5 ways to get the most out of ChatGPT

I've been using ChatGPT as a secretary, advisor and expert within various fields, both in my line of work and privately. It has really improved my productivity, but most of all, it's improved joy. By delegating boring or repetitive tasks, and using ChatGPT to create a starting point, rather than start from an empty canvas, I need to force myself a lot less to get started on tasks that I'm not really super excited about.

There are a few gotchas, though. While impressive, ChatGPT is far from perfect, and in order to get the most out of it, here are some tips.

First, a couple of warnings

While OpenAI claims to not store any data after a chat is deleted, treat ChatGPT as you would someone you just met at a bar.

Secret or sensitive information

Don't enter company secrets or restricted or sensitive content into ChatGPT. And don't ask leading questions that might give a hint to those secrets.

Don't trust ChatGPT's answers

You know that person who will happily and confidently pull answers out of their behind on topics they know nothing about?

Well, that is ChatGPT. It will confidently give you answers that are completely and utterly wrong, and there is no indication of it's confidence level.

Of course, no single source should be trusted, but some sources are more reliable than others, and ChatGPT definitely falls in the less reliable camp.

That doesn't mean it's not useful. It just means that you need to very carefully validate the answers you get, and proofread any content that you had ChatGPT create for you.

ChatGPT is not responsible for what you do with the answers you get from it. You are.

Understand the limitations

Artificial Intelligence - at least in the forms we have now - isn't really Intelligence, at least not in the same way we talk about human intelligence. In fact, it could be argued that simulated intelligence would be a better term.

This includes ChatGPT as well. ChatGPT is a trained model, with very impressive capabilities, but it cannot reason in the same way as humans do.

While it might at the surface look like the capabilities are similar to the capabilities of humans, they are vastly different. And that is, in a way, precisely what makes it so useful.

Humans and AI models, such as ChatGPT, complement each other's capabilities in a lot of ways.

You or I could never read and retain as much information as ChatGPT has been trained on, and that is what it's strength are. In combination with your reasoning abilities, ChatGPT is great help in sorting out information that has the potential to be relevant, write drafts or do a lot of repetitive or boring tasks.

Without your creativity, reasoning and judgement it's not as useful. It's when you learn to combine each others strengths and bridge each others weaknesses that you can become a really strong team together.

Give context

ChatGPT has been trained on a lot of data, so it can give really great general answers to general questions, such as when some king died or which battles he fought in.

And that can be useful. But you can do that with Google and/or Wikipedia as well. Sometimes better than with ChatGPT.

But what makes ChatGPT so powerful, is that it understands context, so if you give it the right context, it can give answers that are really specific and relevant to a very narrow situation.

For example, compare these two questions:

"Give me a recommendation for a movie?"

or

"Films that I like include Pulp Fiction, actually most of Tarantino's movies and the Kevin Smith's movies. In fact, I like movies with great, witty dialog and don't mind violence or controversial topics or foul language. Can you give me a few recommendations with an explanation on why I would like them?"

While ChatGPT knows a lot about movies, it doesn't know you. So you need to provide that context. And the better the context you give ChatGPT, the better the answers will be.

And by the way, you don't need to give all the context up front. You can ask the question and then refine the answers by providing more context afterwards. In fact, just like any other chat, it can be a back and forth, where you can comment and give feedback on the answers you get such as "I've seen that movie, but I didn't like it because it was too long and nothing really happened" or "I didn't like it because I thought it was a stupid twist that Bruce Willies character was dead all along".

Ask specific questions

This is somewhat related to giving context, but by asking more specific questions, you get more specific answers.

Again, compare these two questions:

"What would be a good car for me?"

Or

"What would be a good car for a family of three in Amsterdam with a small child, where I commute twice/week, 50km/way, and take a couple of longer road trips per year, used is OK, I have a budget of €20.000?"

By asking specific questions, you indirectly give context that can help ChatGPT match the answer to your needs.

Ask follow-up questions

ChatGPT is a chat, and you should really take advantage of that. Don't just ask question after question, use ChatGPT to have a conversation, where you can narrow things down together to come up with a great answer.

You can, for instance, start with a general question, such as what car you should buy, and based on the suggestions for what you should think about when selecting a car or the suggestions for models, you can give feedback, such as "The Golf is too small", or "I like the Idea of the Mercedes, but I think it would be out of my budget. What would be good alternatives that is of similar size and feel, but cheaper and with a bigger engine?"

This way, to interact more like a brainstorm, and shoot ideas back and forth, is how we would do if we asked the same question to our friends. It might seem unnatural to do this with a computer, but it is a really great way to arrive to a conclusion, especially when you're not really sure from the beginning what it is you want or need.

Use separate chats for different purposes

Again, to help ensure that ChatGPT has the right content, it's a good idea to keep different chats for different purposes. This way, you can ensure that each chat has the right context. And by reusing these chats, the context will get better and better over time.

And perhaps more importantly, by separating different topics into different chats, you will ensure that ChatGPT only has the context that is relevant for that topic.

Too much context, especially context that is not relevant, will "confuse" ChatGPT and lower the quality of the answers.

So, working on a home improvement? Use a separate chat for this, with information about your house, the climate, the size and build of the house.

Working on a project? Use a separate chat for this as well, with the context that is relevant for the project.

Conclusion: Context is all you need.

Spot a pattern between the answers here? While "Attention is all you need" might be the key to training models like the one used by ChatGPT, "Context is all you need" is the key to getting the most out of generally trained models like ChatGPT.

So let that be your main take-away from this article: Make sure ChatGPT always has relevant context for what you want it to do for you. No more, no less, and only context that is relevant.