Ways To Improve Using ChatGPT For Better Results
ChatGPT is the next “Google” or tool that will shape the Internet. Whether you are doing research or just curious about learning more about something, ChatGPT can help provide answers. From making a chicken salad to complete programming code for an application, it is easy to use but are you getting the most out of it?
There are several ways to effectively interact with ChatGPT to get the best experience in your interactions. You can simply create a statement like in a conversation, or be more specific to get the answer you need.
Techniques called prompt engineering can be used, which embeds the input (i.e. prompt) that follows a methodology rather than just an explicit message. Certain parameters can also be included to further improve the results. These are ways to optimize AI-generated text prompts for more accuracy and relevancy of answers.
What Is ChatGPT?
ChatGPT is a chatbot, in its basic form. You interact with it in a conversational manner, like using a chat application. What sets it apart is an AI-generative NLP (Natural Language Processing) model engine that can respond to human requests in a natural way. It uses a very large training data set that is like the world’s database of all or most of the information known to mankind.
While Google returns a list of search results from queries, ChatGPT returns a summary of the information instead. The generated result appear like they were written by a person, which makes it a remarkable tool for learning and education.
While it is easy to interact with ChatGPT, users may not be getting all the information they need. That is because lack of detail or specificity does not unleash the full potential that ChatGPT is capable of.
Here are some ways to improve ChatGPT interaction:
Do Not Think Of It As A Robot
Although it is a form of robot (i.e. chatbot), when interacting with ChatGPT do not treat your interaction like human to robot. Use natural language when you chat, like a normal conversation.
Instead of using monotonous robot-style commands like most users do with search engines, message ChatGPT like having a conversation with another person.
Being Clear and Specific
Always (as much as possible) provide clear and specific instructions to form the context of your message. It performs best when there is a clear understanding of what answer you want it to return.
Avoid using vague prompts like “Filing process for driving license”. Use a more specific message to get a clearer answer like “What is the process for getting a driver’s license in the state of California?”. If you did not specify California, ChatGPT will return a more general answer or response that may miss important steps related to that particular state.
The less specific your question is, the more general the answer you will get. That is why it is important to add specific details in the prompt to get ChatGPT to return the best answer.
One important thing to note, do not ask questions regarding sensitive information. For example, data about personally identifiable information regarding health records or private data is not revealed by ChatGPT. For obvious reasons, these are things that are not available to the public.
Your objective is to gain knowledge from the information provided from the conversation. If the response is unclear, don’t hesitate to get clarity on the information by making a follow up question. You can request a further explanation or additional examples to get better understanding.
Use System Messages
You can use what are called system messages to instruct ChatGPT on the context of your conversation. This sets the behavior for the language model to follow.
To get into a conversation with ChatGPT that is focused on scientific topics, start the conversation with a system message like “You are an intelligent chatbot that can answer detailed questions about science.”
What the system message does is set the role for ChatGPT, and is recommended for developers who are creating assistant software that provides support for users. By focusing the role on science (per example above), ChatGPT will respond more directly to the point.
Provide A Context
As mentioned earlier, providing more context helps to obtain the information you’re seeking. This helps ChatGPT to better understand the queries in the message.
You can use as references, prior messages in the conversation. Include the relevant details or summary. ChatGPT will then use the context based on the role you want it to assume (i.e. system message) to respond more accurately with the answers required.
For example:
“Based on the events in computer history that led to the invention of the microprocessor, what was the significance of the development of the transistor?”
The context is framed in relation to computers, with reference to transistors in relation to microprocessors. ChatGPT will interpret this based on this specific question. If you do not mention transistor or microprocessor, it could return a different result.
Experiment With Iterations
The responses from ChatGPT can vary based on the input phrasing or order of words. If you’re not getting your desired result, experiment further using iterations.
Rephrase your prompt by breaking it down into smaller sentences. This will make ChatGPT address the message in parts rather than as a whole. It is like breaking down the whole of the query into smaller parts.
According to ChatGPT, iterations refers to the technique of having a back-and-forth conversation with the model by providing multiple prompts or queries in sequence.
Here is an example:
Iteration 1
User: Can you tell me about the traffic commute for today?
ChatGPT: Sure, where is your location?Iteration 2
User: I’m in Hermosa Beach, California.
ChatGPT: Which freeway and where are you driving to?Iteration 3
User: The 405 Freeway going to Los Angeles Downtown.
ChatGPT: The traffic commute from Hermosa Beach on the 405 Freeway to Los Angeles Downtown is ….
Iterations can take longer to get to a specific answer though, but it flows like a natural conversation. In the real world, this is how people usually speak to one another. With AI, by building up on the phrasing of a query it can arrive at the desired result.
Using Temperature and Max Tokens
ChatGPT uses 2 parameters that can influence responses called the temperature and max tokens. Temperature controls randomness of the output response with a range that exists from 0 to 1.
Having lower values (e.g. 0.2) produces focused and deterministic responses, while higher values (e.g. 0.8) result in more diverse and creative responses.
Here is an example:
Max tokens limits the length of the response generated by ChatGPT. For the GPT-4 language model (used by ChatGPT) the maximum number of tokens is 8192. The GPT-4-32K model increases the maximum number to 32768. The tokens can be used to represent the number of word fragments (which are converted into primitive inputs called tokens) that determines the length of a conversation.
You do not have to use these parameters at all, unless you plan on having a long conversation with many required answers. ChatGPT has default values for these parameters. If you want a response that you feel should be longer or better with more detailed information, you can modify the value of this parameter.
The problem or issue a user faces with a low max tokens value are the results could be truncated or incomplete answers. It can return only a small portion of the required information to the user. Higher values may produce the best responses, but can be more expensive (with paid subscription of ChatGPT services) and longer to process.
Let’s take a look at this example:
Software developers have more control with these parameters when used with an API. This provides functions that allow developers to call from their application.
Final Thoughts
ChatGPT is a use case for AI, that provides information much like search engines but in more detail. These are some optimization techniques to help improve interaction with ChatGPT to get the best results.
Using ChatGPT can be a digital tool for learning and enhancing your knowledge. The better your style of asking questions or phrasing conversations, the better the results you will get. Just remember, it’s always important to verify information from reliable sources and use critical thinking before accepting an answer.
Photo Banner Credit: cottonbro studio
