The Differences Between ChatGPT 3.5 and GPT-4

And that's why I would say, "I prefer this  prompt in GPT 3. 5 as opposed to GPT4. " So,   here's the thing: I see nobody talk about  ChatGPT 3. 5 is actually better in some use   cases than GPT4. Not many, but there are some,  and I think that you should know about them   because new things may be exciting, but they  may not always be the best solution. And look,   GPT4 is undoubtedly the better solution  in 9 out of 10 cases. It's more creative,   the reasoning is better, it hallucinates less,  and it's more reliable. But there are still   some areas where GPT 3. 5 performs better, and I  will tell you about the underlying lessons that   each one of these examples reveals. All in order  for you to get a more in-depth understanding of   this technology. But that presents the question:  how do I know about these use cases? There's no   OpenAI blog article, there's no Reddit post about  this. No, all of these are my very own findings,   and I derive those from updating my very own  content. If you're following the channel closely,   you'll know that I just updated my free ebook. I  tested every single prompt in there in GPT 3. 5 and   GPT4, all with the goal in mind to upgrade them  and make them even better inside of GPT4. And   in the process, we (me and my team) re-ran every  single prompt multiple times from multiple OpenAI   accounts to see how they perform in different  situations, with different context added or   subtracted from them. And we were surprised to  find that a selected few prompts did better in   3. 5. That's what this video is about. And if  you're completely new to the channel and don't   know what book I'm talking about, I'll leave a  link in the video description below. And as I   know many of you right now, I realize that a  simple ebook with some basic use cases might   be too basic for your current needs, and you  want to discover the process that goes into   creating your very own prompts. I now also offer  my Prompt Engineering Course, which is designed to   teach you exactly that so you can achieve your  goals and ultimately save time for the art of   prompt engineering for ChatGPT. And with that  being said, let's learn about the differences   between GPT 3. 5 and GPT4 on some actual examples. Prompt number one: Simulate job interviews. So,   what's the point of this prompt and how does  it differ? If you've ever been nervous about   a job interview and wondered what might the  employer ask and what should you answer, well,   fear not because with the help of ChatGPT, you  can simulate job interviews and practice your   responses to potential questions before the big  day. But now the question presents itself: What   model do I want to use to simulate this interview?   So, if we go to the old version of the free   ebook (which is not available anymore because we  upgraded it), you will find this prompt. Okay, if   you open a new thread and copy this into ChatGPT  3. 5, forget "Can you simulate a job interview for   a sales manager by asking and answering questions  as if you're a potential employer? " And then it   does it. It does exactly what we wanted it to do,  right? By asking and answering questions, we want   both of those, so it essentially gives us a script  for how a conversation like this could go. So,   you can fine-tune your expectations on what is  coming at you, And this is fantastic! Look at that,   five questions, and it's really good at assuming  the exact context that is needed here. So,

you don't even need much more context for this  to work well. And now, let's run it in GPT4.    Okay, typical AI language model disclaimer, and  then it goes ahead and runs this interview. And   I'll scroll through this so you can read all  the details by yourself if you care to do so.   But the gist of it is that GPT4 has higher  guardrails, as you might already know, which   means that it's more limited with its freedom of  thought and its ability to think outside of the   box. And it really reflects inside of this  interview scenario. And don't get me wrong,   a lot of the time when it's supposed to come  up with answers, it's so much better and even   more creative. But when you put it into  the scenario of conducting an interview,   it's just less human. I mean, look at this answer.   This is so obviously AI-generated and carefully   crafted. "I have over 10 years of experience  in the marketing industry, working in various   roles such as marketing coordinator, marketing  specialist, and eventually moving into management   positions. " Or down here, "As a result, the team  members' performance improved significantly,   and we were able to meet our project deadlines  more consistently. " It's just a little more   rigid as opposed to the 3. 5 generation, which is  a little more human, I suppose. "In this position,   I would use my experience to create effective  marketing campaigns that align with the company's   goals and objectives. " Or here, "You can really  see it. However, I'm aware of this weakness and   actively work to overcome it by keeping the  company's goals and objectives in mind at all   times. " And look, I'm aware that this point is  a little subtle. If the next one, it's going to   become more obvious. But if you really read from  this and the entire team agreed, here is that the   interviewer questions are actually better in GPT4,  and the answers of the marketing manager are less   human and more rigid. They're super fact-based and  they lack that human touch, whereas GPT 3. 5 has a   bit more of that. This is also probably the reason  why it's so damn hard to use GPT4 to get around   ChatGPT detectors because the guardrails are so  high and the structure in which it generates by   default, without us specifying any style or  giving any additional context here, it's just   quite rigid. Also, 3. 5 is much more concise with  the responses. So, what we did to remedy this in   the new version of this prompt is that we improved  the context and made it more specific. So, here it   really says, "I'm taking the role of an employee,  and you ask increasingly hard questions to screen   my competence, but only after I respond. Start by  introducing yourself. " And this one just works so   much better in GPT4, especially this aspect of it  where it's increasingly hard questions because the   point of this prompt is really preparing for job  interviews. This is going to allow you to insert   your own questions and get the maximum out of  the interaction with GPT4, as it's really damn   good at asking those questions. Alright, back on  track. Second prompt where I would prefer using   GPT 3. 5 over GPT4 is this prompt from the Payday  book, which I call "Uncover Facts. " It's really   fantastic at uncovering unique and interesting  information on any topic you can imagine. Simply   use this prompt of "Can you provide something  not many people know on the topic of... " and   then pick your subject. So let's just take the  first example here of beer, and let's run this   inside of 3. 5 and see what we get. Alright, so  on the topic of beer, it says, "One interesting   fact about beer is that there is a type of beer  called Lambic that is spontaneously fermented,   meaning that the beer is fermented using wild  yeast and bacteria that are present in the air,   instead of using a cultivated strain of yeast.   Okay, it's unique to the Belgian region. Another   interesting fact is that beer is one of the oldest  alcoholic beverages in the world, with evidence of   beer making dating back to ancient civilizations  such as the Sumerians and Egyptians. " Okay,   super good answer. I really enjoyed that. I  feel like it perfectly answered this, right?   And then, on the other hand, we run the same  thing inside of GPT4 and see what we get here.    "Certainly, many people are unaware of the ancient  Chinese beer recipe that dates back around 5000   years. " Okay, so far so good. That's really  interesting. "In 2016, researchers discovered   pottery fragments from the Miyaya archaeological  site in northern China that contain remnants of   a fermented beverage. The pottery dated back to  around 3400 BC. " Okay, so here we enter subjective   territory, but I suppose I personally found this a  little less interesting than the fact that you can   use yeast that is in the air to ferment your beer.   But alright, maybe it's just a preference thing,   and I'm not that much into Chinese history.   Let's move on to the second paragraph.   Upon analysis, the researchers found evidence of  grains such as barley, millet, and Job's tears,   along with tubers like yam and lily. Steve Job's  tears? There was Steve Job's tears in there? Okay,   no, but seriously, this is getting way too  detailed for the fact that I just asked about   something that not many people know on the  topic. Then it says that this suggested that   people of this time brewed this type of beer by  fermenting grains with tubers. Okay, honestly,   I'm not a big beer brewing guy, so that might be  the cause of me not knowing what tubers are or   what Job's tears are, or honestly, I don't even  know what barley is. So, I think it's fair to   say that subjectively, I'm getting a little lost  inside of this answer. If you look at the ending,   it says, "This discovery highlights the long  and fascinating history of beer brewing,   predating even the ancient Sumerian beer recipes  from Mesopotamia. " Cool. So, I don't know, I guess   it's a good answer, but it's just a little heavy.   It's a little too fact-paced and detailed, and   that's not really what I'm looking for with this  prompt, right? I'm just looking for something that   not many people know on the topic. And I suppose  it did give me that. Not a lot of people are gonna   know about all these details. But with 3. 5, it  just had more of that human touch. I mean, just   imagine if you recited this paragraph that 3. 5  delivered to you about how there is yeast in the   air that can ferment your beer or how it's one of  the oldest alcoholic beverages in the world with   evidence of beer making dating back to ancient  civilizations such as the Sumerians and Egyptians.    I would bet that this would interest way more  people than this long-winded response that is   super specific to the different components that  they found in the exact years when it came out.   I realized that this answer might be beneficial  in many other use cases. But if you asked me,   not in this one. I just want something that not  many people know, not a detailed history lesson.    And that's why I would say this one performs  better in 3. 5. And this goes for many other   topics, not just beers. Feel free to test it with  as many topics as you want. It always yields this   type of result. And that's why, if you look at  the updated version of this prompt for GPT4,   what we changed is the wording. Well, so if I copy  this new GPT4 optimized prompt in here and I say,   "What is an aspect of beer that is commonly  misunderstood or overlooked? " We're being way   more specific of what we want. Concretely, we're  saying "commonly misunderstood. " This implies the   context of us communicating to humans and them  misunderstanding our message. This is not "What   is some fact that not a lot of people know? "  No, we're concretely asking for something that   is misunderstood or overlooked, which implies we  want to use this for human conversations, which is   how that last problem was originally intended.   It just didn't work that way in GPT4 anymore.   So, it says, "One aspect of beer that is commonly  misunderstood or overlooked is the role of yeast   and its impact on flavor profiles. " Alright,  we're not getting a detailed history lesson.    This is great. "This is a crucial ingredient in  the brewing process. " Some scientific details.    Alright, "It also contributes significantly to the  flavor and aroma of the beer. " Now, see, this is   the type of stuff that you would actually want to  use at a dinner conversation. And in typical GPT4   fashion, it goes into much detail, which is a good  thing if the direction that it's going into is a   good one. It talks about fermentation temperature  and how it can make the beer more fruity at lower   temperatures, getting you a cleaner, more natural  flavor. Interesting. And in the end, we really   see the power of the shifted context that comes  with the words "misunderstood" or "overlooked,"   which again just implies a human interaction.   It tells us understanding the impact of yeast   in beer can not only help individuals appreciate  the diversity and complexity of beer styles,   but also enables brewers to fine-tune their  recipes to achieve specific flavors and aromas.   And in here, you can see it realizes that we're  in the context of helping individuals appreciate   something about beer. So, as always, you have to  be super careful with the words used. Otherwise,   you're gonna slightly misguide it, and it's just  gonna run off into that direction. And often,   it's as subtle as using "misunderstood"  versus "something not many people know. "   But let me tell you, for this particular prompt,  I still like running it in GPT 3. 5 because just   the results here are a little more unconstrained  than out of the box. Look, if I run this, it says,   "The importance of proper serving temperature. "  Alright, that's even more practical when drinking   beer, right? You can't really influence how it  was produced and how the yeast affected it, but   you can totally influence the serving temperature.   And then it goes into details and tells you how   much you would want to cool down different types  of beer. So, lighter beers are served at colder   temperatures, while darker beers like stouts  and porters are better served at slightly warmer   temperatures. And then the result: they taste flat  and unpleasant if they're too warm. And then it   also tells you about the shapes and materials of  the glassware. All things that you can actually   influence. This is not just theoretical knowledge;  this is something where you can stand up from   the table, make the beer cooler, pick a different  glass, and get results in that moment. And that's   why I would say I prefer this prompt in GPT 3. 5  as opposed to GPT4. Okay, now let's move on to the

third one, and this one is one of my favorites,  so listen closely. What this says is, "In your   opinion, what is the most important quality a  person can possess in the context of a situation? "   For example, you could say a prompt engineer  can possess, and why do you think that is so?    It's not exactly a situation, but it works super  well because it makes it go deeper. Otherwise, you   can say, "In the context of working with adults"  or running a startup. " So,   to test this, let's take this one and let's  go with the leader that is running a startup.    So, let's run this, and it goes ahead and it tells  it, "Your opinion. " And it's really important to   note that I included this part, "Your opinion. "  Okay, we're not asking for the objective truth;   we're asking for the opinion of the language  model, AKA we're kind of loosening the guardrails.    We're saying, "Alright, you don't have to entirely  rely upon facts. Try to read in between the lines.    What do you actually think? What is your  opinion, GPT 3. 5? Tell me. I'm actively   prompting and allowing you to tell me. " And then,  3. 5 does it. It says, "The most important quality   a leader can possess in the context of running a  startup is adaptability. " Okay, and then it talks   about the process of starting a business and how  it's a dynamic and unpredictable process. And then   how you need to have the ability to pivot and  you need to be open to feedback and new ideas,   quick decisions. It also talks about building a  team that shares this quality and how it's a team   effort. But how really adapting to the changing  circumstances is essential for success. And then,   again, it just reaffirms that belief.   Adaptability is the most essential   quality of all. A fantastic answer, in my opinion. Let's try the same thing inside of GPT4. So, right   away, it tells you it's not even gonna attempt  at giving you its personal take because it's   not allowed to do that. Again, higher guardrails,  less out-of-the-box thinking, and consecutively,   less opinions just because it's not allowed  to give you those anymore. But it comes to   similar conclusions. So, let's have a look.   It says, "In the context of running a startup,   adaptability is often considered one of the most  important qualities a leader can possess. " And   you can see the wording here is a little less  confident. It's "often considered," and in 3. 5,   it's just confident. It's like, "Hey,  it's adaptability, no doubt. " Okay,   let's keep analyzing this result. "This quality  enables leaders to navigate the ever-changing   landscape of the business world, adjust to new  challenges, and embrace opportunities for growth. "   Other key qualities include vision, decisiveness,  effective, and it's... Come on, it's just diluting   its response at this point. This response was so  much better. It was like, "Hey, it's adaptability.    Here's the way it affects you and your business.   your team. And that they   need the same qualities. And then it also lists  other things here that are important qualities,   but it takes a step back and says, "However, in my  view, adaptability is the most essential quality   of all, as it enables a leader to navigate  the many challenges and uncertainties that   come with starting a new business. " And usually  in GPT4, you're gonna get longer, more detailed,   more creative answers. But whenever you ask for  the opinion of the language model, which can be a   valuable approach to prompting because it's going  to give you more of a subjective take that you   might get from an actual person, and it's very  untypical for GPT 3. 5 to produce longer results   than M4. But in this particular example, whenever  you're looking for an opinion, do not use GPT4.

And here, for the more advanced folks, I have to  say, yes, you could definitely craft a persona   where you start this out, and you in detail  lay out the type of persona, its backstory,   its skills, and then you move into this prompt,  and GPT4 is going to respect the context of it   not being vanilla GPT4 anymore, but actually  now having a personality, and it's going to   give you more of an opinion. But first of all,  that takes a lot more work. And secondly, and   probably most importantly, I'm just looking for a  general opinion here, right? In a certain sense,   I want to explore the entire data set, which is  the wider internet that this has been trained on,   with just asking for a general opinion, AKA,  across everything that you have seen, what are the   most important qualities a person can possess in  the context of this specific situation? And when   doing that, ChatGPT 3. 5 is preferable over GPT4.   But with that being said, in like another 97 out   of 100 cases, GPT4 does produce better results. And if, at this point, you still wonder how to   craft prompts like this yourself and how to  go even beyond to make it super specific to   your use case, well, in that case, we're leaving  copy-paste territory where it's all about taking   presets and maybe customizing one or two things  about them. And we're entering the territory of   prompt engineering, where it's all about taking a  prompt like this and maybe making it three times   as long so it can fit your specific need, save  you time, and get your results consistently. And   if that's what you care about, then that's what  you're looking for. I finally have a solution

exactly for that problem. My brand new course  that I've been crafting for months teaches prompt   engineering from the ground up, and we go all the  way from basics to advanced prompts and workflows.   And look, I'll be the first to say that,  "Oh, look, another course by a creator that   is trying to sell. " Now, look, legitimately,  the quality that you see inside of all these   videos and the thought that goes into crafting  some of these, I've put that much care and   thought into every single one of the 32 videos  that you get in here. But hey, if you're just   looking for another YouTube video that advances  your prompt engineering skills today and shows   you a bunch of more tricks, well, this one is  exactly for you. People have called this the   most underappreciated video on all of my YouTube.   But let me just say that all these tips are only   relevant if you're intermediate to advanced.   So, if that's the case, I'll see you there.