4 common mistake when using ChatGPT and other LLMs

Most people make 4 basic mistakes when using ChatGPT and other LLM

The good news is fixing them is easy once you are aware

I’ll walkthrough them and show how you can fix them to produce much better results

Let’s dive in

Providing insufficient context

ChatGPT generates responses based on the context provided by the user. Without sufficient context, the responses can be off-topic or nonsensical.

An actionable tip is to always provide enough background for the AI to understand the topic. For instance, if you're asking for book recommendations, specify the genre you're interested in.

It can be difficult to provide context on topics that you don't know much about. In that case, ask ChatGPT to ask you the context it needs to complete the task. For example

Ask me as many question as you need to complete the task.

When in doubt, provide more context

Using ambiguous language

Using ambiguous language is a common pitfall when using ChatGPT. Ambiguity can confuse ChatGPT, yielding unclear responses.Be specific in your queries.

For example: Use delimiter to indicate different parts of the prompt and clearly identity expected output

Read the following customer review and output the 
sentiment of the review. 

Reviews are delimited by """

Sentiments can be one of positive, negative, neutral

"""
<review>
"""

Remember, clarity > clever

Not setting the temperature and max tokens

Most people haven't even heard of the temperature settings of ChatGPT

Temperature affect the randomness of the output. Here is how

Low temperature from 0.1-0.4 creates expected, conservative output. Use this if you want ChatGPT to follow instructions closely and provide precise answer

High temperature 1.0-2.0 generate creative, diverse and unexpected output. You may want to use this for brainstorming, copy writing for social media

You cannot set this setting in the prompt. However, you can use OpenAI playground

• Go to https://platform.openai.com/playground?mode=chat
• Login with your OpenAI account
• You can enter the system prompt or leave it as is
• Select a model and temperature settings are on the right
  

I asked ChatGPT to write the headline for this post. Here are what it comes up with at different temperature

Low temperature: 0.1

ChatGPT: Avoid These 4 Common Mistakes for Optimal AI Conversations

High temperature: 2.0

"The Pitfalls to Dodge While Harnessing ChatGPT:
Read These 4 Barriers You'll Leap Over Towards
Stellar Machine Conversations"

Adjust these settings according to your needs. Proper adjustment optimizes ChatGPT's performance.

Expecting human-level understanding

As powerful as it is, LLM and ChatGPT in particular don't possess human-level comprehension.

• Keep your questions simple.
• Break down complex tasks into smaller sub tasks
• Give the model time to think by prompting Let's do this step by step. Research has shown adding this simple instruction improve result by a whopping 70%

Keep these 4 things in mind the next time you use ChatGPT and you'll be surprised by what it can do