5 Key Prompting Lessons I Wish I Knew Earlier

No matter your experience level, ChatGPT is so  deep that you can always get more out of your   prompts. That's right, today we'll be looking  at five mistakes you make when using ChatGPT,   and more importantly, I'll show you how  to fix them. So if you've been struggling   with getting the most out of your ChatGPT  experience, this video is for you. Oh, for me,   all of these I consider fundamental to ChatGPT and  GPT-4 usage. So no matter your experience level,   think of them like a checklist to consider with  every prompt you run. Alright, let's get started.

The first one is the most common mistake with  beginners. If I were to oversimplify what a prompt   is, it comes down to two unavoidable elements:  instructions and context. ChatGPT really works   best when users provide sufficient context  for their questions. So let's specify this:   What is context? Context is providing background  information or explaining the situation so that   ChatGPT can better understand your question. It's  taking out all the guessing work. The more context   and detail you give, the better the responses you  will get. If your question is vague or ambiguous,   ChatGPT, no matter if GPT-3. 5 or GPT-4, may not be  able to provide an accurate or relevant response.   Here's a real-life example. A user asks ChatGPT  for workout tips but doesn't specify their fitness   level, goals, or any physical limitations. Well,  obviously, ChatGPT will do what it can. It will   provide you with a generic workout routine  that doesn't align with your actual needs,   such as targeting specific muscle groups or  addressing some pre-existing injuries. So,   what do we do here to get the most relevant  responses? Always make sure to provide clear   and detailed information when asking questions or  making requests. It could be as simple as giving   your instructions and then saying, "Now consider  all the details of the context. " So for example,   if your first prompt is, "Write me a tweet  about penguins," you can always follow up with,   "Now consider that my Twitter posts are typically  photography-related memes," and it adjusts its   output to the context you just provided. Another very basic framework to start with   here would be simply separating the two, saying  instruction and context. So this is really one   of the most straightforward frameworks on how to  get started with prompts That include multiple   building blocks. There are many, but these  two are always the first ones to consider. So,   if this structure becomes really easy to reuse,  this template on whatever your instructions or   context might be. Okay, on to mistake number two. A lot of people overestimate ChatGPT's

capabilities. Look, the reality is most people  come in with sky-high expectations because of   all the hype, and especially with GPT-4, they  expect some sort of magic bullet. What really   happens is that your personal assistant ChatGPT  just increased its IQ by 50 points, but   that doesn't mean it's going to be able to turn  water into wine. And especially for beginners,   I find this visualization of pretending that  ChatGPT is your newly hired assistant extremely   helpful. She has every imaginable university  degree and read the most books that exist, and   her sole task is to help you out. Help me. But no  matter how great the assistant, if your requests   are not crafted with laser-like precision, you  won't get all the results you're looking for.   If we take our "you just hired the helpful  assistant" analogy a little further, don't   expect to get perfect results on her first day of  work. Give it time, and if something doesn't work,   communicate it. The beautiful thing about this  little analogy is that you get to leverage all   the social skills every single one of us has when  using ChatGPT. If you're talking to a human and   you have some sort of question, well, you're gonna  ask. And while the "imagine ChatGPT is a helpful   assistant" framework is great, you still have to  remain realistic about ChatGPT's capabilities.    Keep in mind that it is an AI language model,  and while it can provide valuable information,   it may not always have the answers you're looking  for. Especially if you are looking for a very   specific answer and all you're doing is throwing  one-line prompts at it. No, if you're looking for   very specific answers, you will need to get longer  prompts most of the time. And most importantly,   when it comes to facts and data, always,  always double-check your answers or consult   experts when needed. Even the most advanced AI  can't replace human expertise. Not yet, anyway.

Okay, the next mistake is quite simple. It's  ignoring ChatGPT's training data limitations. So,   this one is all about remembering that ChatGPT's  knowledge is up to date only until September   2021. This is relevant for both GPT-3. 5 and  GPT-4. So, keeping in mind September 2021,   that's where its data ends, is an absolutely  essential piece of information when using ChatGPT.

Alright, and the next mistake is something I don't  see people talk about enough. It's neglecting   user privacy issues. You know what's super  important when using the internet? Protecting   your privacy. Why would that change with AI? If  anything, you should be even more careful. It's   easy to get carried away and share personal  details, but that's not the best idea. We're   still in like day one of the AI revolution,  and no matter how useful these tools are,   you have to realize that eventually, once the  competition really heats up, these companies   will use every advantage they can get, and  all the data we're typing into it on a daily   basis now might just be utilized in ways that we  wouldn't agree with right now. Think about it,   all companies that come out with large language  models, they are all kind of similar. Trained on   the wider internet, they utilize a Transformer,  and then the outputs are rated by humans,   and the model is readjusted. One of the biggest  opportunities for these companies is going to be   getting better data, and chances are that your  data is more valuable than you think. So, it's   your duty to protect it. Never share sensitive  information like your financial details or social   security number with AI. A good rule of thumb is,  if you wouldn't share it with a helpful stranger,   don't share it with AI. Protect your privacy. Okay, and this is mistake number five, and it's

probably the biggest one. It's not asking the  right questions. "Where is Gamora? Who's Gamora?    Why is Gamora? " And I know you've heard about  this before, but hear me out, this runs really   deep. So, the biggest problem is if people are  not specific enough with their question, ChatGPT   is an incredibly powerful language model, but it  can't read your mind. If you ask vague questions   to get big answers, but how do you know if you're  asking the wrong question? Look, it's actually   very simple. If you're not satisfied with the  result, you're probably asking the wrong question.   Sometimes ChatGPT may not understand your  question on the first try. To fix these mistakes,   experiment by altering or extending your prompt to  improve the output. You can also consider adding   even more context as previously discussed in this  video. You might already know that the easiest way   to avoid this one is having quality prompts that  you can copy and paste, just like the ones that I   provide in my free ebook. But for many users, that  won't be enough. They have specific use cases,   and they want to squeeze the maximum out of every  prompt. And for those people, I'm creating a   ChatGPT course that dives deep into every single  point in this video and about 20 more topics,   including advanced prompts and step-by-step  guides to crafting world-class prompts.   Look, it's really simple. If you like these  videos, you will love the course. Developing   the course really allowed me to lean back, think  through everything related to prompt engineering,   and go step by step. Anyway, if you just  keep the five points that I mentioned here   in mind whenever you run a prompt inside of  ChatGPT, you'll instantly get more out of   your ChatGPT experience. And if you want  a bunch of prompts to copy paste, then   this is the video for you. I sincerely hope you  learned something today, and I'll see you around.