AI-Prompting is an approach to elicit solutions or answers from so-called Large Language Model AI systems. What these systems do is basically to construct a chain of output tokens (words or, say, programming language statements) according to their pre-trained sequencing probabilities in accordance to the input question. Thus, the results these systems produce are highly sensitive to the question or ‚prompt’ that triggers the generation of the output. This is easiest to see, I think, when you use different languages in the prompt: A question asked in German will result in an answer in German, and, as the sequencing rules for German are based on a different training corpus than those of e..g. English, the contents of the output will differ from the output to a prompt in English as well. Further, it cannot be stated often enough: These LLM systems are not Dialogue systems. They do not ‚understand‘ tasks and therefore they are not able identify which information is lacking to solve the problem posed and request it from the user. So, if your want such a system to deliver a solution to a specific task, the instruction for problem solving needs to cover all relevant information in such a way that permits the LLM system to do its best. So, here it is good if the human side is capable of Dialogue, i.e. understanding the AI’s output, evaluating it with respect to the intended task, and, if necessary, reformulate the prompt to achieve a more appropriate result, perhaps also comparing the results for different language inputs.
Thirdly, as these LLM-AI systems are opened (via e.g. by a chatbot-like interface ChatGPT to the GPT-3 family of LLMs) to the general public for (somewhat) supervised learning, their behaviour changes over time rather quickly. So, you need to keep tracking the development of these systems.
And yes, when Google releases its ‚Taskbot‘ based on ‚LaMDA’, things may take a whole new turn again!
Below, I give you examples above from ChatGPT.

• I know what the problem is with LLM-AIs and what you need to get better results
• I have 30+ years of experience in AI, in particular in all aspects of Dialogue systems (cf. my publications and patents), as well as many contacts in the industry
• I am fluent in English, German and French, and have decent knowledge in some other languages (e.g. Dutch, Italian, Danish, …)
• I have a large body of common knowledge to judge the facticity and credibility of LLM output