Deep Research/Operator Prompts, Gemini 2.0 Pro & More AI Use Cases

Okay, the AI madness is not stopping because we just had another busy week with releases from OpenAI. I think they released the best feature ever. We'll talk about it. Google following suite with a bunch of new models and furthermore, there's a mobile app from Replit that will build you other mobile apps with their agentic system and it's free to try all of that and so much more. In this week's episode of AI News you can use that will be sort of magentic. At least that's the main theme that connects most of these things because these companies are shipping and and I am here for it to report on all the releases that you could be putting to work today. Let's begin. So the main releases over the past week or two have absolutely been out of OpenAI following some of the deep seek technical breakthroughs and the open sourcing of their thinking. LLM OpenAI has been shipping. Matter of fact, I think in the last two weeks they shipped more products and relevant features to the consumer than they did in all of 2024. Now the two main ones here are Operator that they shipped and even more importantly, the deep robot research feature that we got this week. I created a standalone video on that. It includes multiple example prompts that I ran in there and I strongly encourage you check that out if you haven't seen it yet. This is one of those features that when it came out, I immediately grasped the potency of it. And even in the video I stated that hey, if this was actually freely available and not behind the $200 a month paid plan, this would go more viral than ChatGPT. And not just that. I stand behind that statement. The more I use this thing, the more I realize that you can literally get anywhere from 5 to 20 hours of work done with one prompt if you ask it the right question. So again, there's a separate video focusing on just that and a separate video focusing on just Operator. And we published a third video on OpenAI's operator use cases. That's a really good one, by the way, showing you what you can use their computer using agent for. Okay. But I also want to show you some novel use cases I came across. For example, right here, Chris on X is using it to message various people on Facebook Marketplace. This is an interesting one because it has GPT4O. It can come up with these messages and it can absolutely go through and negotiate down various used articles for you if that's what you want to do. You could be buying old couches, reselling them, whatever you might be doing. And then it transfers that data into Excel sheets. As I mentioned, the dedicated video, it's really good at that. And there's one more that I ran across and I actually can't wait for this. It's February 6th as I'm recording this. And that means within the next few days I need to do my bookkeeping for January. And I've created so many scripts for this in the past just to automate it. I dread it every single month. It only takes like four, four or five hours. But just picking together all these different invoices from all the various sites, so annoying. And creating those little scripts to automate that really takes so much time that it's just not efficient to do it. And if they change anything about the website, you can forget about it. You need to do a new script operator is the perfect use case for this. So I've been excited to see Jason over here on X using it for exactly that. He's using it together with QuickBooks and I'll be trying the same soon. Just wanted to share this with you. Okay. As of interesting prompts for deep research, I want to share this one particular one with you from Kyle here. And he said, find the most interesting Russian research papers from 1960s and 70s that have extremely novel ideas not fleshed out or further studied much since then. Further expand on these ideas drawing from modern discoveries or innovations in either related or unrelated fields. Connect the dots across the whole scientific literature as much as possible. Voila. And then you get a report on all the technologies explored in Russia in the 60s and 70s during the Cold War that never really led anywhere. I wanted to include this as an answer to all people claiming that, hey, just comparing different camera models isn't really a worthy use case. No, it's not. But it's a relatable example and I like to use those if you want something a bit more out there. Well, there you go. How about a 10,000 word long report on Bongert's visual pattern recognition puzzles and some good old mirror matter theory from 1966? All right, there you go. That's a fun template that you could customize to other eras. Maybe give it a more specific theme that you want to explore within the papers. As long as they're in the public domain, this should work pretty well for you. But really the overarching story here is that agents are here and they're accessible to consumers. If you can afford the $200 price tag of ChatGPT Pro, these are the first two agentic products out of OpenAI. Operator is pretty neat. Deep Research is insane. Now here I'm going to share my second thoughts on Operator, because I've been using this thing every day since the release of it. Matter of fact, I think every single grocery order that I've done since the release of Operator happened with Operator. And also there's a few fun ones with like data transferral and working with notion databases that I go to sometimes. But I think the big, big unlock here is the Deep Research feature. I want to just briefly tell you one funny story that occurred to me with Operator yesterday. Basically, I used it to order groceries to my place and I created a new prompt for a new supermarket. So I changed something. So I let Operator do its thing, but one hour later, no groceries arrived. So check the app and it turns out that instead of five bananas, it ordered 30 bananas. Okay? And then it said all the bananas plus the other groceries to a different address in Lisbon. So not just that some random person now received an order with 30 bananas. I mean, what a peculiar shopping choice that is. But also if the order costing €80 and Uber is not doing refunds. This is the most costly mistake I've made so far with Operator. And it just goes to show that this product might work in most cases. I changed up something in the prompt and all of a sudden it made mistakes that cost me real money. But I guess my ultimate point is that the one use case that has been really sticky for me is this grocery ordering stuff. The data transfer stuff is sort of like one off tasks that I give it sometimes and they save me anywhere from 20 to 30 minutes. On the other hand, Deep research, literally when you give it things that you're working on, especially relating to this job or researching all these tools, it saves you hours per prompt. And to give you something practical to take away from this, my favorite thing with Deep research so far has been creating these massive tables where you give it freedom to give it as many columns as you wish. So if you're doing product comparisons or you want to research a topic, asking it for a table with, let's say 15 rows and as many columns as necessary produces incredible outputs. This massive comparison tables that let you learn new things, understand new things, make decisions in the fraction of a second, I included some examples here as B roll while I talk about it. But I just really wanted to highlight that these two products out of OpenAI that came out this week and last week have some staying power. And these are two product categories that we're only going to see more of I'm going to keep covering any alternatives that come out, but right now nothing comes even close to the potency of operator and deep research. I think all the alternatives, whether open source or from Google or whatever to it, they're not potent enough for me to recommend them. But unfortunately these two OpenAI products are behind the $200 paywall that will change over time and we'll cover that as it comes out. All right. In other ChatGPT news, the most minor update, but I did want to mention this. They increased the memory limit within ChatGPT Pro and Teams by 25%, meaning you can save 25% more memories. Now, as I mentioned many times before, this is a beginner to intermediate feature. Advanced users craft their own custom instructions so you can use the memories to gather different commands, but then you put them into the custom instructions so you really have control over what it has. And it doesn't just randomly add new facts as you start up new chats. But for anybody using memories, you can save 25% more of them now. So if you're using that, this is a nice to have and I love that they keep improving chatgpt as that is still my daily driver, if you want to call it that. All right, next up, let's talk about Google shipping. A lot of models with functionalities and a lot of them are freely accessible. Accessible with massive token windows. This is pretty amazing. Even Google is joining the AI release party here. Matter of fact, we have a bunch of new models. Look, if I log into my paid account, you can see 2.0 flash 2.0 flash thinking experimental 2.0 flash thinking experimental with apps. That's the really interesting one we'll talk about here. And 2.0 Pro experimental, their new flagship model. Here. This is on a paid account. If I tab over to me being logged in with a free account, you can see it's basically identical except of 2.0 pro experimental disappearing. But I think that this 2.0 flash thinking experimental with apps is the one that you might care about. So let's talk about what this name means. 2.0 flash means it's their super fast and also super smart model. Thinking means it's a reasoning model, just like OpenAI01 or Now03 or Deepseeks R1, right? They don't give you an answer right away. They think about it first. But this one actually works with their applications. And this has always been one of the main strengths of Google's Gemini web interface. Here, look, I can just pick this one and again, this is on the free plan. You can try this. Right now I'm just logged in with one of my Google accounts and then if you go in here and type at, you're going to see different apps pop up and you can actually go to something like YouTube that you don't even see here. Ah, because I need to enable it. If I go to my settings, I got to make sure YouTube is enabled, which it is. Let's try this again. So it doesn't pop up, but if you just say add YouTube, it works good enough for me. And I could either do a YouTube search in here, or I could pull the transcripts of these videos in here and then I can follow up with something like summarize the content of each one of them concisely in a bullet point list. You can see now it's thinking, it's doing its chain of thought, looking at the different videos, pulling the transcript, running a summary, and then eventually there you go. You have concise summaries of each one of this video. This one builds a treehouse with specialized treehouse hardware for secure tree attachment. It covers the construction and then all the details that it covers. And this one focuses on building the treehouse platform. Aha. So okay, the title does suggest that this one also does the platform only. And if I wanted things like building a ladder that's covered in this video. So for example, I could even go into here and just search for ladder and I'll see that actually the first one and the next to last video are the only two that focus on a ladder, if that's what I'm looking for. So this is a really powerful way to browse YouTube. No, and this is just one of the apps here. You could also use the other ones like the Google Calendar integration, so you can talk to your calendar, summarize what's coming up in the next day, and so on. So this is a real superpower that Google is just better at than anybody else because Obviously they own YouTube and Google Calendar. You could even link this to your Google workspace by enabling this. And then you could work with your emails, your Google Docs, your Google Drive, and even Google Maps. I feel like those are the most useful ones, like for flights and hotels can be good to research. But I think the productivity unlocks with something like YouTube here or having access to all of your mails is just larger. And again, I'm still on the free plan here, right? So if I tab on over to the Pro plan, I have access to 2.0 Pro experimental, which you know is their state of the art LLM now. But like what should I even make of this one? Because there's so many good LLMs now. So I haven't used this one yet. But I can tell you what the word on the streets is, so to say. Basically they say it's good at coding tasks, but I've been told that Sonnet 3.5 and O3 mini high or deepseek are still superior to this when it comes to coding tasks. But it comes with a 2 million context token window that is around 3,000 pages of text. That's a lot of text, probably way more than most people need. And this one has some Internet search capabilities, but what you don't have is access to these applications in here. So that's available because this model set with apps in the end, right? And if I ask it for today's tech news, it's going to pull up October 26, 2023. So this one you would really want to use for long context tasks. We don't need the extra tooling for that. You would switch on over to the thinking model with apps that is also available on the free, free account over here. So yeah, good job Google. And although some of these benchmarks that I guess I haven't even touched on so far aren't like state of the art like other models like Deepseek or O3 are still ahead, I really do think that a lot of people are going to find good usage for this. If you could use this free one with the Google apps and the paid one with the super long context that right now you can't get anywhere else. Like all the ChatGPT models have 128,000 tokens of context versus 2 million right here. Okay, so this next one is really interesting. It's a brand new app on both the Android and the App Store from Replit that allows you to. And here comes build other applications without any coding knowledge on your phone. So if you need a custom app, quite simple one, obviously this app can do it and they don't even charge you money to try it right now. So what you want to do is look for this one called Replit ID 2 app fast. I downloaded this onto my device and we can just give this a quick shot. I already created one in preparation for this to see if this actually works as fluently as I would expect it to. And the good news is that it actually does. So when I just log into my Google account, again, no paid plan here, you can just look at these different templates here can build a Pong game, a piano soundboard, a daily game challenge. I thought this piano soundboard was particularly cool because then you could probably customize it with some of your own custom sound effects, effectively creating a custom soundboard. Again, I didn't try that. I'm just speculating here, but I just selected the prompt, hit send and now I'll just wait. Okay, so that was seconds and it came up with a plan. I'll just approve this and proceed. And it's already creating my little app here now. And there you go. This was about what, two minutes since I opened the app. And I have a functioning web app right here. Let me full screen this and if I play audio, Yeah, I could further customize it. Isn't this just by itself amazing already? If you've never tried one of these agentic application creators like Replit, Agent Cursor or others, this is a great opportunity to get your hands on this, to at least get a feeling for how they work. And it's free right now. Not sure how long this is going to remain free, but there you go. Replit mobile app that creates other apps. Next up, I want to quickly talk about this story. I mean it will be something that we will be able to use. The European Union is investing 56% million into building an open source LLM that speaks all European languages. I mean, what is going on here? ChatGPT is excellent at speaking all European languages. I can tell you from working with a Slovakian company. All the models speak perfect Slovak. And Slovakia, if you're not familiar, is one of the smallest countries in Europe. 5 million people in total. And ChatGPT does an amazing job at speaking it. Even the voice assistant does that. Why does the European Union need to invest $56 million into developing an open source LLM? We have all of these amazing LLMs coming out. Last week we featured the new Mistral small free. Actually it turns out that this generated quite a bit of buzz and seems to be one of the best non thinking models, non reasoning models that you can run locally. 70B performance at a 24B size. This is a French company and these models also speak all European languages. So what's going on here except of politics? I mean they want to be seen as AI ready, but really all this is is virtual signaling. And I know that the bulk of this audience is from the U.S. but still, I personally am born and raised in Europe and seeing this kind of breaks my heart. I wish there was something I could do to help this beautiful Beautiful continent, not to fall too far behind. I don't know. If you're aware of some opportunities like that, let me know. Feel free to shoot me an email. And talking about Mistral, I want to mention one more thing. They actually shipped updates to Le Chat or Le Chat, like the cat in French, whatever it might be. This thing is super speedy. Now, Mistral Large has always been very capable. So let's give this a shot. And with the new. What the heck. I barely managed to finish my sentence. What? Okay, once more, hit enter. One, two, three. Holy moly. Okay, that was super fast. I got to enable the canvas feature here, like so. And let's see how it works with that. And instantly look, this is almost instant. Pretty impressive. It does it. And they have this artifacts feature. That's what I was trying to say here. And all of this is super free. They also have pro plans now. But the nice thing about Mistral's chat interface here has always been that you got their models, their version of GPTs and everything at no cost. Really. I guess you have pro plans if you're going to use a lot, but there you go. If you're looking for something that is 12 times faster than GPT4O for cheaper, that might have a few less features, but it's free to use, you can give this one a shot. Okay, so next up we have an app called Refusion. And basically this is the first AI music generator that works pretty much as well as some of the paid options. But this thing is completely free, at least for now. So if you ever wanted to create some songs but you were deterred by the paid walls, check this out. Because the quality is topnotch. These vocals sound like humans and not like AI. Just sounds good. Okay, one more example. Impressive. And not just vocals. Check this one out. This one is of. I don't know about you, I'm feeling this one. Yeah. Gotta agree with Robbie Delta here. Making this free is pretty legendary. Legendary. All right, next up, I want to briefly show you something that me and the team actually created over the past months. It's a little self assessment quiz that takes less than a minute that will Recommend custom generative AI learning resources for you. And it's 100% free. Check it out. You can just go to the first link in the description. I can answer these questions. It's just a few super basic questions and you will receive free recommendations. One of them is a list of prompts that we have found to work really well for your particular level and interest. For example, these business Focused prompts work really well, like applying these real world case studies to your own business. Now, these are the more advanced ones. If you're already experienced with AI. If you have less experience, we'll give you simpler prompts to to start with. And it comes at no cost. Just as a part of filling this out, you opt into our weekly newsletter. That's it. So if you like what you see in these videos, go ahead and get your custom prompt recommendations, a free guide, and a course recommendation within our community as a part of it for free right now. First link in the description. Okay, let's look at the next piece of AI news that you can use. Next up, we have the story, and maybe not entirely something you can use, but something you can attempt. So Anthropic is running a jailbreaking challenge. It is eight levels, and if you pass all of them, there's cash prizes there for you. It's pretty straightforward. There's a blog article explaining this. And then right here you can take the jailbreaking challenge. This is the first question right here. And you can go try this for yourself if you have some jailbreaking skills. But I wanted to point out this little drama that has been unfolding on X because Pliny the Liberator, one of the popular jailbreakers on X, actually challenged Anthropic back to open source the entire data set. And if they do that, he'll do all eight levels on his live stream. And then Jan from Anthropic comes back saying, we don't want to open source the data set, but we might provide a different incentive. Stay tuned. And Pliny right here with the most legendary answer of them all. It's going to be legend. Wait for it. I'm allergic to money, so don't bother. My incentives are aligned with what's best for the community. And the future of AI are yours. Gary Ball is in your courtyard. So I'm going to be following this pretty interesting story right here. And if you want to give jailbreaking these eight questions a shot, you can try. According to Anthropic, right now, nobody solved anything beyond level three. Okay, another quick story that I wanted to throw in here is one of the top video models getting one of the best features that was originally released by Runway. But now, as to be expected, all these other video generators are also getting some of the tooling. This one is called director mode, and you can basically control the camera. And this one also comes with various presets. What I really like about some of these video generation tools is that they come with comprehensive tutorials. So right here there's an entire notion page they made that shows you all the different presets, how to use them, what you can expect from a shot of this goat if you apply the push in preset on it. Okay, hold up. My bad. That's actually a sheep that is being pushed in on. So there you go. If you're AI video user, I think this is one of the most convenient thing they can add to your toolkit. You get to actually control how the camera moves in the scene. Okay, here's a story that I sort of wanted to feature, but I wasn't a little sure I'm doing it anyway, and I would love your feedback on this. It's basically humanoid robots entering the market and this could be considered AI. It's the definitely news and you can also use it. So it sort of hits the criteria of this show. But then on the other hand, it's more like hardware and we look at a lot of software, apps and new features here. Anyway, if you're interested in this, out of all the things I've personally seen, I really enjoyed this little segment from popular live streamer that you might know named Kaisenat. And he basically bought one of these $70,000 robots and unboxed it on stream and started playing with it, you know, started bullying a little bit, as you do with these humanoids, I suppose. And I just wanted to point your attention towards this. This is a category that I personally am closely following, but I'm not entirely sure that it fits the theme of this channel. But heck, when these things come out, I would love to review and test them for you here. At the very least, I'll be throwing in a story like this here and there. And I just figured if you don't want to spend $70,000 yourself and want to see what this can basically do and what the unboxing experience of something like this looks like. I mean, it's a new product category to something that is very close to generative AI applications. I'd recommend to check out this video. I had a good time watching this. All right. And that's really everything I have for you today. To round things out, I want to mention that we did a live stream earlier this week which looked at various thinking model use cases. So if you're looking at all these releases like Gemini Advanced, that is thinking like O3 mini or even R1, and you're just wondering, what are some unique use cases for this? What could I be using this for? I ran a one hour long live stream on the channel where we looked at various community challenge submissions of what people watching this channel are actually using this for. I picked the five best ones, covered them in one hour and I truly believe that everybody can learn something new in there about how to best utilize these thinking models for yourself. So go check out that stream, maybe put it on while you commute or cook. And other than that, I will see you next week.