GPT-4.5 Is The Most Human AI EVER & More AI Use Cases

Okay, generative AI is going crazy right now. OpenAI released GPT 4.5. Anthropic released Sonnet 3.7, and both of them are state of the art. That's specific things. And we have new audio transcription models, the best video model finally being available, and so much more. In this week's episode of AI News, you can use the show that looks at all of the Generative AI releases from the past week and filters out only the ones that you can put to work today, starting with ChatGPT 4.5. We just gotta talk about this. I'm recording at midnight. This one came out a few hours ago. I'm not doing a dedicated video on it because I have some content ideas that take a little bit of time to realize. And I figured the first look we can do right here in News you can use. So let's do exactly that. But first let me briefly summarize the key facts here. Because this thing, unfortunately, as of today, Friday, February 28th, is only available to the pro users. That's a $200 plan. But no worries. Within a week they want to ship this to all teams and plus users, that's a $20 plan. And after that, they're going to ship it to all the educational and enterprise accounts. So this thing's not going to come to the free tier anytime soon, but within a week, everybody should get it at the standard premium price of $20 with the ChatGPT plus plan. Now, why should you care about this model? We're not even going to pull up benchmarks here. Okay? This is the first time in a while where I'm just going to completely ignore those. Because really recently that's really been the story. It's been about pushing those benchmarks further. It's about performing better on all of these coding and mathematical evaluations. That's the story behind all the thinking models. This not a thinking model. So, look, I know some of this, especially the stuff we're about to talk about here, can get a little confusing, especially if you don't follow this day by day. I'm going to create a separate video very soon, breaking down all the different LLMs and telling you which one we use and which one people use for different use cases. But the very brief version of that and what I've heard about 4.5 and how they presented it so far is that, hey, if you wanted to write or brainstorm with you, this is the best model in the world. GPT 4.5. I'll verify that myself in a second here with my first look, but that's what they claim and that also seems to be the consensus across the Internet. So I have high expectations there when it comes to coding. Everybody always loved the Claude models. Sonnet 3.5 or now Sonnet 3.7 we're going to talk about Sonnet 3.7 in a second here. I created a separate video but have some follow up thoughts that we'll get to in a second here. That model is really the king for coding and if you want up to date data, well Grok3 is your best choice because it's plugged into the Twitter feed. It's going to be the most current model and then if you have mathematical or science related problems you want to look at one of the reasoning models, probably O3 mini high or deep seq R1 and if you want to research something, well then deep research by OpenAI is just the undisputed king there right now. That's a very rough overview of the different models that people are using. Of course there's way more again already creating a separate video to kind of give you a lay of the land when it comes to all the LLM platforms out there. So back to GPT 4.5. This is not one that will crush benchmarks because it's not a thinking model, it's a normal model like we're used to with GPT4 GPT4O Cloudsonnet 3.5. All of these models don't think before the answer, they just give you the answer straight away. So they're not going to crush benchmarks, but their main claim to fame here is that it has higher eq, better vibes, it feels more human, it's a better brainstorming partner, it's a better writing partner, it's a better creative partner. Now I personally love to hear this because that's my main use case for LLMs. I love it for brainstorming, I love it for writing, rewriting. I love to use it as a creative partner, as an assistant to my different workflows. Now with all that being said, let's do a very first taste test here. I'm going to go into my ChatGPT Pro account, that is the $200 subscription and I'm going to switch over from 4.0 to GPT 4.4.5 and I'll run one of my standard benchmarking prompts to see what the tone of this is like. Write an email to my boss about the broken coffee machine in our office and I'm super curious to see how it handles this very standard request. Okay, this formulation is very different from what other models give. You can already tell you that. Perhaps organize a repair. That is so kind. I'm happy to help coordinate if needed. Also offering the help. This is different. Okay, so I might be a little biased because I heard that it's a bit more human, but this is way more human. And you know what? For each one of these prompts that I run here, we'll put up a comparison on screen so you can Compare it to 4o, the previous model and Claude Sonnet 3.5, which many people up until now considered to be the king when it comes to the writing style. People just love writing with Claude. So from these comparisons you can make up your own mind on what works best. So yeah, what can I say? Except that 4.5 is clearly the best here. It shows empathy and offers help to the boss with solving the problem. And it also keeps it super concise by default. Okay, I'll do another one of my standards. Write me an essay about penguins. Okay, this is the prompt that I probably ran the Most across all LLMs, so let me just subjectively judge this. Known primarily for their upright posture, tuxedo like coloration and charming waddle, penguins are a testament to the marvels of evolutionary adaption. Oh, this is just way better. It sounds less like a textbook and more like something a person would write. Yeah, I mean, that's what this sounds like. Yeah, it just sounds less boring than the other alternatives, don't you think? Definitely more human. I mean, that's lovely. Okay, what if I add in the style of Bukowski penguins, little tuxedoed bastards sliding on their bellies across a frozen wasteland like drunks leaving a bar at closing time. Wow, that's really good. I mean, if I only were to Compare that to the 4 output here. Penguins. Christ, if there ever was a bird that embodied the cosmic joke of existence, it's these little bastards. I suppose in this style, actually both of them perform super well. So it's really this default tone that sees an upgrade. Interesting. Okay, for the idea generation, I'll go with something very standard that I've also done a ton helping it to brainstorm me titles for YouTube videos. LM Search, generally speaking, not very good at this. Include a few keywords here that I really like there. Like generate emotionally engaging ideas for a YouTube video title that compares various LLM platforms like ChatGPT, Cloud or Gemini. Let's run this in 4.540 and Claude. Okay, instantly I can tell this is Better. So look, these are just the first few prompts and I realize that this is not a comprehensive test, but I've ran these things so many times. I used prompts like this in every LLM out there. It's just higher quality. These are actually usable titles. This is really good. Also this is really good. Yeah, also this is really good. If you look at some of these other ones, this one's good. Oh, look at that. Actually 4. 0 upgraded. Also, I gotta say, these results changed from a few months back for sure. Okay, actually I like 4.0 here. And then Claude is a bit too verbose on this one. Generally speaking, for ideation, it never used to be my real go to model. It's really good. Don't get me wrong, Claude was the one that I just really like for writing and rewriting and code, of course. Okay, what about a cheeky little follow up prompt here? What other similar videos could I create? Okay, 4.5 comes up with categories. It talks about the target audience, which is something I should have mentioned, but I consciously didn't. Best AI for work. That's a good concept. Everyone's wrong about ChatGPT and Claude. Heck, I want to click that myself. Wow, that's a lot of ideas actually right there without me prompting anything intricate. Very solid results right here. Let's see what 4 came up with. Ooh, similar structure. Look at that. And there's a lot of these challenge type videos. Replacing a virtual assistant. Now I have to tell you, like, there's maybe one or two usable ones in here from my perspective, just purely personal point of view for this channel. Whereas this, every second one of these feels like I might want to pursue that idea further. I mean, damn. What can I say? 4.5 is just really good here. And Sonnet 3.5 as expected. Nah, it just doesn't fit the content of the channel at all. I mean, these are very general ideas. No, there's not a single one here that I would want to make. Whereas 4.5 got me, well, at least inspired, but also I feel sort of motivated with all of these good ideas. So there you go. The first model that really sounds like a human. It has empathy, the number one thing that I was looking for in these models. That's why I liked Advanced voice so much. And it seems to be baked into 4.5. I realize it's just a few prompts, but hey, it just released. I'm just testing it out. You already know, I'll spend my weekend throwing a bunch of different prompts at this and then I'll come back with a more in depth evaluation of these different models, how they compare, when to use, which one. But yeah, as everybody seems to agree, and as OpenAI mentioned, this does seem like the very best writing model and also the very best brainstorming and creative model. They did mention that this is probably the biggest model that has ever been trained in human history and that seems to line up with the quality of the results here. So hopefully within a week everybody on the plus plan will get this. And meanwhile I'll do the same thing I did when Operator came out. If you have a prompt you want me to run, you can just leave a comment below and we'll take some of them, run them through 4.5 and just reply with the answer so you can test some of your own prompts already, even though you might not have prepared Pro. Of course we can't guarantee an answer to everybody, but within the first day or two we'll do our best to reply to as many comments as possible. One more thing about GPT 4.5, this is wild. If you want to use this model through the API, this is the wildest LLM pricing that we've ever seen. 1 million input tokens cost $75. Just to compare, if you were to use DeepSeek R1 through the API, 1 million input tokens cost 14 cents. Okay, the output, so when it replies to you $150 per 1 million tokens, $150. That's virtually unusable for like 99% of the use cases. That's crazy. So really we're blessed with the fact that this is available through ChatGPT because the API costs are just not feasible right now. All right, so that's 4.5. That's really the main story of this week. But then we also got soned 3.7 and I created a separate video on that. Now that's an incredible model, especially for coding. It's a thinking model and I've been using Claude Coder ever since a bit. So I just want to follow up on that video. If I' seen it, I give you an overview of what Claude Coder does and an overview of Sonnet. One thing that I failed to mention is the API cost that comes with Claude Coder. I said Sonnet is free, which it is. You can freely access it on Claude AI. Also, Claude Coder is free. There's no subscription. But using Claude Coder takes up API credits and I've been using it every day for about 30 to 60 minutes to build little Chrome extensions or little dashboards with different AI apps in there and presets for myself. And the average usage for me over the past five days has been $10 per day. I mean, $10 for a little application is sort of neat, but most of the times these are just crappier versions of what you would get for, I don't know, a 10 or $20 subscription from some other service. So it's definitely not cheap. And I wanted to make that clear here. Claude Coder itself is free. I think they put it behind the waitlist now because of the popularity, actually. But the API requests aren't. Let's move on to the next one. Okay, so our next segment is something for everybody who's using the code Chrome browser. And this right here is a nifty little Chrome extension that allows you to use various reasoning models right inside of your browser. It's called Monica, and they're the sponsor of today's video. And when I first saw this, I was genuinely interested because this is one of the most interesting Chrome plugins that I have seen so far. But rather than me telling you about it, let me show you what I'm talking about here. So I'm just going to pull up a random Wikipedia article and pretend like I'm writing a research paper for school or something. So if I wanted to use LLM with all of the information in here, there's multiple ways I could use Search. I could select all of this and copy it over into a large language model, or I could just stay in the window that I'm in already rather than managing multiple windows and tabs. Because the Monica extension allows you to run models like deep seq R1 directly on this page. And if I want to work with this tab, all I need to do is go over here, say, chat with this page, and hit summary. That's it. And it starts a brand new conversation. I get the full summary right here and I get to model switch between all of these various models, including Deepseek, O3 Mini and more. Let me be clear, this extension is very extensive. They built so many things into here that this chatting with the page is literally just scratching the surface. Because if I wanted a different type of summary, I could just go down here to my prompt library and I saved this density summarize. It is actually a prompt I use myself in my own workflows. It creates five summaries that are increasingly dense and usually somewhere around the fourth or fifth one, I find the very best summary that AI can produce. And here in this case I can just use the preset prompt in the conversation with the Wikipedia article and without me having to type, copy, paste, anything like that. I just applied my prompt preset and here I have my five summaries. Five increasingly then summaries. Usually I look at number four and five and those are just way better than anything you will get from a simple prompt. That gives you one then summary. And then obviously you could add your own custom prompts to this library, Use voice inputs, search the web. All of the good stuff that you might be used to from your favorite LLMs, but it's all right here in this application. And what I was showing you here was the free trial of it. Now, if you want to fully use all the features and get more than 40 requests a day, then you'll have to subscribe for the paid plan. But if you've been following this channel for a while, you might have noticed that this is literally the first browser extension that we've ever featured. And we get a lot of requests for these. I mean, people use AI tools to just build them quickly. But I figured that this one in particular is super neat and might even cater to the needs of some power users here because it has so many other features that you can dive into. So go ahead and test it yourself on the free plan. Download the Monica Chrome extension today from the link at the top of the video description. And if you do sign up, use the code TAA10 to get 10% of all plans. And they also have a second offer right now. If you subscribe within 24 hours of signing up and get the annual plan, you'll get 25% of that. All right, now let's get back to more AI news that you can use. Our next segment is that they're finally, finally rolling out ChatGPT deep research to all paid users, not just the pro plans. So if you're on the $20 ChatGPT plan, or the Teams plan, or the educational plan, you now get 10 deep researchers a month. I still think this is the most significant AI releases for most consumers that we got ever since GPT4, maybe. And if you haven't given it a shot, this is the time to try it. I've been talking about it regularly over the past week, and I created a separate video looking at 12 incredible use cases for this tool. So if you haven't seen that video, or if you didn't watch the video because you didn't have access to the tool, now is the time we'll link it in the Description below, all ChatGPT plus users, you can use Deep Research now finally. And they also made some changes to all other users. So all the pro users who are paying 200amonth now get 20% more deep researches a month. I think this actually matters as we at the AI Advantage hit the limit on multiple accounts already and it hasn't even been a month. And they shipped some updates where the Deep Researchers now include include images and their outputs images that are directly cited from the sites just like most of the content. And they're saying the Deep Research has a better understanding of uploaded files. And this is great because a lot of times I gave it a document and it's sort of ignored it and really focused on the links. So I'm glad to see that if I do something like uploading the five page document with my company context, it will be able to reference that better in its Deep Research. So there you go. The, in my opinion most powerful AI tool out there is now finally accessible, possible to a wider audience and go save some time with it. And on a similar note, and this will be super brief, Perplexity has sort of a weaker version of Deep Research that I mentioned last week, is now freely available on a free account with Perplexity and this week they're releasing it through their API. So there's no OpenAI deep research API yet. So if you want to call it programmatically built it into an automation or whatever it might be, can't do that yet with OpenAI's version. It's only available through chatgpt.com perplexity shipped this and I expect or hope for a Deep Research API soon to come because in a lot of my automations I would love to start with step one, go and scour the Internet for half an hour and then start the automation right now. You can start implementing that with the Perplexity Sonar Deep Research API if you want to. Let's move on to the next one which is the release of VO2. So this happened towards the end of the last week, so we didn't catch it in that news roundup video. But VO2, the best AI video model as most people agree, that's Google's model, is now finally available. Up until now we only had previews and a lot of examples and some select people who were in touch with Google had early access. But now we can finally use it. And there's two sources. One of them is on freepik, the second one is on File AI. I'll include both links in the video description below. And we did the classic thing that we do here at the Advantage and that is running the basics test prompts. So you can compare this to some of the other state of the art model while watching the video right now on screen you'll see comparisons of some of the clips we have done. Like this wonderful sloth hanging out in a pool. And let me tell you, all of these examples VO2 just nails. It's better at human expression than pretty much all other models. The water looks really realistic. I know that you must have heard this a hundred times before, but this is the best AI tool in this category. And the reason you keep hearing that statement so often is because generative AI is just progressing so damn fast. So if you kind of lost track of all the video generators and you were wondering what's best, it's this one. And you can use it on these two sites just in case you want a more nuanced answer to that. That's why we publish monthly video tool rankings in a free area for community. I'll also include the link to that below. But basically every month we sit together with the team and we re rank all the video tools, add the new ones that were released and posted in sort of this blog article with an image at the top that instantly shows you what the best tools are. We also do this for image tools and all the LLM platforms completely free. We do this to keep on top of this ourselves anyway and just publish the results for anyone else too. Okay, so this is just a quick release from Pika that is a video generator and I thought this one in particular was quite cool. Funny because it allows it to replace specific objects from a video with AI video. And what we're going to do here is I'm just going to throw this over to the editors, but let's give them something to work with like this frying pan that I have right here. So. So have fun replacing this for Edit. I wonder what you will come up with. All right, let's move on to the next one. All right, next up we have a new model from eleven Labs. This is a state of the art speech to text transcription model. Beating out all the other models on the benchmarks and sounding kind of crisp. Hey, can you put me through the sales please? So to be clear, this is a model that transcribes text. So I could take these words that are coming out of my mouth right now and turn them into text in the most accurate and flexible manner. So 99 languages beats every other model on Transcription accuracy, which I guess soon is going to power everything that can use subtitles, from YouTube videos to Netflix videos. I don't know if you noticed, but recently Netflix made some updates and a lot of the movies are now available with 30 different languages of subtitles. This wasn't the case a year ago. Same thing for the voices. A lot of them used to have two, maybe three different voice tracks. I had like English, German and maybe the native language of the film. Now there's like 10, 15 choices of different audio tracks. And Netflix talked about the fact that a lot of that is AI. I don't think all of it, but models like this are going to bring this to every platform. Eventually you can expect YouTube videos that are just transcribed and dubbed perfectly into all languages. It's a wild world we're entering and I just wanted to cover this because it's available now. And also one more side note, I created a video with 11 labs on the channel this week and I just wanted to mention it one more time as the comment section has been extremely positive. People seem to love that video. If you haven't seen it, I recommend you check that out too. In one sentence, it's a free mobile app that takes written text, a link or a PDF file and turns that into a voice file or a podcast that you can listen to to completely free, like legitimately, without restrictions, without a credit card. Nothing is quite amazing. And I love to see that a lot of the viewers of this channel enjoyed that video are enjoying the app. So I just wanted to reference it one more time. 11 labs is just killing it right now. Okay, on to the next one. This one I just had to include in this week's video cuz I just thought to myself that this must be the coolest release ever. And maybe it's just me because I grew up with these games, but they made an agent that is powered by the new Claude and that has access to an emulator of a Game Boy Play one of the original Pokemon games. And not just that they showed this off. There's also a live stream that is actually running ever since. I hope it's still available by the time this video is up. But as of now, there's 1900 people watching an agent that uses Claude play Pokemon in real time. You can look at the Pokemon battles, you can look at it, pick the attacks, navigate the maps, get through the gyms. It's actually doing it and you can see the decisions in real time. So any Pokemon fans out there, I would kind of recommend to check out the stream and see what this is about in real time. Right here. It's thinking about the Pokemon team's health, removing unnecessary information like the previous battle logs, and considering the use of this cape rope to get out of this cave. I don't know, I just thought it was fascinating to see. And when I was five to eight years old, we were obsessing over this. I remember us trading all the Pokemon with the link cables in class. All the girls were kind of mad because they couldn't even walk through the class, but we were just nerding out at 5 years old trying to catch them all. And now AI is doing it by itself. So there you go. The whole thing comes packaged in a beautiful thread that I highly recommend you read. If this caught your attention and if you don't care about Pokemon concretely, see this as a demo of what these models can do today. Because something like this just did not work a year ago. But AI is kind of going for this trajectory. So expect more and more use cases like this being unlocked and we'll be reporting on them here. And now let's move on to the next one, which is also a bit peculiar. This one is called Signs and it's essentially an app that helps you learn sign language with the help of AI I just really wanted to demo this quickly because it's a super fun app and experience here. So I'm just going to start the tutorial right here. Give it access to my cam. Ah, there we are.

Make sure you have the space to do the signs. Adjust your camera and move either closer or further away so that your face fits within the box. If you are right handed, put your right hand into the blue. Excellent. Let me teach you the basics first. The first word I will teach you is the word hello. Let me show you how to do it. Stretch your fingers and tuck your thumb in. Now move your right hand so it touches your head. And now move your hand outwards. And that's it. Now it's your turn to try. Let's go. Start by copying the fingers. Great. Now move your hand outwards. Awesome. Now try doing it two more times. Well done.

Just said hello.

The next word is thank you. With a flat hand, start by touching your chin. And now move your hand outwards towards the person you are saying thank you to give it a go. Excellent. Here is the sign for who. With your finger pointing, put it near your mouth. Now bend your index finger. Straighten it and bend it one last time. That was who. Pointer finger near mouth, index finger down.

I'm not sure what to do with my hands.

Straighten it. One last bend. And that was who? Well done. See, that was easy.

There you go. That's science. I just thought the usage of the camera and how I recognize all the fingers was really interesting. And I suppose you can only expect more and more interactive learning experiences like this popping up as all of these tools become more widely available. If you want to learn sign language, this is a great way to do it, I suppose. And that's all I got this week. I can't wait to bring you a few more videos that go a bit more in depth on these various topics. I hope you find something that was interesting to you and I will see you very soon.