chatgpt prompts ai llm

This is an article for people allergic to hype. I get it. As soon as something gets big every last wannabe influencer and side hustle guru pops up with a million social media posts about it.

In the case of ChatGPT and similar apps we’ve been completely inundated with posts about “500 prompts” and explainer videos that all say the same thing. Most of it is fluff. And even very well meaning and honest people are confused.

So here are some no-nonsense suggestions about getting started using interactive LLMs like ChatGPT, minus all the marketing speak, exaggeration, and just incorrect information.

What Do I Know?

Yeah, everyone’s an expert. There are so few real experts on this that a lot of people are successfully acting like one.

I started using ChatGPT the day it was released. I’m a web developer and was immediately charmed by it’s ability to spit out code. I spent the next month doing projects that would previously have been a bit over my head. I made a text to speech drum machine. I made a simple video game. I made a fractal viewer.

And I started using GPT every day on the job. I told my boss I was using it and that they could expect me to work faster now. Embrace the future. No going back.

And so in roughly five months I’ve used ChatGPT for maybe 300 hours. I’ve also built a custom version for the educational project which is the focus on this site. I’m no expert. But I also know the thing.

Just Use It

The most important thing I have to say here is that people should just go use it. The whole entire point is that you can simply talk to it. You can ask it about itself. Ask it how it works. Ask it what it’s good and not good at. Ask it about unique things it’s able to accomplish. Just ask.

I find ChatGPT really fun to use and kind of charming because it’s a pretty good chat bot. It’s been trained to be courteous. It’s been trained to try to help. That means that you really don’t need instructions. You don’t need a guide or explainer because that’s what it already is.

What It Can’t Do

Interactive large language models are not really artificial intelligence. They are not conscious in any way. They do not understand things or have the ability to reason. In essence they are really advanced and well trained auto-complete.

That can result in some amazingly human-like conversations. But don’t be fooled. They just spit out the most appropriate words based on their training data.

So don’t ask ChatGPT about feelings. Don’t ask it for life advice. Don’t ask it to solve ethical dilemmas or questions about politics. It’s really just an interactive encyclopedia with the ability to correlate.

Also, most of these applications don’t have access to the internet. And even those that are a part of search engines like Bing Chat only work along side it and don’t actually have the ability to find content online.

In addition, most of these language models are trained on data that is not entirely current. For that reason most LLM chat bots can’t answer questions about things that are more recent.

The Only Prompts That Really Matter

Please ignore just about everything you see related to prompting. “Prompt engineering” is something for doctors of linguistics and computer science. Generally speaking, you can just ask LLMs for what you want in your own words.

That being said, ChatGPT (and the rest) are able to do specific tasks that you should know about. Here’s a non-comprehensive list of things you might want it to do for you … by just asking.

  • Generate or edit text (creative, marketing, educational)
  • Summarize text or concepts for your level of education
  • Generate lists (top 10, reading, instructions, creative)
  • Provide code (web, Excel, Shopify, AWS, Unity)
  • Create tables and charts
  • Translate language

So then for example you could ask GPT for “a chart of some of the best marketing action words by demographic.” Here is the result.

Demographic Action Words
Young adults (ages 18-25) trendy, cool, vibrant, fresh, innovative, edgy, fun, bold
Parents (ages 30-50) reliable, trustworthy, practical, family-friendly, easy, effective, convenient, affordable
Seniors (ages 65+) safe, secure, comfortable, easy, healthy, dependable, quality, affordable
Women empowering, beautiful, sophisticated, elegant, nurturing, graceful, radiant, confident
Men powerful, strong, rugged, dependable, masculine, adventurous, bold, confident
High-income earners luxurious, exclusive, premium, elite, sophisticated, top-tier, high-end, prestigious
Budget-conscious shoppers affordable, bargain, cheap, value, discount, low-cost, cost-effective, economical

Or you could ask it to give you “five sentences to sell a ceiling fan.” Here is the result.

  • “Stay cool and comfortable all summer long with our energy-efficient ceiling fan, which features multiple speed settings and a sleek, modern design.”
  • “Beat the heat without breaking the bank with our affordable ceiling fan, which boasts powerful yet quiet operation and easy installation.”
  • “Transform your living space with our stylish and functional ceiling fan, which provides optimal air circulation and complements any decor.”
  • “Create a cozy and inviting atmosphere in any room with our versatile ceiling fan, which comes in a range of finishes and sizes to suit your needs.”
  • “Experience the ultimate in convenience and comfort with our remote-controlled ceiling fan, which allows you to adjust speed and lighting from the comfort of your couch or bed.”

You could ask it to give you “the Javascript to add a class of ‘button’ to every html button” in your code. Here is the result.

const buttons = document.querySelectorAll("button");
for (let i = 0; i < buttons.length; i++) {

The point here is that I didn’t go read about any of those examples. I just thought of them and asked for what I wanted in my natural language. That being said, it’s helpful to know about the basic capabilities that are possible.

Chat Memory

One thing I didn’t understand for the first couple months that really confused me is that LLMs like ChatGPT can only remember a certain amount on any one chat thread. And the more you feed it or the more it spits out, the quicker that memory runs out.

Basically, they can only keep so many words in mind before they start ‘forgetting‘ things. And they can only handle so many words of input in the first place. In the future you’ll be able to feed an LLM an entire book to summarize or rewrite. But that time is not now.

The takeaway here is that if you find that the chat bot starts to get confused in any way, if it starts getting off topic or gives you answers you find incorrect, start a new chat. Using these tools well means starting new chats often.

You can feed it text from another chat to get moving in the right direction again. But unfortunately, these tools just aren’t able to do a lengthy process of development around a single idea. For example, they aren’t able to help you build long blocks of code.

The workaround is to work on long text or code in smaller pieces.

Incorrect Answers

The last thing that you need to know is how to deal with wrong answers. If you spend any time at all using LLMs you will quickly realize that they can be wrong.

Now, actually they aren’t often wrong about basic facts, unless you are on a long chat and they somehow get confused. Again, start a new chat. But these tools are fallible. For that reason they aren’t really appropriate for giving medical advice and making decision that could greatly impact your health or livelihood. If the stakes are high around any task, you should search online to verify facts.

Also, ask again. If you ask an LLM the same question many times, you will likely get slightly different answers each time. So if an answer is wrong, ask again and you will often get the right one. It can also be useful to ask many times in order to choose the response you like the most. This isn’t a bug but a feature.

And keep in mind that the more complex the topic and the more logic needed to answer a question, the more likely it is that a response will be wrong.

So for example, it will never be wrong if you ask it for state capitols. But if you ask it about the historical ramifications of the Civil War, you might get something less than valid.

Have Fun

Finally, I’ve found this tech entirely captivating at times. It can give you very surprising answers and can take you in directions you hadn’t anticipated. Test its boundaries. See for yourself what it can do and how it acts. Don’t take what it says too seriously and you will find it quite entertaining to use.