How To Stop Chatgpt From Going Off The Rails

As a passionate AI language model enthusiast, I have encountered various experiences using ChatGPT. While it is a remarkable tool for generating text that resembles human expression, there are instances where it may deviate and generate untrustworthy or unsuitable responses. In this article, I will provide tips and recommendations to effectively guide ChatGPT and avoid any unexpected outcomes.

Understanding ChatGPT’s Limitations

Before we dive into specific techniques, it’s important to acknowledge the limitations of ChatGPT. While it can generate coherent and contextually relevant responses, it lacks real-world experience and common sense. ChatGPT’s responses are purely based on patterns it has learned from the training data. This means it may sometimes provide inaccurate information or exhibit biased behavior.

Being aware of these limitations is crucial for effectively managing ChatGPT’s responses. With that said, let’s explore some strategies to keep ChatGPT on track.

1. Be Clear and Specific in Your Prompts

When interacting with ChatGPT, it’s important to be clear and specific in your prompts. Avoid ambiguous or open-ended questions that can lead to unpredictable responses. Instead, provide detailed instructions and ask direct questions to guide ChatGPT towards a desired outcome.

For example, instead of asking “What are the pros and cons of technology?” you can ask “What are three advantages and three disadvantages of using technology in healthcare?”. By specifying the context and desired format, you can increase the chances of receiving accurate and relevant responses.

2. Use System Level Instructions

OpenAI has introduced the concept of system level instructions, or “system messages”, which allow you to give high-level guidance to ChatGPT. These instructions appear before the user message and help set the behavior of the model throughout the conversation.

By using system messages, you can provide guidelines and specify the desired tone or style of the responses. For example, you can include a system message like “You are an assistant that only provides technical information” to ensure that the generated responses stay within the intended scope.

3. Experiment with Temperature and Max Tokens

Temperature and max tokens are two parameters that can influence the output of ChatGPT. Temperature controls the randomness of the generated text, with higher values (e.g., 0.8) leading to more diverse but potentially less focused responses, and lower values (e.g., 0.2) resulting in more deterministic and conservative responses.

Max tokens, on the other hand, limits the length of the response generated by ChatGPT. By setting an appropriate max tokens value, you can prevent ChatGPT from generating overly lengthy or irrelevant responses.

4. Flag Inappropriate Responses

If you encounter an inappropriate or biased response from ChatGPT, it’s important to flag it and provide feedback to OpenAI. Reporting these instances helps OpenAI improve the model and refine its behavior.

OpenAI has implemented a Flagging API that allows developers to report issues directly. By actively participating in this feedback loop, you contribute to the ongoing efforts of making ChatGPT more reliable and safer to use.


ChatGPT is an impressive tool that can provide valuable assistance in generating human-like text. However, it’s important to understand and manage its limitations to ensure reliable and responsible use. By following the strategies and best practices outlined in this article, you can guide ChatGPT effectively and prevent it from going off the rails.