10 things I learnt from using ChatGPT
I remember my first interaction with a search engine circa 1995. It was a revelation. Suddenly I no longer needed to know or guess internet addresses that may contain the information I needed. I could describe the information I needed and the search engine presented me with a list of relevant sites. Internet search became my own personal librarian! However, I had to learn how to create good search queries, how to use different search engines, and discern between good and bad results.
In November 2022, ChatGPT burst on to the scene. It's responses are almost magical. At first I experimented with it to create future "AI for Identity" features, but it did not take long to realise that this was so much more than a technology for building better products. It is a new kind of productivity tool.
Just like the emergence of search required new skills to make the most of it, we have to learn new skills to make the most of generative AIs like ChatGPT (or at the very least revisit existing skills). As I experimented with ChatGPT, I was constantly amazed at how my colleagues are using it. Don't underestimate the creativity of humans!
So what did I learn from my own and others' ChatGPT explorations?
- Question quality matters: In computer science there is a saying "garbage in, garbage out". This is true for interacting with ChatGPT as well. If you ask vague questions, the quality of the answer may disappoint. If your questions are clear, purposeful and specific, the answers become more useful. Luckily ChatGPT is patient, so you can iterate and practice your questioning skills.
- Treat the answer as an opinion: Treat your conversation with ChatGPT much like you would a random conversation with a stranger in a public space. The information may or may not be true, it may have elements of truth, but before accepting it, double check it against sources you trust. If your using ChatGPT as your private tutor, stick to things that are easy to verify, like running a bit of sample code if your learning a new programming language.
- Try interactive brainstorming: Use ChatGPT to explore ideas, much like you would with your colleagues, or better still, along with your colleagues. Ask it how it would solve a problem, design a system or even to describe a problem and discuss the answer and brainstorm the next question with your colleagues. It may be dead wrong, but if your brainstorming, even wrong answers are interesting.
- Bad answers are interesting signals: ChatGPT is trained on data about things that already exist. If you are brainstorming with it about a problem that is well understood and have known answers, you will get good quality answers. When you get bad answers, it may be that your questions are bad, but it may also be an indicator that you stumbled onto something new. Explore further!
- Beware the echo chamber: When your interacting with ChatGPT, especially in a longer session, it may start re-enforcing your point of view. To counter that, ask it to present the counter points. What are the downside of the idea you are exploring? Why wouldn't it work?
- Your personal editor and summarizer: ChatGPT is very good at generating or improving text. As a non-native English speaker, I find it useful to help me refine text and suggest alternatives. It is also a great tool to overcome "blank page" syndrome. It is effective at providing summaries, either of supplied text, or about subjects that I need a quick refresher or briefing. Remember to treat it as an (untrusted) opinion and always validate or review the content.
- Practice full disclosure: I adopted a "full disclosure" approach to content generated with the help of ChatGPT or other generative AIs. Disclosing the use of ChatGPT to your audience may be unsettling to both you and them. You may get questions like "so what do you really think?", or "what part of this idea is yours?". Unsettling or not, I am choosing transparency, and hope other will as well.
- The buck stops with you: ChatGPT and its future incarnations can provide you with recommendations, summaries, or help spark an idea. It is still up to you to evaluate them, apply critical thinking and exercise good judgement on what you do with the information you received.
- Look behind the curtain: AI is a new technology with known failure modes, so we should be cautious, but also remain optimistic that it will continue to evolve and improve. Understanding more about how they work is also helpful in understanding their limitations.
- Ask others how they are using it: I continue to be amazed at the ingenuous ways in which my colleagues and friends are using ChatGPT. Share what your doing and whether it works or not with others and listen to how they are using it. This is a great time to adopt a growth mindset!
I continue to be amazed at the uses people are finding for generative AIs like ChatGPT, both in work, and for leisure. There are many excellent ideas of how to use them. We now have our own assistants, editors, tutors and a new kind of collaborator in addition to our own personal librarians. Making the most of these AIs will require new skills and a new mindset as we use it to augment our own abilities. I look forward to learning more about them and would love to hear about your experiences with this intriguing new tool. What are you doing with generative AI and what have you learnt?
Generative AI Disclosure:
- The ten points above came form observations and discussions between me, my friends and my colleagues after using ChatGPT in different ways.
- The image at the top of the post was generated by Dall-E from OpenAI.
- I used ChatGPT, in much the same way that I would have used a search engine, worked with an editor to refine text, or bounced ideas of a colleague.
Standards & Technology Diplomat at Microsoft Privacy and Regulatory Affairs / Corporate Standards Group
1yAll 10 advices make total sense to me. I would add "Be creative and try pushing ChatGPT to its limits and see what happens".
Helpful article. Like all tools, ChatGPT requires some skill to correctly - and for the right problem.
//we all depend on each other
1y#5 is especially interesting, thanks for sharing this Pieter Kasselman