Is ChatGPT 4.5 Even That Good!

We've been dealing with AI in a lot of different ways that has truly better or worse conversations. And when I say conversations, I mean that. I mean in the conversational sense. For example, when it comes to ChachiPT, I'd say it's always sounded pretty onhuman. You have to do a lot of prompting in order to get it to become human. Then there's products like Claude, which we're going to cover 3.7 in another episode just to break down the benefits of that. But simply put, chat GPT4.5 is the first step forward in a human sounding model. Hi, my name is Dmitri Bonichi and I'm a content creator, agency owner, and AI enthusiast. You're listening to the AI agents podcast brought to you by Jot Form and featuring our very own CEO and founder, Idkin Tank. This is the show where artificial intelligence meets innovation, productivity, and the

tools shaping the future of work. Enjoy the show. Hey there, my name is Dmitri and as many of you know, this is a podcast about AI, AI agents. And in this episode, we just wanted to talk to you about Chat GPT4.5 as it has made waves in the community, but not in the ways that I would have liked it to make waves. Also, for those of you that are a little bit new to the podcast, we haven't really covered a lot of AI news, but we are going to be doing more of that soon here, and I recommend that you subscribe if you want to keep up to date with all of the latest news going on in the world of AI and AI agents. First and foremost, Chad GPT 4.5 came out in February at the end of it, and it probably was something

you might have missed. The goal of Chad GPT4.5 is that it is going to have not only just a better capability than GPT40 in regards to accuracy and that would be higher and lower hallucination rate but it also has something that other options in Chad GPT do not have. This is the human factor. So, as you can see here on the lefth hand side, it really does talk a little bit more like a person. Whereas on the right hand side, it talks like AI. You've experienced AI. Look, like literally you just wrote, I'm having a tough time after failing a test. And on the right hand side, it says it's sorry to hear that. So, as you can see in the article, it points out that it doesn't think before it responds, which actually makes it stronger than 01 and 03 in regards to having

a general purpose AI. So, I'm just going to show you some examples of the differences between what it was like to write with 40 and 4.5 and as well as three. I'm going to ask it the exact same question. And this actually comes into play, I think, with content writing a lot cuz most people use 40 for content writing because 03 is just ridiculous and 01 as well is ridiculous cuz it's reasoning. So, we're just going to blast something out that's over the top. I want to write a social media post about the benefits of chat GPT model 4.5. And I can do like a little internet search. Make it optimized for LinkedIn. And I'm going to just copy this so I can redo the prompt every time. So, first we're going to get a 40 response. Embracing. I can't stand when he uses words

like this. Okay, this is a pretty decent response. It talks about how 4.5 exhibits a deeper understanding. Exhibit's a really not human word in the context of a social media post. And I can even try to say, make this post sound more personable, like you were talking to a co-orker you are friends with. And I'm going to turn off the search. Hey friends. Yeah. No. Who talks like that in a social media post? Game changer. I can't stand when it always uses game changer. If you've been using AI for work or just out of curiosity, you'll definitely notice how smoother and smarter this version feels. It added emojis this time. As you can see, it's got better context, fewer hallucinations, stronger reasoning, and plays well with other tools. Okay, cool. So, let's try this out again to prompt sequence-wise, but with 03, just to point

out the difference. I just I can't stand and and this is this is nothing against there. It goes over the top so much when it comes to length, the hashtags, all this type of stuff when we talk about four and 03. So, let me grab the second prompt and see how it goes. the three mini it goes. I love how by the way it says reasoning. Hey there, I just had to share. Okay, not bad. I just don't think the formatting of this writing is that good. The way to understand context and nuances and conversations feels human. It's made brainstorming and problem solving so much smoother. Fits right in whether I'm tacking a new project. Okay, this is probably a little bit better than 40. Granted, 03 came out way later. But let's try this one more time with 4.5. Okay, let's do another section.

Okay, first of all, it got rid of the weird bullet point system that it had before. Let me add a second prompt like I had before about personality, and we'll see how it does. And I like how it didn't do the whole breakdown. Ooh, put it in the canvas, too. Yeah. See, this this has some human writing to it. And honestly, I'm pretty impressed. It's a big upgrade. Better conversations, fewer mistakes, and way more intuitive. Way more is not something AI would have written previously. For anyone who's been frustrated by AI hallucinations, I know I have. This model's accuracy boost is super refreshing. Definitely recommend checking it out. It's available now for TGBT Pro users and select developers. So, as you can see, we're just noticing a different writing style here. And in the same sense, I think it's just going to be interesting to

know we've been dealing with AI in a lot of different ways that has truly better or worse conversations. And when I say conversations, I mean that. I mean in the conversational sense. For example, when it comes to chatbt, I'd say it's always sounded pretty unhuman. You have to do a lot of prompting in order to get it to become human. Then there's products like Claude, which we're going to cover 3.7 in another episode just to break down the benefits of that. But simply put, chat human sounding model. Now, I don't really think that's the case though in the entirety of models that exist. You could use different tools like Jasper AI or even just clawed in itself. I think just sounds better. Personally, we're really stuck on chatbt as a community right now. It's ingested in so many different places. Apple has a relationship with

them and they're trying to get that into the entire ecosystem with things like if you go to the top right of my screen, you can see this has a connection with OpenAI. Like for example, I'd imagine this is going to be much better for Siri and things like that. So for example, I can ask what are the top five things I can do to save time. It says I need to use Chachi to write that. It's going to work on it. They've integrated this into the capabilities of Chacht and imagine a model that just talks like a person and they can fine-tune it. This is fine. This is a good list. I appreciate it. However, if we're talking to this product and we want to interact with it inside of our system, I don't want to get a 45word response from 40 when it could

be a 20-word response or a 10-word response. Like you'll see more examples of these when I say things like I'm looking for fasting options for nonan animalbased food that I can have breakfast, lunch and dinner with. Also no olive oil as I can't have that. Please give me some ideas. And sure, research is great, but sometimes people are just kind of want to interact with something and especially when we get more into the verbal side of it because we are going to get more into the verbal side of it. This is a good breakdown. Chickpea salad wrap, lentil and vegetable stew, good ideas. However, the hallucinations are less and the personality is better. So, if I put this out, just take a look at the writing and see the way that it writes slightly differently. So, for the basics, it's okay. But if I ask

questions like, "What would be your top three pieces of advice to handle this new type of eating?" What's going to happen is it's going to respond slightly shorter and slightly to more to the point with more human type words. So, prepare large quantities of staples like grains, legumes by educating yourself and decking. Yeah, it's just a little bit different. Whereas if I ask the other one, especially the reasoning models, it's just going to be expansive. And what I have noticed, by the way, with 4.5 is there's not a lot of errors, uh, sometimes when chatbt releases new models, it gets pretty annoying when you're essentially just trying to work with a product and it'll just crash. Not great. Now, okay, so you can see here, understanding the principles of whole food, plant-based, no oil, educate yourself about plant-based nutrition, meal planning, and oil free cooking

techniques. Slightly easier to read out. Not bad. giving nice succinct bullet points for the sections rather than these huge text blocks. I actually think that was a much more readable experience. However, I do have to call this out. If you're looking for a model that's essentially just like, wow, they managed to supercharge 40. Like the leap from 3.5 to 4 was a pretty dang good leap. Yeah. No, that's not it. This is slightly better. I mean, this is noticeably better, but it's one of those things where is it significant or is it statistically significant, right? Is it going to feel significant? No. Is it statistically significant based on their own numbers? Yeah. That's something that frustrates me about a lot of these updates. However, same thing goes with phones. In the moment, we're annoyed that the iPhone 15 is only a little bit better than

the 14. However, when you look at the 10 versus the 15, it's better. And that's kind of just how progress has to work with these things. And it is weird because the leap was so big from 3.5 to 4 and you think things get exponential. However, they're working on deep research. They have 01, they have 03. There's a lot of really nice models coming out. Apparently, five is going to come out at some point this year, we think. And that is said to be a much bigger leap than what you're seeing here. I've noticed a lot of people on the internet basically clowning this, saying how bad it is. I don't really think it's anything to scoff at. The only thing that does suck is that this is essentially something that could write slightly better. And for content, an irritating thing for me is that

the API on this is essentially not good. Let me explain what I mean by that. Same context length. That's fine. Look at the cost difference. What is that? Who that who in their right mind would pay for this right now? 40 and 40 mini are not that much worse. And you can just do great prompt engineering and get similar text outputs. you can do multiple prompts in a row to get it to sound human. And since they released the responses API and stuff like that, you can add knowledge bases and context. So, I just don't really think it was great of them to release this, especially with the API at that cost, and be like, it's a new update. Like, no, I don't I don't see it. They claim that it's good for, you know, agentic planning, too. I think that's a bit of an

overstatement. like you have the other models that can do better reasoning and think of things and you can do deep research on agentic research, right? So, I don't I don't really I don't really see how this is a huge leap forward. I do see that it is a solid leap forward and if you have it on your plan, might as well use it over 40 if you're writing, but I'm not really into the way that this was announced. But I'm going to be reasonable and say similar to whatever happens with your iPhones, be happy that it improved. Okay, I'm happy it improved. You're happy it improved. I hope. And with that being said, please make sure to leave a like, a review on Apple Podcast, and let us know if there's anything else you'd like us to review. We are going to be releasing more

short news-based content here on the channel. Thank you so much for watching, and we'll see you in the next episode. Bye.