Congratulations! You’re about to get a new intern! She may be named Microsoft Copilot, OpenAI, ChatGPT, or Google Bard. Whatever her name, she’s a large language model that can make you more efficient and creative.
Rather than thinking of this new assistant as just another tool, I’ve found it useful to think of her as a virtual intern. She’s a Large Language Model (LLM) built from neural networks that simulate the human brain. This makes her very different from other virtual assistants like Siri or Alexa. Rather than just providing you with the answers to questions, she can collaborate with you and help you develop your ideas.
But she’s different from your typical intern. She’s been trained on the entire internet. She’s read everything from Shakespeare to Seinfeld to quantum mechanics, and she’s lightning fast at pulling all this stuff together. And while she’s surprisingly creative, she has a habit of making stuff up. Overall, she’s very diligent but a bit quirky which makes her difficult to manage. To help you manage her more effectively, I’ve put together these five tips:
1. Set clear objectives
She’s a great researcher and she’s very educated, but she needs to be pointed in the right direction. You should do this for any employee, but your virtual intern takes it to another level. Tell her to do something, and she will execute the hell out of it without a fuss—whether or not it makes sense. She has no clue what to prioritize. If you tell her to “Explain quantum mechanics as if you were a dog,” she will come back with “Quantum mechanics is when you can be in two places at once, like when you chase your own tail and catch it.” It’s important to remember that, despite her name, she’s not actually “intelligent” like a person. People often get confused by the words “Artificial Intelligence,” and think that a computer program can “just figure it out for them” for them without any guidance. She’s more like a GPS that can help you get where you’re going more effectively.
2. Take responsibility for her work
I don’t mean that you should plagiarize her work. But once she gives you an answer, you need to make sure it’s the right solution to your problem. This can be tricky because she writes in a very compelling fashion with lots of confidence, even when she’s wrong. Remember, she’s trained on the entire internet. But before you share her ideas with others, make sure you think them through and check that they apply to your situation. Today, many institutions rely on AI to make decisions for them. They rely on AI systems today to make lending decisions, hiring decisions, and parole decisions. Just like with a human intern, you own that decision and can’t pass off responsibility to someone else. “Don’t blame me, the AI said so” is not an acceptable answer.
3. Check her work carefully
She works really hard and bullshits really well. She does both of these things literally a billion times better than a human. Instead of saying “I don’t know,” she may create long expositions that are made up of whole cloth. These may even include references to fictitious websites, articles, or even legal cases that are completely made up. Recently, one of her fellow virtual interns worked for Roberto Mata when he sued the airline Avianca. His ChatGPT intern cited a half dozen imaginary cases. Some people call it AI hallucinations, but the interns prefer to call it creative writing.
4. Learn to communicate better with her
Like any new super smart intern, she thinks she knows everything and doesn’t know when to ask for help. Therefore it’s useful to guide her with prompts and examples of how to do things. As an example, virtual interns are notoriously bad at math. This is because she likes to jump to conclusions based on what she’s seen before. When I asked her, “What’s (7 * 4) – (16 * 2)?” she told me the answer was -12, which is wrong. The right answer is -4. But it is the right answer to “What’s (5 * 4) – (16 * 2)?” but in math, close isn’t good enough. Remember, she doesn’t think the same way a person does. But when I change the prompt slightly to “What’s (7 * 4) – (16 * 2)? Go step by step,” she provides a flawless answer and shows all the intermediate steps. There’s a whole new field called prompt engineering to help communicate with her.
5. Brainstorm lots of ideas
Use her as a brainstorming partner, not a ghostwriter. She’s great at generating ideas, so use her explore different angles, perspectives, and examples. But remember, while she comes up with a lot of good ideas, she also comes up with a lot of bad ones. When asking for her to write something, don’t be proscriptive and rely on just one answer. Ask her for a few different options. Ask her questions like, “Give me 10 ideas of how AI can be used to help me at work.” You’ll get some good ones, like “Automate repetitive and tedious tasks such as data entry, scheduling, invoicing, etc.” but also some bad ones like, “You can use AI to impersonate your coworkers and spy on them.” I’m just kidding. For that one I asked her for 10 “bad” ways to use AI and picked the worst one.
Just try it
Congratulations! Now you’re ready to work with your new intern. Don’t be scared. The best way to master this new world is to experiment. She’s not real person so she won’t actual intern so she won’t mind if you send her on crazy missions or try out new things. t’s a great way to test your ideas and see what she can do. You can play with ChatGPT or Bing’s AI Chat for free. It’s also really fun to see what she can do. For example, I’ve found that she can have the cast of Seinfeld explain difficult concepts. Here’s how she explains LLMs through a Seinfeld script.
Remember to think of her as a partner rather than someone to fetch your coffee. She can help come up with new and innovative solutions that you’ve never thought of. Leverage her correctly and she will surprise you, inspire you, and challenge you.
Explore, Create, and Have fun!
You can see more of my experiments and writing on LLMs on my website. If you’d like to know how you can partner with large language models to get things done, check out Impromptu (especially the first 50 pages) by Reid Hoffman, one of the founders of OpenAI, who co-authored the book with ChatGPT. If you want something more basic, check out John Maeda’s book How to Speak Machine. If you want something more hardcore, read through Stephen Wolfram’s essays What is ChatGPT and How does it Work and Will AIs Take All Our Jobs and End Human History—or Not? Well, It’s Complicated… To learn more about prompt engineering, watch a The Past and Future of Prompt Engineering by Riley Goodside, the world’s most famous prompt engineer.