All NEW ChatGPT Features Explained in 2025! (Full Guide for Beginners)

0:00 There's been a lot going on with Chad GBT lately. GBT 4.5 got released, four brand new Reason models, something called autonomous operator plus canvas, search, deep research, basically a ton of cool stuff. If you really want to get the most out of Chad GBT, you've got to know how to use all these features correctly. So, in this video, I'm going to walk you through all the latest changes, show you how each feature works, how the interface has been tweaked, and best of all, how you can do more with Chad GBT now than ever before. Just a few weeks ago, the best model got even better. Chad GBT 4.5 outperforms the 40 in every single way. Look at that. Smarter, more accurate, and way more humanlike. Was trained on an even larger data set with more parameters.

0:46 So, apparently, OpenAI took the 40 model and pumped it full of steroids, but it still doesn't have reasoning. It's just a more powerful version of 40, nothing more. If you actually need reasoning, you still have to use the 01 model. Check out our guide on that first. Otherwise, every prompt and trick you've learned for 40 will work exactly the same way for 4.5. The next big update is the new custom instructions interface. If you remember how things used to be, we only had two little boxes for prompts. One for whatever info we wanted Chad GBT to keep in mind and another explaining its tasks. Well, that's totally changed. Now to set them up, you click on your profile and then choose customize chatgbt. You'll see four input fields instead of two. The first field is just

1:32 for your name or alias. Chat GPT will use that when it replies to you. It doesn't really affect the actual answers. It's more about giving it a personal touch. Kind of like telling your phone to call you Captain Marvel, but hey, no judgment here. The second field is for describing what you do and what you need from Chad GBT. It's still mostly for context. It helps Chad GBT figure out how formal or casual to be. For example, in the old days, I used to recommend adding audience context in every prompt. Now, you can just type it once in these custom instructions and be done. The third field is all about the personality or traits you want Chad GBT to have. You can either type them out yourself or pick from the predefined options. By default, you just write it in plain English. No fancy system prompt magic words needed. Still, I suggest tossing in a quick explanation of Chad

2:17 GBT's role here, too, so it knows what vibe to go for. Finally, the fourth field is where you can add anything else should know about you. In my case, I often list the exact kinds of tasks I wanted to handle. Since there's no separate field just for instructions anymore, it's now all about us, the users, which is a bit of a bummer. I've noticed, and I'm not the only one, that if you don't spend enough time setting this up for your own goals, the output can be well kind of meh. But if you put in the effort, custom instructions can still be awesome. Over in Geek Academy community, we just posted a bunch of example instructions for all sorts of different scenarios, and they are super handy. Below all that, in the advanced section, there's a new setting called advanced voice, which makes Chat GPT's voice responses sound more natural and humanlike, but that only works if you're

3:02 actually using voice interactions, otherwise you won't notice a difference. These new custom instructions features are available to everyone, free users and subscribers alike. Have you ever record a video, maybe in a Messenger app, only to find it saved all blurry and pixelated? It's so frustrating, right? For the longest time, I've been wishing Chad GBT could handle videos the way it does photos, but that day still hasn't come. Until then, I rely on video editors on my phone. And the newest one in my lineup is called Wink, and I'm loving it for three main reasons. Let's say I save a video from a chat and it looks like it was recorded on Potato, totally pixelated. With a Wink, I just tap a button and boom, the quality jumps right up. There are four quality settings, HD, Ultra HD, portrait, and AI

3:48 UHD. The portrait and AI UHD modes are special face tuning algorithms. So, not only does the overall quality get better, but people's faces actually end up looking better than they did in the original. They're also specific scenarios, games, animation, product images, and text. So, if your footage falls into any of these categories, Wink knows exactly how to handle it. The second big thing I love is the repair features like a Swiss Army knife for videos and photos, too. You can dial in the final quality, resolution, color gradient strength, stabilization, you name it. When you're done, you end up with a video that's amazingly smooth and crisp. The third standout feature is the upscaler. Got a 720p clip you want in 4K? Wang does that in just a few

4:33 seconds. It not only ups the resolution, but also clears up any blur and restores details like magic. And if that's not enough, Wink is packed with a bunch of other cool options. There's even a fullblown video editor included. You can fine-tune your own face and body, change proportions, adjust skin tone, all that good stuff. It's like having a mini Hollywood studio on your phone. I will drop a link to link in the description. Check it out and let me know what you think. [Music] Right after Deepc caused tech stocks to nose dive by trillion dollars, OpenAI introduced two new models, 03 Mini and 03 Mini High. First, Only Plus members get to try them out. Then a few weeks ago, all free users suddenly had a new button appear called Reason. This means

5:20 you can now dip your toes into Advanced Reason and features without paying for a subscription, which is awesome. Essentially, this button activates the 03 mini model for a limited number of questions. Since that model focuses on reasoning, complex math, and coding, you'll need to tweak your prompts a bit differently than you would with 40 or 40 mini. Let me give you quick example. If you turn reason on and ask why do leaves change color in the fall? Yes, it's a very basic question, but still, you'll see a difference. The 4 answer is short, sweet, and to the point, but a little too generic for my taste. Switching on reason and asking again produces a longer, more detailed response that goes deeper into the explanation, making it easier to follow. The question I used was super simple, so just imagine how

6:05 much more noticeable this is on tougher problems. Because the reason button used the 03 mini model, all the usual prompting tips still apply. It's basically an upgraded approach to the 40 style of prompting, but with extra emphasis on reasoning. Truth be told, the list of models available gets longer and longer and they're all so similar. What I can do is ask myself. I just open AI lab at Geek Academy and call out to myself. Now I can ask questions about AI, any questions, what tool to use, how AI works, and so much more. Or I can go to talk to AI master, click the link and talk to myself directly. What do you think about AGI? Ah, AGI or artificial general intelligence is the holy grail

6:50 of AI research. Me doing this looks a bit surreal, but for you, this is a great way to get in touch and get personal help. I trained it on every piece of data we have in our videos or guides so can guide you through everything like explain how to prompt for a reason. Tell me how to prompt for reason in chat GPT. Here's a simple approach. One, be clear and specific. Clearly state what you want to achieve with your questional prompt. See you later. All right, see you later. And about that, instead of just saying, "Thank," you'll want to ask for a step-by-step breakdown. For instance, how can a person learn AI fast? Approach it step by step, show the thinking process and your thoughts. See what happened there. Now, we'll see the logic Chad GBT is using and it will handle the question in a more structured way. I personally love asking Chad GBT to show

7:35 it thinking process. Deepseek does something similar and it really helps you see how the model works. Another thing, Chad GBT lets you switch on both search and reason at the same time. So don't forget that you can get more out of it by asking for detailed source analysis and quoted information. Chad GGPT's main goal is to give you an answer no matter what. So sometimes it may skip over logical connections if you're not specific. That's why I suggest adding something like analyze the sources and their usefulness for the response which leads to better, more reliable answers. Also follow-up prompts become super important. Make sure you keep the reasoning element front and center. maybe by saying reason on or give me reasons why or anything along those lines. The big selling point of 03 Mini is that it's better at coding and

8:21 math. So, it's worth taking advantage of that. Asking it for programming help is straightforward. Just be clear about what you need. You can paste long code snippets and it'll debug them or have it generate code. It won't create a full-blown app, but it can give you a decent starting template. The same goes for math. It can handle pretty complex stuff and give multi-step solutions. Though it's not going to solve super extreme equations. Honestly, I recommend using Reason whenever you're working with your own data. Upload a file, switch on Reason, and let it analyze everything. You'll see much better results that way. So, there is a limit to how many times you can hit that reason button, which means you want to use it strategically. We've pulled together few situations where reason shines. Of course, you can do whatever you want,

9:06 but here's what I recommend. One, decision-making. If you need Chad GPT to weigh the pros and cons, compare multiple options, or just give a balanced perspective, definitely use reason. For example, should I buy an electric or gas car? Consider cost, maintenance, and long-term value. This is perfect for choices that need a deeper thought process or have bigger consequences. Two, finding and fixing errors in math or code. Here's a sample prompt. Why does this Python code give an error? Then just paste the code and get a detailed analysis. I actually mentioned this earlier in our full guide to the 01 model. So might want to check that out. Three, factchecking and verifying. Reason doesn't just make chat GBT think more deeply. It also helps it handle sources and data more carefully. If the answers are coming across too

9:52 robotic, switch on reason and ask again. You can toggle it on whenever you like. You could first gather info with the regular 40 model, then flip to reason mode for a deeper analysis. These models were released in response to Deep Seek R1. And naturally, you might be wondering if you should stick with Chad GBT or try out DeepSeek. The first thing I tested was whether I should buy an electric or gas car considering cost, maintenance, and long-term value. Chad GBT was way faster, and both AIs gave pretty similar answers. I do enjoy how Deep Seek lays out its thinking process. It's like listening in on an internal monologue. Next, I tried a coding question. I gave both models a little Python function and

10:37 asked them to fix some mistakes. Chad GBT not only spotted an error, but explained why it happened and told me how to fix it. Deepseek did basically the same job, but was a lot wordier and took much longer to respond. Right now, their servers are slammed, so you'll probably see error messages here and there. We've also put together a short guide with prompt templates for deepseek plus tips for midjourney claude chat GBT and other models all available to Geek Academy members. You can chat with AI master directly, talk to me with your voice and ask anything AI related. You can also generate images there and the quality is really good. If you don't know which AI tools you need, there is a dedicated tollfinder bot that can help you out. In Geek Academy, we teach people how to use AI effectively, share hidden tricks for popular tools, help you find the perfect one for your needs,

11:24 and so much more. We offer special discounts and encourage everyone to jump into giveaways for prices. With Geek Academy, you don't need a ton of subscriptions to learn AI. Everything you might need is right here, so click the link in the description to join us. Finally, the 03 Mini Reason tends to keep things concise and well organized, which is great all around. Deepseek on the other hand goes into deeper explanations. That can be a good thing. More detail often means more insights. It's really up to you which tile you prefer. There's one major perk though that DeepS has over 03 Mini. You can actually run DeepSeek R1 on your own computer with no internet required. We show you exactly how in our full guide to Deepseek. [Music] If you remember our older Chad GBT

12:10 guides, we used to tell you to explicitly ask Chad GBT to do a web search, but now you don't need to type a special prompt for that. The search button more or less forces Chad GBT to use the web to answer your question. If you leave it off, Chad GBT decides on its own whether it should search online, which usually leads to outdated answers. The moment you turn search on, you'll see a list of trending searches below the prompt window. Not super helpful, but hey, it's there. To get the best results, I recommend keeping your queries direct and simple. No fancy words or multi-step instructions. For instance, what is the latest research on electric vehicle range in 2025? Chad GBT will take a few seconds to search, then show you the results. One advantage of Chad GBT's built-in web search is that it actually summarizes what it finds. Instead of just tossing you a bunch of links, it gives you a real answer with

12:57 clickable links right next to the info. You can still ask follow-up questions or get clarifications, which is awesome. Just be sure to check the sources because Chad GBT can't really judge whether a website is credible. So if something sounds fishy, click the link next to it and see if the site looks legit. That's your best bet. After Chad GBT gives you an answer, you can also ask it to quote specific info. For example, if you asked about quantum computing breakthroughs and it gave you a summary, you can follow up with quote the part of the report that mentions the statistics. This way, you'll get direct, less processed info that's easier to verify. And always remember, fact check everything. That's AI 101. Because the search feature uses the 40 model, you can do cool stuff like ask Cad GBT for a specific number of sources, set

13:42 standards for those sources, etc. But keep in mind, if it can't find anything or gives you completely useless answer, just run the search again. That usually fixes the problem. There are a few search limitations though. This site needs a login, Chad GBT can't get in. Same with those pesky cookies that block AI from looking at a page until you accept them, like on Medium for instance. Aside from that though, it's pretty handy. Another major addition to Chad GBT is Canvas, and it's seriously big. Canvas is like a brand new way to team up and work on projects. Switches on automatically when Chad sees a scenario where it might help. That's when your screen splits into two. Chat on the left, results on the right. If you want to launch it yourself, just add use

14:28 canvas to your prompt. Here's how it works. You type instructions on the left side. Chat GBT does his thing and the results show up on the right. You won't see the entire chain of replies like normal. Each time you make a change, the document on the right gets updated. The general process is basically the same as the normal chat. Canvas just gives you a more convenient layout. All the fancy new interface tools sit on the right and clicking the clock icon lets you see exactly what changed in your last edit. After each edit, ChatGpt also tells you what it did in the chat window. Back and forward buttons let you jump through older versions of your file. The copy button copies your results. If you see failed to copy, just try again. Finally, the share button generates a link to this canvas so others can jump in and collaborate. That sharing feature is brand new, by the way. Working with text

15:14 is exactly how you did in the chat. There aren't any fancy new prompting tricks here because Canvas is essentially the same back end, just a different interface. But if you're a writer, we've posted a special guide to Chad GBT prompts for authors in Geek Academy, where we show you how to harness AI's full potential. You can still save a bit of time using Canvas's quick edit features. Look in the lower right corner for a pencil icon and click it. A pop-up will appear with options for one-click edits, like adding emojis. You can choose from four options. Add them to words, sections, lists, or remove them entirely. There's also final text polish feature plus sliders to adjust reading level or text length. These are super simple to use, even if the edits are pretty basic. Another neat trick, if you hover your mouse over a paragraph, you'll see a button pop up.

16:00 Click it and you can type in exactly how you want that paragraph to change. Maybe you want to add some new details, change the formatting, or even rewrite it completely. It's all up to you. Where canvas truly shines is when you're dealing with code. The Hall canvas mode was originally built with writers and developers in mind. It even used a specially trained model at first. It's just running on a trusty old 40. The code and tips are pretty much the same as they were for 40, but now it's a lot simpler to do everything. Those buttons on the right side when you're in coding mode are different from the usual ones. They let Chad GBT add comments, insert logs, or fix books in your code. You can also pick which language you're working with by clicking forward to language and choosing PHP, C++,

16:47 JavaScript, or TypeScript. The last button at the bottom is code review, which makes Chad GBT check over your code and give suggestions. It's super handy for troubleshooting your own work or trying to learn something new. Another useful trick is selecting a specific part of your code and asking ChatGpt about it. Just like with text editing, chat GBT can edit or explain the highlighted piece and the explanation will pop up in the chat window on the right. The most important thing though is that you can actually run your code right inside canvas. Click run and the interface changes to suit whatever you're running. For instance, if the code needs a console, it will show up at the bottom of the screen. Now, let's shift gears and talk about changes in the mobile app. There's been one update you'll definitely like.

17:33 Advanced voice mode. Before looked like a white circle and a black background if you're still seeing that means you haven't updated. So go to your app store and grab the latest version. Then go into your app settings. Tap personalization then customize chat GBT. Scroll all the way down to advanced. Make sure advanced voice is switched on. And while you're there, you may as well enable any other features you see just in case. After that, open a new chat and hit the wave button. You'll notice the interface has changed. There is now colorful circle in the middle where you can mute your mic and the chat or change the voice. To pick a new voice, tap the settings icon in the top right corner. Play around until you find one you like. From there, just speak to it like you would to any other person. You can interrupt it, ask for clarifications, whatever you need. Don't expect any

18:19 special powers, though. It's still using the same 40 model as before. The only real difference is the voice quality. But do keep in mind that once you run out of advanced voice time, switches back to the standard more robotic voice. And trust me, compared to the advanced one, it's pretty dull. One more thing to remember about voice mode is that it can't access Chad GBT's memory or store anything in it. Custom instructions won't work either. On the upside, there is a neat feature where you can hit the share button to send someone a little screen recording of your conversation if you want them to see or hear what just happened. [Music] One cool feature you might not have noticed yet is schedule tasks. Basically, you can tell Chat GBT to send you information at a specific time on a

19:06 specific day, and it will send it to you by email. This could be a daily summary of your calendar, or a short list of news articles. When the time comes, you'll get an email, a phone notification if you have the app, a browser alert, whatever you've enabled. If you don't allow notifications, you'll still get an email which looks super basic, just a big green view message button that opens a new ChatGBT tab. And just to clarify, Chad GBT isn't trying to become some sort of autonomous agent. It's literally just scheduling tasks. Using Chad GBT tasks does require ChatGBT plus subscription. Open Chad GBT, start a new chat, and pick GBT40 with scheduled tasks. Type in a prompt that specifies both what you want and when you want it. For example, every hour starting at 1:00 p.m. today until 12:00 a.m. tomorrow, send me a fun fact about koalas. Chad GBT will then create

19:54 a new task in your chat, and at those specified times, you'll get the fun facts. Tests can be one time or repeating, but don't plan things too far in the future because the system can still be a little quirky. You can see all your scheduled tasks by clicking your profile icon in the top right corner on the web and choosing tasks. If you want to pause or delete one, hover over it, click the three dot menu, and pick pause or delete. You can also change a task's name, instructions, or schedule from there. Another quick way is to click the three dots next to the task in the chat, then select edit task. That'll also take you to the full task list. Just keep in mind, Chad GBT only lets you have 10 active tasks at once. If you hit that limit, you'll need to pause or delete something or wait for a task to finish before making a new one.

20:42 Let me show you a couple ways you can use tasks to boost your productivity. But first, here's a little pro tip. When you create your first task, schedule it to happen in 1 minute. That way, you'll see how it works right away and can tweak the prompt if you want. Something like, "In 1 minute, give me the current weather report for tomorrow morning." One, meal planning. That way, you'll get a custom weekly meal plan right before grocery shopping. For tests like this, it's a good idea to include more detailed info about your dietary preferences. Two, regular work break reminders. Even though CHBT timers aren't super fancy, you'll still get two notifications. One to take a break and another to let you know when to get back to work. If you're someone who does most of your work inside ChatGBT, you might want to duplicate your schedule here.

21:27 For instance, I use Google Calendar, but often forget about start time. So something like uh every workday from 8:00 a.m. till 400 p.m. I'm working. Send me a message to start work and to end it. It's a simple setup, but super useful for urgent tasks. Just be aware that chat GBT can glitch sometimes sending only one message instead of two. So you might want to schedule them separately. Another button that all paid users have likely spotted is deep research. This one is a serious upgrade for anyone doing in-depth research. Think of it like having an entire perplexity built right into Chad GBT. Once you flip on deep research, Chad GBT isn't just doing a quick web search for a couple of sources. It actually launches a multi-step research process

22:13 pulling info from web pages, PDFs, and other sources to give you one concise, organized response. The trade-off, it can take some time. Depending on how big your research question is, you might be waiting a few minutes or even up to half an hour. For each answer, Chad GBT includes all the source links and it works with any model. That means you can easily combine deep research with the 01 model for fullon reasoning, not just data. With 40, you'll get results a little faster. With 01, you'll get a deeper, smarter analysis. Personally, I suggest always turning on 01 if you're doing serious research. It's just that good. Before I talk about how to prompt it, here's a key thing to remember. Always phrase your question clearly and directly. Avoid fluff sob doesn't go chasing the wrong info. And if your

22:58 research project is huge, break it into smaller parts and then use all one's reasoning to piece everything together. [Music] Again, prompting for deep research is like another layer on top of the model you're using. Whether that's the 01 with its strong focus on reasoning or the 4 with multimodality and speed, it will be adding the same phrases to your prompt. First off, ask for relevant data. Specify time frames if you want truly upto-date info. Chad GPT wants to answer and it will use any data that fits its algorithm, even if it's seriously outdated. A prompt like, "What's the best food to eat while gaining muscle mass?" will give a balanced response. But adding use data newer than 2020 makes it more relevant and factual.

23:45 Second, ask for citations. By default, all Chad GPT with deep research enabled only adds links to sources. This means it takes the raw data, processes it, and then uses it for the response. What's the best food to eat while gaining muscle mass, use data newer than 2020, add direct quotes, will now show you pieces of articles and research papers that were used. This is especially helpful for students or researchers who need quotations. Next up, ask ChatGpt to doublech checkck the credibility of sources. You can provide a whole list of criteria on what useful sources are, or just let ChatGPT figure it out. Just be sure to turn on the 01 model before typing a prompt like this. Operator was easily the most buzzed about Chad GBT feature, an agent

24:31 that actually browses the web for you. Rather than just given answers, operator can click, type, scroll, and basically complete tasks almost like a human. At least that's the idea. Right now, it's only open to pro users and only in the US. So, if you're not in the US, you'll have to use VPN and drop $200 on a subscription to give it a spin. According to OpenAI, operator can autofill forms, book flights, hotels, and restaurants, restock your groceries, order food, and even whip up memes. In my test, it's been a bit hit or miss. Still kind of buggy, but hey, it's in beta, so that's what you'd expect. To access operator, you'll need to switch tabs in the menu on the left. It'll open a new window with a slightly different Chad GBT interface. You won't see deep research or web search anymore. Instead,

25:17 there are buttons like dining and events, delivery, local services, and so on. Each button fires up a specific operator algorithm, and these preset scenarios work best. Below them, there's a list of sample ways to use operator. Give them a look to spark some ideas. The layout kind of reminds me of canvas. The chat is on the left, the output on the right. You can click to expand the generated response and see operator's thought process as it goes. On the right, you'll watch a live preview of the actions it's taking. If you see it's struggling, you can hit take control to do the clicking yourself, which is handy if a site is giving Chad GBT trouble. The first thing I found operator can do is price research. Here's a prompt. This task would have taken me a solid five minutes and operator took roughly 5

26:04 minutes as well. However, it's results are a bit more structured than what I would have done myself. Prompting operator is pretty much the same as talking to chat GPT. Normally, you can say something casual like order me sushi and it'll ask for your address, preferences, portion size, etc. That said, if you're placing an order, it might be simpler to do it from your phone. You can also use it to check the weather, plan travel, or anything else. But to be honest, I've noticed that sometimes using operator takes about the same amount of time or more than doing things manually. So, at 200 bucks, I'm not convinced it's worth it yet. Hopefully, once it's publicly released, it will have memory and actually get to know us better. So, ordering sushi could be a matter of seconds, just like in the sci-fi movies.

26:52 Before we got the 03 models, we only had the 01 and 01 Pro. All that I told you about the 03 is true for the 01. Both models do exactly the same thing, but in different levels. 01 and 01 Pro may be less adept in coding and math, but they are much better general reasoning. Just like 03 models, the 01 models are quite limited. You only get text, textbased images, and PDF inputs on 01. But the context window is huge, which partially makes up for the lack of proper multimodality. If you are on the plus subscription for $20 a month, you get limited weekly messages, 50 per week, but you can unlock GBT1 Pro with a $200 plan that gives you unlimited access. Also, more sore credits and an overall

27:38 upgrade in performance. Since 01 digs deeply into topics, it's important to guide it with the right prompts. First, clarify the level of detail you need so it knows to provide an in-depth answer. Including detailed report in your prompt forces 01 to pick a more structured approach. Second, if you wanted to break things down step by step, just say so. This helps the 01 model structure its reasoning so you can see exactly how each part of the solution fits together. And since it can take a while to respond, try to group related questions together instead of sending them as different prompts. Be sure to specify how you want the final answer formatted, especially if you need tables or headings. If you're working with code or big chunks of text, you can feed them to 01 all at once thanks to that large context window, which means you can

28:25 debug larger scripts or analyze lengthy data easier. The trade-off is speed, but you'll usually get fewer errors and a more logical breakdown of each step while using the 01 models. I also want you to remember that if you need extra data, input it manually. The 01 can fetch realtime info on its own unless you turn on deep research. Overall, prompting for the 01 is easy, but for more information, be sure to check out our new full guide to the new reasoning models. The projects feature has been out for some time, but to some people it's still new. This feature is super useful for anyone who works on multiple projects in one chat GBT account. With projects, you can keep all the chats for that project in a separate tab complete with its own custom instructions,

29:11 memory, and so on. It's basically a way to have multiple separate chat GBTs. To create a project, find the projects tab in the menu on the left. Then click plus and give your project a name. Once it's created, open it and hit add instructions. These custom instructions look different from what you're used to. There's only one field. How should CHBT respond? But that's actually simpler. It's easier to give direct instructions on what and how to do things this way. At the bottom of a project, there is a familiar prompting window. Just start prompting and all new chats will appear in the list of chats. However, projects don't have all Cad GBT features. You won't see deep research here. Only file uploading, web search, and temporary chats. At least reasoning works. if you select the 01 or 03 model in the model selector. Interestingly, this organizational feature is something many

29:58 people want for free, but at the moment, projects are only available to plus and pro subscribers. In terms of actual use cases, it's pretty straightforward. There's no special prompting involved. It's purely an organizational feature. You create a separate project for each distinct task and each time you start a new chat within that project. And that's basically it for projects. [Music] So, thanks for hanging out with me. Thanks for being awesome. Thanks for watching and see you in the next one. [Music]