Get the best out of ChatGPT with these golden clues

Welcome back to On Tech: AI, a pop-up newsletter that teaches you about artificial intelligence, how it works and how to use it.

A few months ago, my colleagues Cade Metz and Kevin Roose explained the inner workings of AI, including chatbots such as OpenAI’s ChatGPT, Microsoft’s Bing, and Google’s Bard. Now we’re back with a new mission: helping you get the most out of AI.

People from all walks of life – students, programmers, artists and accountants – are experimenting with using AI tools. Employers post jobs looking for people adept at using them. Soon, if not already, you will have the chance to use AI to streamline and improve your work and personal life.

As a personal tech columnist for The Times, I’m here to help you find out how you can safely and responsibly use these tools to improve many areas of your life.

I’m going to spend today’s newsletter discussing two general approaches that will be useful in a number of situations.

Then, over the next few weeks, I’ll give you more specific tips for different aspects of your life, including parenting and family life, work, organizing in your personal life, learning/education, creativity, and shopping.

A few common sense warnings to start:

• If you’re concerned about privacy, leave out personal information like your name and where you work. The tech companies say your data is used to train their systems, which means other people may be able to see your information.

• Do not share confidential information. Your employer may have specific guidelines or restrictions, but generally entering trade secrets or sensitive information is a very bad idea.

• Hallucinations: Chatbots are powered by a technology called a large language model, or LLM, that gains its capabilities by analyzing massive amounts of digital text from the Internet. Many things on the internet are wrong and chatbots can repeat those falsehoods. Sometimes, while trying to predict patterns from their huge training data, they can make things up.

The golden clues

ChatGPT, Bing And Bard are among the most popular AI chatbots. (To use ChatGPT you need to create an OpenAI account and need a subscription for the most advanced version. Bing requires you to use Microsoft’s Edge web browser. Bard requires a Google account.)

Although they look simple, you type something in a box and you get answers! — asking questions the wrong way produces generic, useless and sometimes downright incorrect answers.

It turns out to be an art to type in the precise words and framing to generate the most useful answers. I call these the golden clues.

The people who get the most out of the chatbots use variations of these strategies:

“Pretend.” If you start your prompt with these magic words, the bot will be instructed to impersonate an expert. For example, typing “Pretend to be a tutor for the SATs” or “Pretend to be a personal trainer” guides the bots to model themselves around people in those professions.

These prompts provide additional context for the AI ​​to generate its response. The AI ​​doesn’t actually understand what it means to be a tutor or a personal trainer. Instead, the prompt helps the AI ​​draw on specific statistical patterns in its training data.

A weak prompt without guidance produces less useful results. If all you type is “What should I eat this week?” the chatbot comes up with a generic list of meals for a balanced diet, such as stir-fried turkey with a side of colorful veggies for dinner (which sounds very “meh” to me).

“Tell me what else you need to do this.” To get more personalized results, for example health advice for your specific body type or medical condition, invite the bot to request more information.

In the personal trainer example, a prompt could be, “Pretend you are my personal trainer. Create a weekly workout regimen and meal plan for me. Tell me what else you need to do this. The bot can then ask you about your age, height, weight, dietary restrictions, and health goals to tailor a week-long meal plan and fitness routine for you.

If you don’t get any good answers on your first try, don’t give up right away. Better yet, in the words of Ethan Mollick, a professor at the Wharton School of the University of Pennsylvania, treat the bot as if it were a human trainee: “If it makes a mistake, point it out and ask to do it better.” Be forgiving and patient, and you’ll probably get better results.

Have your chatbot conversations

After you master the prompts, you can make your chatbot more useful over time. The key here is to avoid treating your chatbot like a web search and starting a new search each time. Instead, keep several conversation threads open and add more over time.

This strategy is easiest with ChatGPT. Bing requires you to reset your conversations periodically, and Bard doesn’t make it as easy to jump between conversation threads.

Natalie Choprasert, an entrepreneur in Sydney, Australia who advises companies on how to use AI, uses ChatGPT as a business coach and executive assistant. For each of these roles, she has separate conversations side by side.

For the business coach thread, she shares insights about her professional background and the company’s goals and issues. For the executive assistant thread, she shares schedule information, such as the clients she’s meeting with.

“It builds and trains well, so when I ask it a question later, it will be in context and will give me answers that are close to what I’m looking for,” Choprasert said.

She shared a bonus gold prompt that trained her assistants to be extra helpful: Apply a framework. She recently read ‘Clockwork’, a book about setting up a business. When she asked ChatGPT-the-business-coach to provide advice using “Clockwork’s” framework, she was pleased to see that it was able to incorporate principles from the book into an action plan for expanding her business.

Share your clues

What are your golden clues that have given you AI’s most impressive, helpful results? Email us your samples. We may use your submissions in future editions of this newsletter.