If you’re using ChatGPT, or similar services, then this is not for you.
I recently have had more than one conversation with folks from varying companies and industries who haven’t used ChatGPT yet. As I explain to them what it can do they get interested. I will attempt to introduce this thing called ChatGPT and point you to why should you use it, how to get started, and hopefully enable you to leverage a really powerful tool!
If you’re a developer, and not using ChatGPT or GitHub CoPilot, DO IT!
Let’s start with…
What can ChatGPT do?
Before we get our hands on it, let’s talk about what it can do. ChatGPT can interact with humans by responding with well-built (accurate, spelling, grammar, etc.) sentences. The computer can understand what you’re saying and respond as if you were talking to a friend.
However, this friend has an enormous amount of knowledge, it knows about virtually anything that is publically accessible. It has read books, websites, social media sites, legal documentation, Wikipedia, etcetera a thousand times over. We don’t know the full reach of its knowledge, we just know ChatGPT knows a lot, like, a lot.
Check out this chart, published by openai.com:
The colors are different versions of ChatGPT, ignore that for a moment. The lines across the bottom are different exams they had ChatGPT take, and the height is its exam results. Incredible, it can pass the Uniform Bar Exam and many others. This also highlights some areas it just doesn’t know enough about, there’s always room for improvement.
It knows all of this content and can answer you immediately, with mostly correct information (more on accuracy later). Ask it how to bake bread and how to change the oil on your car. You can search for these with Google or Bing, and then find a site, then read it, or, ask ChatGPT and just have it answer you.
You can ask it anything, with any nuanced variation, and it’ll answer you: “how can I bake bread to taste like bacon”, it may answer: “To achieve a bacon-like taste in bread, you can incorporate ingredients that mimic the flavor profile of bacon, such as smokiness and saltiness, along with a hint of sweetness. Here are a few suggestions:…”.
Pretty cool. So for now, let’s say ChatGPT knows everything. What would you ask someone who knows everything? What kind of help would you ask for? Here are some examples I’ve used and heard of others using ChatGPT for:
- Explain a topic to a 2nd grader (helps to break down complex topics for me to better understand it)
- Create a recipe with these ingredients…
- Summarize this email into bullets
- What is the proper order to watch Marvel movies
- In the book ABC, who is this character? (Halfway through a book I thought there was a new character, there wasn’t, I just forgot who it was)
- I’m getting this error in my code…
- Help me write feedback for my colleague
- Help me write a function in Python that can do XYZ
- Write a poem about Darth Vader
- Help me plan a surprise party for my wife/husband
- I’m going to Disney, what are the secrets of each park I should explore?
Some basic examples, and they can get incredibly more complex, and technical. We’re just getting started.
So you might be asking…
How does ChatGPT do what it does?
Magic.
“Any sufficiently advanced technology is indistinguishable from magic”
Arthur C. Clarke
I don’t know all the inner workings of ChatGPT: AI modeling, large language models, transformers, etc. Based on the definition above, it’s magic to me, and I’m okay with that. But here’s what I do understand.
The computer has been trained on countless amounts of data. Training a computer is like training a dog. You give it a command, and then it responds. If it does good, you praise it. If it does bad, you tell it no and try again. ChatGPT has been trained against billions of commands (its knowledge) and has received signals when it responds correctly, and incorrectly. By the time we get our hands on it, it’s so well trained that it more or less does what we expect.
What it’s really doing is guessing. ChatGPT guesses the next word in its sentence. Based on all of its knowledge, it guesses that the next word should be the word it has picked. Wild, right? Consider the sentence “the quick brown fox jumps over the lazy dog”. This sentence is in its knowledge, probably many many times. You may ask “what does a quick brown fox do?”, it’ll respond with “a quick brown fox jumps over the lazy dog”. It guessed the next word because it’s been trained on it.
No, it’s not alive, it’s just predicting the next word, with incredible accuracy.
But sometimes it’s wrong
Sometimes, because it is just guessing, it can be wrong. Sometimes it’ll just guess the wrong answer, it may say “a quick brown fox crawls under the truck”. If you think it’s wrong, you can challenge it, and ask it more questions, even a simple “are you sure?” can get it to produce a different answer.
It guesses everything, but because its knowledge is so vast, it’s normally very accurate. Its knowledge is vast but dated. The task of training the machine on content is very time consuming, so ChatGPT is not as up-to-date as possible. You can ask it “what’s your last training date” and it’ll tell you (as of now, it’s April 2023).
A big challenge is when it’s wrong, ChatGPT thinks it’s right! Like your friend who knows everything, it’ll tell you the wrong answer, confidently. It won’t tell you “I don’t know” unless you ask a super obscure question or you explicitly tell it “Tell me ‘I don’t know’ when you don’t know.” (even then, it’ll still try to tell you a correct answer). Because of this, it’s highly recommended to verify all of its content if you’re going to use it for work, life-changing events, publishing, etc. There are stories of ChatGPT conjuring up new legal cases, using copyrighted information, and straight-up creating false content.
In my experience it’s reliable. When I use it for serious work, and the work I do with my clients, I verify anything it gives me. In my personal life, if I’m asking about a recipe, a new medical condition the doctor on the TV says I have, or help me rewrite an email. It’s usually spot on and little to no verification is needed.
Okay, are you ready? Let’s do this.
Get started with ChatGPT
Go to chat.openai.com and log in. It’s free.
Start chatting! After you log in you’ll see a chat box at the bottom, start typing and get going!
Have fun with it, it is incredibly powerful.
Assign ChatGPT a personality
Creepy? Yes. But it works. Tell it what you want it to act like:
- You’re a senior developer…
- You’re a marketing specialist…
- You’re a CEO at an XYZ company…
- You’re a dad of a teenager…
Try “you’re a jr high math teacher, explain derivative calculus to me” and “you’re an ivy league professor, explain derivative calculus to me”
Tell ChatGPT who you are
You can also tell it about yourself and how you want it to respond. This can help it think about how best to respond to you:
- I’m a senior developer, I only want code, do not explain…
- I am a home chef aspiring to be an executive chef, tell me…
- I manage a team of engineers at a retail company…
- I have owned this house for 11 years…
- Tell me in 2 sentences what…
- Speak like Shakespeare and tell me…
Try “I’m a 4th grader researching the Boston tea party. Please provide a basic outline for my report on the subject” and “I’m a junior in college researching the Boston tea party. Please provide a basic outline for my report on the subject”. Dramatically different requests based on who you are.
Provide an example
If you want it to create something in a specific format, like an outline, or a comma-separated list of values to insert into Excel, or whatever, you can provide it an example of what you’d like, and it’ll follow that. For example:
“Provide me a list of Marvel movies in order of release date. Format it as a CSV like: movie name, date released, box office numbers” and it will spit that out for you. This example may not have all the movies, again, training dates, and all.
You can provide it any format you’d like if you provide a clear example. Sometimes you may have to add a little more data to your example if there’s complexity.
Give ChatGPT specific content
Because of the training date, ChatGPT doesn’t have the latest and greatest content. If you use ChatGPT via bing.com, or pay for the premium from OpenAI.com, it will perform web searches and retrieve newer content for you.
If a web search doesn’t cover it because your data isn’t public, you can still use it with GPT.
IMPORTANT content you put into GPT could be used to train it later on, meaning the content you put in would be available to others. OpenAI has an option to not use your content. I don’t think Bing.com does yet. I don’t recommend putting any sensitive content, company content, client content, etc. into ChatGPT without fully understanding the risks. Review their privacy policy and ask your legal and risk teams if using company content.
You can copy/paste any content into the chat box. If you are paying for OpenAI, you can upload images and files, and it’ll read it all. With that content there you can ask questions about it, ask it to summarize, reword, etc.
How are you going to use ChatGPT?
Go on, go use it. Head over to chat.openai.com and get started. Just talk to it. Go crazy and ask it some weird stuff. Ask it for ideas on what it can do. There are limitations of the service, things it can’t talk about, can you find them?
I have found the hardest thing with ChatGPT is the limitations of our own creativity. I have shared a very limited list of things to try. The real magic is when we start thinking more creatively about what ChatGPT can do for us. It’s not alive, it doesn’t want your job. I believe ChatGPT and other AI technologies can greatly supplement our lives by taking away mundane and tedious tasks so we can use our ingenuity, creativity, and curiosity to do even greater things (things AI can’t do, yet).
I hope you found this helpful and has piqued your interest in using ChatGPT. This is just the tip of the iceberg. There is so much we can do with AI, across every aspect of our work and personal lives.
Subscribe to my blog and get posts like this in your inbox. Share your email below, or follow me on Threads, LinkedIn, or BlueSky.
David Lozzi 