Limitations of ChatGPT and How to Make It Work for You

ChatGPT is a great Al platform that has taken the world by storm. And that's no wonder since it brought Al to the "masses" for the first time. As a result, professionals from various fields use it to become more productive and do new things. Some obvious ones are, for example, marketers using it to generate unique content ideas. Other examples include teachers using ChatGPT to create better learning objectives for new subjects.

Yet, amongst all this hype, we must be well aware of its limitations. ChatGPT and AI systems in general, as Stuart Russell pointed out in an interview with Miriam Vogel on the "In Al We Trust" podcast by the World Economic Forum , are very good at regurgitating information, but that doesn't make them real "thinkers". In my personal experience, ChatGPT is a good example of that. Most of the time, it does its job well. However, sometimes it can produce an impressively inadequate answer, while at other times, the answer is average at best.

Thus, it's a good idea to understand the limits of ChatGPT's capabilities and to develop and follow a process to counter them. Here are some of these limitations:

1. It's no replacement for domain knowledge - you cannot use it to generate a list of treatments for a patient with depression if you are not a mental health professional. While it will produce the list, you need to be able to evaluate the options given in it and select or discard them as needed.
2. It's a black box - you don't know precisely how the answers are produced. Without going into detail on how ChatGPT works, we need to be aware that a degree of unpredictability in its output always exists since it uses probabilistic methods. Consequently, the same question run several times can produce different results.
3. It has been trained on data until 2021 - ChatGPT doesn't know about events, materials, etc., that appeared after 2021. So, as a result, if you want ChatGPT to base its answer on something that occurred after 2021, it cannot. So, for example, you cannot expect a high-level data privacy action plan to reflect legislation enacted in 2022 and 2023.

Notwithstanding all of the above limitations, using ChatGPT is a great way to become more productive and free the time to focus on higher-order, more creative activities. Remember that it is a tool, not a replacement for human reasoning.

Here are some helpful tips:

• Provide clear and concise prompts
• Continually review and, when necessary, modify the output. I find it helpful to run the result through Grammarly for official documents.
• If you don't get a suitable answer the first time, re-run the same prompt multiple times.

Would you be interested in specific topics you would like me to cover? Please feel free to drop me a direct message or leave a comment.

Finally, share this newsletter with your network if you find it helpful!