Gemini 3 Deep Think Reviewed & More AI Use Cases

Welcome to another crazy week in AI. So the big boys are currently competing for the spot of the best model. So we'll be talking about that because There's a new ChatGPT model, a new Gemini model, and it pushed what was possible on most scales. But beyond that, we have some prompting techniques and really creative apps that I think you will appreciate in this week's episode of AI News yous can use the show that looks at all of this AI madness. We filter it for what matters, try it all, and then I get to you tell you about it. And I want to start our discussion off this week by talking about the big model providers. I mean, it's really the big three of ChatGPT, Gemini and Claude. These are the three big assistants. There's a few more that are worth considering, sure. But most people, if you talk to them and ask them what AI assistant they use day to day, they will say one of these free. And this week we got massive announcements out of both ChatGPT and Gemini releasing their best model that goes beyond anything that their models before did. In the case of ChatGPT, it's 5.2, including a 5.2 Pro and even more capable models in their API. In the case of Gemini, it's their Deep Think model. And then Claude has their opus 4.5 that came out a while ago. It doesn't have anything like the Pro mode. It kind of just thinks for a minute where these think for 15, 20 minutes sometimes both of these. But they're all gunning for the same thing, being the best AI enabled assistant on the web. And I created a standalone video yesterday contrasting them. Okay, so. So if you want to see examples and if you want to see me go into various categories of things these models can do and directly showing you the result in ChatGPT Pro or the normal model that you have on the free plan and then also going into Gemini and using this Deep Think feature which is hidden Behind, I believe $150 a month subscription, something like that, then you can check out that video. I also compare Claude where it's appropriate here in this video I just wanted to give you this brief overview of what has been happening, which I did now. And I also want to kind of point out the verdict of all of that testing. I think the nuances matter and that video is very interesting. I had a lot of fun creating it because we really show and contrast what's possible in specific examples. But overall, I would say this. If you just need one app, it's ChatGPT and that's hard to argue with because of the tooling. They have the best dictation function and the best interactive voice mode. The Deep Research is more solid than the one in Claude. For Deep Research. I think Gemini wins though. The image generation is okay. I think they're also. Gemini clearly wins with nada banana. But the model is smart in a business sense where I love it as a brainstorming partner and planning partner. And then you have things like projects that you don't have in Gemini and it just brings all of this together. It's one app that kind of has it all. And now they have a model that is on par with the best Gemini model. So there's really no reason except of maybe image generation and slightly better Deep Research to go to Gemini at this point. That's why they released it. That's the point I want to get across. They were slightly behind in terms of model capability. Now, now they kind of match Gemini. And because they have all the users and all the tooling and just the best platform overall, I would say there's no reason for people to switch. And that's the thought behind this. Now, that's not to say that the Gemini models are not good. They're excellent. And again, if you want examples and benchmarks, you can go check out the other video I uploaded yesterday. It's a really strong contender. And then Claude, people prefer it for its code generation. It just makes sense the way it works. It lacks some features, it doesn't even have image generation. But the way it writes code and writes text, it's just very human and just sense. So for me personally, I find myself day to day between ChatGPT and Claude and I was going more and more to Gemini, but now with the new 5.2 model, I think it's back to ChatGPT just because I'm so used to it. Before we move on, I want to correct one more thing. I actually made a mistake in the video that I created yesterday where I was saying write me an essay and I made a mistake by saying essay. I wanted to say email to my boss about the broken coffee machine and I was trying to do the video in one flow and I made this mistake of saying essay essay instead of email. So it was needlessly long. And then I was saying I wonder why this email is so long. But obviously it was me prompting it. If I prompted correctly, GPT 5.2 auto shows something that I would expect a very neutral tone, a follow up prompt down here, and placeholders when appropriate. The stuff that I usually look for. So overall, if I were to add something, then in terms of writing, it's very neutral. I don't think it's very human. I think for writing, I still prefer the default what I get from Claude out of the box. And then with Gemini, I don't know, I'm just not feeling this writing style even when I prompt for different styles. Probably a personal preference thing, but I don't know, it's just a bit more or appears to be malfunctioning. I would never say that write that. Anyway, I just sort of wanted to add that to yesterday's video. But yeah, in any one of these you have new models and now you know about it. So go use them. And if you're in chat, check out how well it can create PowerPoint presentations and excel sheets. Now. It's extremely good at that now, which did not used to be the case. Hey, quick reminder that if you're enjoying these stories, subscribe to the channel, because I do this every single Friday. So for the next story, I really want to tell you about this prompting trick that I found. I've been testing and it just has proven to be a gem. It's something that was tweeted by Andrej Karpathy. I've seen variations of this previously. This idea of a council of elders, council of advisors is nothing new, but this way of prompting for it is. And maybe you haven't seen it before. So I wanted to bring it to your attention, which is rather than telling it, hey, you're an expert in xyz. Tell it. What would be a good group of people to explore xyz? What would they say? And I actually tested this inside of the Deep Think model and I ran the same prompt twice. Once I tried this new approach. What would be a good group of people to explore the best way to talk to strangers? A very open and kind of interesting question. What would they say? Right. So this is one approach and I ran this through Google's Deep Think. Same idea, different formulation of the question. You're a communications expert, explore the best way to talk to strangers. So this is the traditional approach and I want you to look at the result here. We're not going to review every line because that would take too long, but basically it goes ahead and gives you an answer that says, okay. As a communications expert, I would recommend using the following frameworks. 1. You need this mindset, you need this type of opener. You need to read these type of signals. Here's a common mistake. Here's some tactics and here's how you exit. All right. Actually pretty good advice. I read through this whole thing. All of this has merit, but as soon as you switch over to this other type of prompting, and I think this works especially well on a model like this Deep think, where it really thinks for a long amount of time and works on it and gets to rethink the perspectives, it gives a completely different answer. It first forms something like a council of connection. That's what it calls it. And it just brings in various experts as if you were having a roundtable discussion. And all of a sudden, these experts give their opinion. And let me tell you, the structure of the advice is the same. Phase one, it starts with the mindset, and then you go into how to approach and how to open and deepen the connection and how to exit. The structure is the same, but each one of these paragraphs is different. Comes from a specific person, an expert on that topic. And I want to say this concept is not entirely novel. Summoning a council of experts is something that I've done before, but it's rather tedious. You kind of have to create these different Personas and turn them either into GPTs or prompt in a way where you set up the Persona and then let them talk to each other. This approach is really simple. And all of a sudden, you're not getting advice from one communications expert, but you're getting these different perspectives, like approach advice from Chris Voss, the FBI negotiator who's on all these podcasts, actually read his book, too, which was amazing. That's a lie. Actually, listen to it. Still a great book. And most importantly, I have to say, I found this version of it where different experts were giving advice actually more actionable. It resonated more with me. I read this, and I was like, wow, that's true. And maybe that's partially due to the delivery, because all of a sudden, sort of the book author that you know is speaking to you or a specific person is giving you advice. But then also the pieces of advice, they were just more varied, more colorful, rather than the communications expert kind of following a more rigid outline. That's what I would say. So both of them worked super well. But in general, I'm going to start using this a whole lot more. And, yeah, this show is about AI use cases and how to use them. And this is a fantastic way how to get more out of the prompts that you already have. Okay, see what's next. Okay, so if you've been following this channel, you'll know that I'm using this tool called Whisper Flow for audio transcription of my voice into whatever tool I'm using. It's one of these things that once you start using it across all your devices, including your phone and tablet, you kind of start wondering how you ever went with without it, because all you need is a microphone and you can type perfectly. Whereas, let's be honest, a lot of the built in dictation tools on the iPhone or in some AI apps don't even get close to being perfect. And at this point, I just want to say a big thank you to Whisper Flow for sponsoring this video and making this production possible. So originally I started using Whisper Flow in applications that didn't have voice dictation at all. And by now I got so used to it that I even use it in applications that have voice dictation. In my experience, Flow tends to work better, which means it's more precise and actually mirrors what I say rather than misinterpreting things then a lot of the existing dictation tools. If you ever use the dictation on the iPhone, you just know how imprecise that is by default. And Whisper Flow nails it. Hey, what's up? I'm just trying the dictation function on my MacBook Pro. This is the built in one from Apple. I wonder how well it works in practice. New line. See you later, Igor. Okay, now let's try Whisper Flow. Hey, what's up? I'm just trying the dictation function on my MacBook Pro. This is the one from Whisperflow. I wonder how well it works in practice. New line. See you later, Igor. And see, this is what I mean. The formatting, the punctuation, the new line, the greeting, all of this is a finished message rather than this being a draft. That is close, ish. But then I have to go in and actually add in the punctuation, separate it out, formatted like a message. Whisper Flow just knows. And here's a bonus feature that I like. Sometimes I could start talking in English above ein meier Deutschland undes function. So just switch to German in the middle of the sentence and it gets it. So as I told you, I've been using Flow for all my voice dictation needs across all apps and all devices. It's just this mental ease of knowing that I can dictate wherever I am without having to consider the specifics of all the different platforms. Just know it's there and it works. And if I have a lot of thoughts, I can type at the speed of my thoughts. Which is something that I can't do with the keyboard no matter how fast I type. So if you've ever been in a situation where you feel like your brain is moving way faster than you can type, this is the tool for you. So go check out Flow today with the link at the top of the this video's description, and you can use the code AIAdvantage for an extra month of Flow Pro. And once again, a big thank you to Flow for sponsoring this video. Now, let's get back to the next piece of AI news that you can use. All right, so next up, we have Adobe photoshop inside of ChatGPT. Yes, I was confused too. They literally added an app from Adobe so you can use photoshop from within ChatGPT. And I actually can't wait to try this. Now, obviously you do need a Adobe Creative Cloud subscription for this, and I know, I know a lot of controversy around their subscription model and all, but Photoshop is the photo editing app, and I really want to see how this works because I've heard good things. So basically you just go to your settings, go to apps and connections, and then you enable Adobe Photoshop like so. Then if I'm in here, I can click plus and down here under more Adobe Photoshop. Okay, I'll just bring in this random picture of Sylvester Stallone that is laying on my desktop. Excellent. Okay, let's just say blur the background. Something very simple that involves a filter and some selection. Go. If you're not familiar, these apps connect third party applications right into here. And usually they also bring the interface right into here. So it's not just working somewhere on the cloud. It also brings some of the capabilities which obviously in Photoshop's case are so, so deep. Extremely deep. And now they're supposed to be in here. So I really want to see how this looks now connected to the app. And okay, that's a blurred background and I get a slider on how much blur I want. This is actually pretty good. And the selection is very good too. Look at that. Even if I'm nitpicky, this is good. Okay, let's try something more. Make his hair blonde. Here's the thinking process. We must use select hair. Use selection by prompt with prompt hair. Oh, so it's using AI tools to complete this. Can auto detect hair as a separate object. Totally normal with older movies still. So, yeah, this is just using AI in the process. Obviously those tools have become really capable in Photoshop and now you have a new way to use them. Interesting. Okay, so it actually went ahead and asked me a follow up on if we want the whole thing or just the hair, which I mistyped too, but that's okay. Oh, and he asked me again. Okay, let's say two. It's apparently having troubles with selecting the hair because, well, you can barely tell that this is hair. It's just a black blob. But. Okay, this might not be perfect, but I want to see what it does. Oh. So I've pushed it beyond the capabilities here. Okay. We can make the mask manually, I guess, and then I would need to change the entire thing so it didn't manage to mask this. So obviously there's going to be limitations here. Okay. We want Hulk Stallone. Look at that. These are confusing times. And if I wanted, I could click Open in Photoshop and move out. So this is really for quick adjustments. And then in here we have the trusty interface. And here I could now probably brush in the masks with something like this. I could invert this and then pencil this in. Well, obviously this is not doing much because we need to colorize the hair. Okay. Human saturation is probably not the way to do this, but you get the idea. At this point, I'm just playing with Photoshop, so the fact that you kick things off, but most of the work is still going to happen in here. So, you know, nice to have. Not a complete game changer. They announced a few more things, actually. You know, I have access to Adobe Express, which is their version of Canva, essentially an all in one image editing studio with a bunch of presets inside of Photoshop. So you could use that as an app too. Probably more suitable than Photoshop because these basic edits are not what you go to Photoshop for. You go for the depth and the ability to manage layers, whereas Adobe Express just has quick presets that AI can probably deal with well. But that's going to be similar to the Canva plugin, which we looked at before, just like all other apps. Good to kick off the process, but you'll end up in the app anyway. So, yeah, that's pretty much it for this one. Okay, next up, I want to show you an update to an app that I've shown you before, but they updated with a feature that I think you should know about. This thing is from Google, it's free, it's called Mixboard. And basically, if you go into it with this new update that now has Nano Banana built into it, you can just add various images and then turn them into presentations with the click of A button and. And it works so well because nanobanana Pro is so good at creating slides with text and all that. Quick note. I did have to use a VPN to access this from Europe, but they are expanding this and a lot more countries have this rather than just us who had it on release. Okay, let's create a new one. And I want to design my new YouTube studio. We'll just keep this brief. I'll go ahead and add a bunch of images and cut to that and then I'll show you the new presentations feature with Nada Banana Pro in it. Okay. So from about three prompts, I arrived at a mood board like this. Now it creates multiple things at once and I'll just take some images that I really like for studio inspiration while holding the Shift button. And now this is the new feature. Up here you can see Transform. And if I click this, you will see it will want to turn this into a presentation. We'll be doing a visual deck. What story do we want to tell? A pitch deck for a new YouTube studio design. All right, and then what do we wanted? Like let's say cinematic. Okay. Transform. Oh. So it did take about 15 minutes, but it's ready now. Let's have a look. Every creator starts with blank cameras. Passionate. Oh. So now it used all of them, maybe some blueprints, and it's really telling a story. Okay. But it's using all the images I had there rather than just the selected ones. Aha. So I slightly misunderstood it, but that's okay because you can learn from this. So basically what he did in the demo is he selected a few images to create a new one. Right. But then in the end, when he goes into presentation mode, it doesn't matter what you select, it just takes all of them and turns that into a story. So if you don't want something in a story, you'll have to delete it off the mix board. And then you too can get a nice little cohesive presentation that tells the full story in the way you want, as it did in the demo video here. Live and learn. But the fact remains that this is really impressive. Obviously, nanobananapro brings this together in a brand new way. And you can see how this is a pitch deck, just like I asked for. And then ends on a call to action. You just click download and there it is. A YouTube studio design PDF with 16 pages just like that. Honestly, when it comes to working with Nano Banana Pro, this is actually so much more efficient than creating images one by one. Just say what Kind of board you want. And it makes 10. Follow up, it creates three, another three. You delete, mix and match. Boom. One button presentation. What an amazing app to have free access to. Okay, so let's have a look at this week's quick hits. Stories that we maybe don't have to try in depth, but are still worth your attention. Starting with the Grok hackathon that happened in San Francisco, I believe. Kind of just assuming, but isn't every hackathon ever in some San Francisco? Yeah. All right, so these are some of the results that really got me. First of all, this one that went Giga viral. Actually, you have to see this. It's using Grok's video model to generate ads that look as if they were a part of the show. I mean, just have a look at the scene from Suits. If you go to the ad segment, all of a sudden Harvey is coming out of the building. Jessica, what are you doing? Oh, something inside of me cringed so hard. Is this the future of advertisement? There's going to be no escaping it. It's easy to look away from a billboard. How do you look away from your favorite show if you don't even know if the scene is real or not? This is not just product placement. This is so much more aggressive. But if you know capitalism, you know that this is inevitable. Well, the cat is out of the bag, I suppose. Congratulations to the hackathon participants here. What else we got? I like this one. Haggle, an autonomous voice agent that calls service providers for you using smart negotiation tactics to secure the best deal. Is that what the future of negotiating and services is going to look like? Is it just AI agents going to be calling each other and haggling the best price? I suppose it's kind of cool to know about this early, maybe get ahead of the trend or. I mean, that's all interesting, but again, this Grok ads thing is still. I don't know, just doesn't feel right. Look at this scene from the Friends where Joey's picking up headphones that didn't even exist back then. That's just wrong. Next up, a report from OpenAI, which is kind of interesting. It analyzed all of their enterprise AI activity. There's a few big takeaways, and I think the main one is this that I want to highlight. You can look at the whole thing, but basically they found that the frontier workers are sending six times more messages, and the frontier firms are sending two times as many messages per per seat than the median enterprise. Meaning that there is a widening gap in between leaders and laggards in the space in terms of usage. And what I found really interesting here was the fact that they measured the leaders in usage by how many messages they sent. And I actually thought about it for a second and it makes so much sense. If you communicate more with the AI, you're going to get more results, you're going to get more experience, you're going to become a power user quicker, rather than somebody who's a bit shy with it and doesn't use it as much. And then also they mentioned that enterprise users report saving 40 to 60 minutes per day and being able to complete new technical tasks such as data analysis and coding, which they couldn't do before. And I think this is accurate. Saving about an hour a day with some of the basic use cases is totally reasonable. I think if you go deeper, kind of create and do new use cases that you wouldn't be doing otherwise, which gives you back more than an hour a day. But this is just across all the enterprises they looked at. I think that makes a lot of sense. And then if you put it in contrast with the fact that a lot of people claim that, you know, well, 10 times your output, meaning you go from eight hours to 80 a day, it's just not reasonable. It's more like, yeah, it'll add 20, 30% to your output, maybe 50. OpenAI agrees. And also Anthropic did a similar report where they interviewed 1250 professionals about how they work with AI. I always enjoy looking at these and especially Anthropic makes these beautiful graphs that really show you where it's making a difference. And this one shows how optimistic or pessimistic people were about various use cases. Yet again, a different lens to look at this. And I really like that it showed that, hey, for professional communication, they're super optimistic about it, right? This is one of the big use cases. It's so damn good at writing and summarizing things, optimizing business, optimizing project workflows, teaching materials. These are the big use cases that we know about. But then on the other end it's something like false stations where they're very unconfident, but they made this small. And also writing independence. Writing is different than professional communication. And I agree, I trusted less with writing something that matters rather than a simple email. There's way more here. Like you can look at a view for creatives or scientists, where scientists have a lot of security concerns, but creatives really enjoy it for content creation, obviously just an overall interesting report to look at all. Right, then. That's pretty much everything I have for you today. I hope you found something that was interesting or inspiring to you. And with that being said, my name is Igor Pagani, and I hope you have a wonderful week.