Chatting with AI: Effective Communication Tips

              AI is growing in relevance every day in every field. But to effectively use a chatbot like ChatGPT-4, you must be able to competently converse with it. We need to learn how these chatbots want to receive our input and refine our instructions. AI and Machine Learning (ML) employ neural networks. Neural networks are “a machine learning program, or model, that makes decisions in a manner similar to the human brain, by using processes that mimic the way biological neurons work together to identify phenomena, weigh options and arrive at conclusions” (IBM).

What is an Artificial Neural Network (ANN)?

              “ANNs have an input layer and output layer. Between these two layers there are other hidden layers that perform the mathematical computations that help determine the decision or action the machine should take. Ultimately, these hidden layers are in place to transform the input data into something the output unit can use.” (7wData)

              Neurons in neural networks contain a value that ‘lights up’ or activates when it reaches a specified threshold. Neural networks work in layers, moving to the next layer when neurons in the first layer activate. For neural networks to be able to analyze and recognize audio, images or text, it needs to be able to break down these layers. The way AI does this is through a linear regression model. In these equations, importance (weights) and the threshold (bias) determine activation through the layers. The way neural networks ‘learn’ is by finding the right weights and biases.

What can Artificial Neural Networks Do?

              With this basic knowledge of how neural networks function, let’s apply this knowledge to talking with AI Chatbots. We will be working with ChatGPT-4 in this blog. The three best tips to effectively using an AI Chatbot are using simple language, providing context, and chatting with the chatbot.

Simple Language

              Even though ChatGPT can give surprisingly human-like responses, it thrives on simplicity. It may even feel as if we are having a conversation with a person, but the reality is that ChatGPT is only as human as it has been trained to be. When conversing with ChatGPT, it’s important to remember that it is an algorithm that does not have common sense or the ability to ‘think’ beyond the data it has been given. We need to use simple, but specific and clear language to effectively use chatbots. Chatbots can only ‘think’ in logical, algorithmic ways. It can’t understand things such as sarcasm.

Chatbots are designed to analyze the data they have received and provide outputs. ChatGPT-4 is more human-like, and therefore more successful, than ChatGPT-3 because it has been trained on language more effectively. Future versions of ChatGPT will likely be able to understand more nuanced topics if they are trained on them.

Using simple language does not necessarily mean dumbing down the language. A better way to think about simplicity is ‘more restrictive.’ Since chatbots run on algorithms, restrictions help narrow down their options for output. For example, if you ask ChatGPT to write a speech, it would be important to include specific and relevant details so it produces a better speech. Like a chef, ChatGPT needs the correct ingredients to make the right dish. Be specific about what you want in a response; what you don’t provide, you likely won’t get back.

Providing Context

              LLMs may be human-like, but they are still a long way from understanding real life complexities the way a human can. As we’ve said before, everything a chatbot outputs is from its training data. With more complex and high-quality training data, the responses will be more accurate and relevant. It is important to note that ChatGPT-4 does not have a live connection to the internet. It works off the knowledge base gained from a past training version. GPT4 was trained on Microsoft Azure AI supercomputers.

              How do we provide high-quality input to a chatbot session? We do this by using specific and simple language to keep the conversation on track. Providing relevant background information can help narrow the chats responses. Giving the chatbot samples of your previous writing can also help it understand the style and tone you are aiming for.

Back Out of Derailed Conversations

              An important point to understand when working with LLMs is that one bad response can ruin the entire conversation. If the chatbot gives an illogical or strange response, it’s prudent to delete that response and try guiding it in a different way. If you allow the bot to carry on with an odd response, it will learn off that response and continue down a nonsensical conversation. If you don’t reverse course, you will be down the rabbit hole like Alice in Alice in Wonderland (1865) by Lewis Carroll. Providing context is an important step to getting the answers you want from a chatbot. The more complex the input data, the higher quality the output.

Chatting with the Chatbot

              The final step to effectively communicating with a chatbot is to simply talk with it as if you are talking with a colleague. While the chatbot is an algorithm, it’s special because it can hold a conversation. In traditional search engines, you would need to figure out what was going wrong on your own. But the chatbot can tell you when and why it is confused. It’s important to listen to the chatbot and take its suggestions and questions seriously.

              The two easiest ways to fix a problem the chatbot brings up is to look at the simplicity and context. Usually, those two areas are where the AI chatbot may be getting stuck. Here is a guide to reviewing and troubleshooting providing prompts to a chatbot:

• Add more restrictions such as the style, the intended audience, or if you want the bot to go deeper into a specific topic.
• Clarify preexisting restrictions.
• Provide more purpose, context and details.
• Break down the input into smaller sections.
• Narrow down the desired output.

Even though chatbots are far from the complexity of humans, they are trained to talk like a human. Don’t be afraid to ask the bot questions or converse with it in a way you would with a human. It can provide reasons for specific outputs or pinpoint what it doesn’t understand about your input. Don’t guess what is going wrong, just ask! For instance, ask it why it gave a particular response.

The ability to chat with AI is an incredibly useful skill when you know how to wield it. Even though AI is continuously being updated and trained to be more complex, it is not always accurate. It’s important to check the bots’ responses and question it when it’s wrong. Avoid letting it continue to train on bad data. Use this powerful tool to help support your work and remember to add restrictions to inputs, provide high-quality context and drill down into specifics.