options”, break it down. Start with “Find me a Mexican restaurant in Los Angeles.”

  4. Avoid Ambiguity: If a statement can be understood in more than one way, the bot might not respond the way you wanted. So try to make your prompts as unambiguous as possible.

  5. Know Your Bot's Abilities: Different bots have different purposes and capabilities. Some are good at answering questions, some are good at setting up appointments, and others excel at finding information online. Keep this in mind when crafting your prompts.

  6. Experiment and Refine: Not every attempt will be successful on the first try. Learn from the responses, adjust your prompts, and experiment to see what works best.

Remember, the goal is to communicate clearly and effectively with the AI to get the best possible response. With these strategies, you should be able to craft better prompts for any chatbot.

* Text generated by XIPU AI.   Try asking these questions in XIPU AI.

How to Write Prompts

How to Write Prompts

Prompts are the input content provided to the AI tool, and guide the AI to generate the desired content. Therefore, creating prompts is particularly crucial as it directly influences the quality of the generated content. Here are Six strategies for getting better results by OpenAI Prompt engineering Guide about writing prompts: 

  • Write clear instructions

Models can't read your mind. For concise outputs, request brief replies. For more complex responses, ask for expert-level writing. If you're dissatisfied with the format, illustrate the format you prefer. The clearer your instructions, the more likely you'll receive the desired output.

  • Provide reference text

Language models may invent inaccurate responses, especially for obscure topics or when asked for citations and URLs. Similar to how notes aid a student in a test, offering reference text helps these models generate responses with fewer fabrications.

  • Split complex tasks into simpler subtasks

Complex tasks tend to have higher error rates, while breaking them into simpler subtasks improves reliability. Complex tasks can often be redefined as workflows, where outputs from earlier tasks construct inputs for subsequent ones.

  • Give the model time to "think"

Much like humans needing time to calculate, models make fewer reasoning errors when given time to process. Requesting a "chain of thought" before an answer can help the model reason its way to correct responses more reliably.

  • Use external tools

Address model limitations by supplementing it with outputs from other tools. For instance, a text retrieval system (such as RAG or retrieval augmented generation) can provide information from relevant documents. A code execution engine like OpenAI's Code Interpreter assists the model in math and code execution. If a tool can handle a task more reliably or efficiently than a language model, offload it to achieve optimal results.

  • Test changes systematically

Enhancing performance is facilitated by measurable metrics. In some cases, prompt modifications may improve performance on specific instances but lead to worse overall results. To ensure a change has a positive net impact, defining a comprehensive test suite (also known as an "eval") may be necessary.

Additional  tactics and prompt examples can also be found on the OpenAI guide page referenced above.

However, it's quite difficult to generate the intended output in a single try, especially for the image and video content. So Keep conversing and asking for changes. Ask AI tools to revise the answer in various ways. Here are some intriguing examples of prompts on GitHub and various websites.

So, keep practicing and start writing your prompts in XIPU AI.