AI - The Prompt, part 2
This is the second of my posts on the Prompt. You can find the first here. One of the things that people can learn to do right now to take advantage of AI applications like chatGPT is to learn to write a prompt. In this post I focus on chatGPT but the points I make are valid for any large language model (LLM).
For those who don't know, an AI prompt is the question or instruction that you give to an AI like chatGPT or Dalle-3 to tell it what to do or respond to. It's similar to asking someone a question and waiting for their answer. For example,
"What is the capital of France? write 500 words describing the capital of France from the point of view of a lost tourist."
That is a prompt. Not a very inspiring prompt, but a prompt all the same. The prompt guides the AI on what information it should provide or what task it should perform. The quality and clarity of the prompt are important because they help the AI understand exactly what is being asked, leading to better and more relevant responses. In simple terms, a prompt is how we, the humans, communicate with AI to get specific information or responses.
This is really important, so I will say it again. A well-constructed prompt is crucial for effective AI performance, it determines the context, scope, and direction of the AI's response. The better the prompt the more chance of ensuring accuracy & relevance in the answer you get. Good prompts enable AI to demonstrate its capabilities fully, whether in generating text, image, or video, providing information, solving problems, or creating content.
Prompts are important, very important.
If your prompt is poor, the answer may be poor, worse still, it may be an hallucination, this is were the AI makes something up. Many of us when we first start out are not very good at writing prompts. It is not as easy as it first appears. Asking some odd random thing like the example above may be easy enough, but is of little help if I work for a telco, a law firm, the oil industry, retail, automotive, health, manufacturing or any other industry.
Many coders now use chatGPT to help the write code and many struggle because they do not ask the right prompt so the code they get is not always that useful. This was one of the problems I discovered early on and it took me a couple of days to really get the hang of asking the right question and sometimes, asking the same question again, and again, but worded differently.
Just an aside note here, chatGPT and similar GUI based applications now have a basic memory, a context. As you continue to prompt the conversation builds on previous interactions and previous prompts. Sometimes this can be useful because you do not have to explain everything each time you write a prompt. On the other hand, if your prompt has taken you a wrong direction it can be easier to reset the conversation, using something like "Let's change the topic", "Start a new conversation", or open up a new chatGPT session.
For a very quick win, if you want your people to start making some use of chatGPT or any large language model (LLM) for your company, give them some training on how to structure and ask a prompt. A great prompt really can be a game changer. This will give you a massive bang for your buck at the moment.
OK, but what if you don't have the time to teach all your people how to write a great prompt, what if some of them are like me, developers, renowned for our in ability to put sentences together? What then?
Is there a way to create the equivalent of a super prompt that can help your staff? Well, it turns out there is, sort of. Here are two things that chatGPT enables you to improve prompting. This can also be applied to any LLM, although at the moment it is easier in chatGPT. Using chatGPT, you can now create your own bespoke chatGPT. This is a chatGPT where you have given it instructions that it should apply to the user's prompt when crafting a response to a question.
The second technique is to use an assistant. Strictly speaking, chatGPT is an assistant, but it turns out that you can also have an assistant to the assistant. You can create your own assistant that you can tune in a similar fashion to the bespoke chatGPT I mentioned in the previous paragraph.
I am going to wrap this one up here because I don't want my posts to get to long. I will go into more detail on the bespoke chatGPT and the bespoke assistant in the next post, including the similarities and differences.
The takeaway from this one is that the easiest way to get better usage out of chatGPT or any other LLM is to learn to create high-quality prompts. This act alone will significantly improve your interaction with chatGPT.
Bryan
As usual, this is written by me, a human, so any shiteness is down to me and not the AI.