o3 Model Announced, Free AI Search & More AI Use Cases

All right, we have a massive week of AI news, mainly from OpenAI this week we're going to be recapping everything they announced during the last five days of OpenAI Christmas live streams with a few tips and tricks on why I'm excited about some of these announcements. And then we have some super fun or state of the art applications in the imaging and video categories all packaged for you. In this week's episode of AI News. You can use the show that summarizes all the AI news from last week that you can actually put to work today, starting out with OpenAI dev day. And let me start with last Friday. The recording cutoff for this is usually on Thursday Thursdays, so although we upload on Friday, last Friday's Day 7. That introduced ChatGPT projects was something that I wasn't able to talk about anymore in last week's episode, but I considered this such a convenient feature that could literally add value to so many people if they realize what features hide in here or how to use this, that I decided to create a standalone video on this. We're still editing that. It's coming within the next few days. It's a full breakdown of various use cases me and the team have for this feature and then also a comparison between Claude and ChatGPT. So subscribe to the channel and keep your eyes peeled for that video coming very soon. That's all I have to say about that for today. Besides this being a feature that I'm already migrating a lot of my custom instructions into. Plus, just like in Claude, I'm turning some of my favorite prompts into standalone projects. Fantastic way to organize your ChatGPT interface. Dedicated video on this coming soon. Okay then on Monday we had day eight which was all about ChatGPT search. It really boils down to three points. First of all, an improved search experience where you have embedded the maps or videos directly inside of the chat chat interface. No need to leave the application. Also, as a part of the search improvement, they improved the latency. If you're watching every week, you might remember that this was my biggest complaint When I tested search for seven days straight. I turned on the extension, didn't use Google for a week straight and my result was hey, it actually finds the links that you need. It's way better and convenient than I expected it to be. Except of the speed. The speed was the reason why I installed the extension. Again, there was just a 1 to 2 second delay between my query and the results. This got lowered. Now, not quite as speedy as Google Search, but Significantly faster. So there you go, some major improvements to the search functionality. Secondly, they're adding search capabilities to advanced voice mode. This still seems to be rolling out, but combined with the Day 11 announcements from today and we'll talk about soon, this is a really amazing upgrade. And then lastly, they're opening up search to all free users. So if you don't have a Pro account and you were Interested in trying ChatGPT search, now you can get the benefits of all these upgrades to search for on a free account. Heck, you don't even need to log in anymore. And I'd probably recommend it because then you have a search history. So if you were ever curious about AI search, well, no better time than now to give it a shot. All right, on day nine we had the mini Dev day with a bunch of developer focused updates. The most significant one that will interest the widest audience is probably the full release of O1 through the APIs. So now it can finally be built into various applications. This includes function calling meaning ow. One can reason over something and then actually do things by calling external code. This is at first only rolling out to tier 5 OpenAI accounts, but I can't wait to see what various developers come up with. This is the first time they can get their hands on 01 outside of ChatGPT. There are a bunch of other updates and I'm sure all the devs will have checked out. Not just the announcement but also a lot of the live demos that they uploaded. Amongst two talks with Sam Altman, they uploaded over 30 videos of different developers showcasing what they built at their different dev days over the past months. So if you're into that, that's definitely worth checking out. And Also there's almost 30 minute long live demo where they build with the new advanced Voice API on stage. Something that might even be interesting to a non technical audience to actually see the difference of how voiced worked up until now and how fluid it is with advanced voice mode through the API. So give this a few weeks that you might see a lot more applications include their reasoning models inside of other apps. Also I wanted to point out this AMA they held inside of their OpenAI community here and and also I would like to kind of highlight that the dev community is so much friendlier than the wide chatgpt audience. If you look at the comment section of the dev day and this ama, it's super civil and constructive, rather just being filled with random demands. Anyway, if you want more details on what they shipped for developers, I highly recommend this ama. Although they did ignore this one answer that I personally was really curious about and that's if they're ever going to do a proper embeddings model or multimodal embeddings sometime soon. But with all this talk about agentic capabilities in the Fireside Chat and the AMA they uploaded with Sam Altman, I reckon this is just just a question of time. And talking about ChatGPT inside of new apps, they tend disappointed a lot of people, which I was a bit confused about. I thought this was really cool. They brought chatgpt to the phone and you could do this from everywhere without having an account. Sort of a good thing to know for anywhere you are. You can access ChatGPT via phone in the US, internationally, you can message it via WhatsApp and it will write back as the experience is free to everybody now, no account required. This is one of the opportunities that open up. I thought it was sort of neat, but a lot of these comment sections seem to be full of hate and people with high expectations or just bitterness around the AI industry anyway. I mean at this point maybe I should just remind everybody that hey, they don't owe us anything. They're doing 12 days of announcements and I personally am just grateful for what they are releasing. Some people just bring these sky high expectations to the table and are then disappointed that they're not launching AGI but just chatgpt via phone call. Anyway, if you're in the us, you can give this a try. I think the number is 1, 800 and then the letters chatgpt which is, which is 2, 4, 2. I have to look at the keypad 1-800-242-8478. All right, as in the letters on the different keys. Which brings me to today's announcement day 11 and this one is really cool. And we're going to demo this right now because they added support for note taking apps off the Mac desktop app. It's unfortunately not rolled out to Windows yet. It's a question of time, which is a bit disappointing. It will take a little bit of time. But what we got here is an upgrade to this desktop app that is shipping shipped now all around the world. And what it can do is if you're inside of another note taking app like let's say Notion here, for example, I could just press option shift 1 to activate this app specific ChatGPT assistant. And as you can see this is working with Notion. Now it can see my screen and now for the very first time it's bringing together a lot of the things we talked about. As you can see, I have Internet search here. I have Advanced voice mode here. It sees my screen. So I could activate Advanced voice mode and and start talking to chatgpt via my voice and have it assist me while I work inside of my note taking application. Heck, inside of Notion, I have the entire operating system for my entire company, for the entire community. Everything is in there. And whenever I need help, I don't need to screenshot, bring things over. I need to Google things. Everything is just within this one interface. Let's give it a shot. Hey there. Let's just start out by what can you see right here? Ho ho ho. I can see you're looking at a Notion page about an AI prompt library. It includes steps, steps to duplicate a template, watch a video tutorial, and access various resources related to prompts and prompt engineering. There's also information on courses, ebooks, and various online resources. If you need help now. Okay, hold up. I don't think I can take Santa here for this. Hey there. Let's go with Spruce here. And as you can see, as you might have been able to tell already, Spruce or the assistant here can see the entire page, not just where I'm at. It could see the top, it could see the bottom. And as it can see the application, not just your screen. Okay, what kind of categories of prompts can you find inside of this Updated template for 2025? The updated 2025 AI prompt library includes categories like education, business, entertainment, health, and more. These categories cover a wide range of topics to suit different needs. Excellent. So as you can see, this works great. Okay, so maybe a more practical example would be me opening up this page with a database of all the prompts of the weeks we featured in our newsletter over the course of the the past one and a half years. I'm just going to say option shift one. I'm going to activate advanced voice mode and ask it something like. All right, chatgpt. So I'm about to do some creative writing. It's going to be journaling. Quite personal. What kind of prompts could help me from this page that you can see, pick out free. For personal journaling, you might find prompts like enhance your writing with constructive feedback, simplify complex topics for any age group, and create engaging classroom activities. Helpful. These prompts can inspire reflection and creativity and personal growth in your journaling. Wow, look at that. What a great way to search over all of this. I don't need to read every line and now I could just go in press command F and search for enhanced personal. Huh? Interesting. There's actually no prompt with that exact name. Let's try this one more time. I'm a bit skeptical here. I'm sorry. Hey. But I'm having trouble responding right now. Please wait a bit. Okay. Seems like advanced voice is having troubles right now. I'm going to give this one more shot just via text. Let's see. Okay. Similar prompt. Pick one of these prompts that could help me with personal journaling. Let's see if it correctly sees the entire page. Okay, I need to be more specific. Pick one from this Notion page. Not just generally, but it is working with Notion and it is paired. It's a bit weird that it can see it. No. One last attempt before I give up. Okay, it's working with Notion. It's a fresh chat. Let's give this another shot. For creative journaling, you might try prompts like reflect on your day, describe a memory, or explore your feelings. How can I assist you further? Interesting. So it seems like if it's a normal page, it can see the entire content, but if it's a database, it cannot actually see inside of the database, which is sort of a huge deal inside of Notion. So I suppose if you use this with something like notes, it would work just fine. But instead of Notion, it cannot see databases, which is a serious limitation. Nevertheless, I think that if you're aware of that limitation and start playing with this, a little bit advanced voice, that will be also able to access Internet search, which brings in verifiable sources, and it sees your screen, not just your screen, your entire application. There's some serious power that can be unlocked here. And just off the top of my head, I think if you're learning something new, this could be an extremely powerful tool. You could have a tutor that talks to you, that sees your screen, gets all the context from the app, and might even have some custom instructions, and that could literally work with you. So yeah, it's good to be aware of this limitation of the database itself side of notion. But other than that, this looks very powerful for all the developers out there. They added support for various terminals now too. So slowly but surely they're expanding the capabilities of the ChatGPT desktop app, starting with the Mac 1 Windows coming soon. Supposedly all of this is live right now. Go give it a shot. And that wraps up everything that OpenAI announced. But we have more this week. Hey, everyone, Post editing. Igor here. I just got back from the mountain to my grandma's place. The point point here is this. They just announced O3 mini and O3. And I wanted to quickly add a little comment to round out the 12 days of OpenAI on this. Have you seen those benchmarks? I mean, we know benchmarks are not everything, but like we usually say that on subjective things like productivity and writing, on things like coding and mathematics, they actually do mean a lot. And the gains in the ELO scores, we'll put them up on the screen right now are in sane. They even introduced these AGI benchmarks where it just blows everything that O1 did out of the water. O3 is essentially 01 on steroids. And my first thoughts on who will this be useful for are, hey, if you're not using 01 already on a regular basis, this won't change the game for you. For now, people might build interesting things with it, but if you're not doing math or coding with the models, it's not a game changer. If you're a developer, big. If you're using it for math or science, also very big. I think what they said sums it up. Well, we'll just have to rethink our AI intuition. We'll have to reconsider what AI can do for you in 2025 and my hope is we'll be doing that together. I'll certainly do my best to take you along for the ride here, but yeah, just wanted to share the news on O3. It's coming in late January by the way, so nothing to worry about for now. But there you go, check out the live stream, check out some of the benchmarks they published and yeah, for now all we have is the benchmarks. Looks impressive. Head and shoulders above anything we've seen yet. Okay, on to the next one. GitHub Copilot actually introduced a free tier. Anybody who has been using this before will know that this was a paid service and there were some other competitors that offered a free one. Now also GitHub Copilot gives you free access to this and this only includes up to 2,000 completions and 50 chat requests per month. So there's a rate limit. Obviously somebody using this for a full time job is not going to be able to use this for free. On one account that is. But it is an improvement. As before it was only paid. This is just for the people out there who might be using GitHub copilot or thought about using it. Well, now there's a free tier. Okay, so next up we have a big update from a platform that we featured a few weeks ago and that is chatbase, the sponsor of Today's video Last time I showed you how to easily build a chatbot, I could connect to a knowledge base like Notion so it can automatically stay up to date with your latest information in real time. That can be very useful in the right situations. But what they added now takes it to another level. Their new Actions feature allows you to build chat bots that can actually do things for you or your customer. You can now integrate chat based bots with tools that you're already using personally. My favorite actions here are the ones that connected to Slack, Calendly and Stripe as these are tools we use in our workflow and being able to set up a suite of chatbots that connect to these tools automatically can easily save you time on a regular basis basis. So for example, if in a conversation with the bots, the topic of hey, when is eager available comes up, the chat based chatbot can pull info from your calendly and immediately provide the factually correct information on when there are slots open in your calendar. If they decide to book a slot, it can do that too. And then you can link it to Slack so your team is notified that a new appointment has been booked. You could even manage payment details or update subscriptions directly from within the bot. And if that isn't enough, they offer an option to connect two custom APIs, meaning that you can connect to any tool that you're using that offers an API endpoint, which is most of them wanted to pull live data from your internal CRM done needed to create support tickets on Zendesk or get the current weather data. Both of those are easy if you spend the time to set it up properly. You can truly build a professional AI assistant with Chatbase today. So if that's something you're interested in, check out the link at the top of the video's description. Thanks again to Chatbase for sponsoring this video. And now let's move on to the next piece of AI news that you can use. All right, next up we have another application out of Google Labs and every time I see a new release from Google Labs I get a little excited because they really know how to cook up a good AI interface that you have probably never seen before. And it's no different in this case. This whisk tool that by the way is completely free geoblocked in some locations, but a VPN makes this accessible even to me from Europe. It's quite simple. You pick a subject, you pick a style, and with their imagen free model in the background, that is one of the most capable models out there, they'll mash these two together and then out of the mix of those two, you can create a subject that you can define. This is the demo screen where the object that you create is pretty fine. So we're going to create a plushie and I just have the screenshot of the X logo laying around. So I guess we're going to do a X style plushie. Let's see what we get here. I don't even know myself. What's up? Whoa. And look at that. That's actually pretty sweet. Okay, so this is what the full interface would look like. So if I really like this, I could favorite it and then I could go ahead and delete these subjects here and stick start from scratch. And having played with this for a little bit, let me tell you, this is a really fun way to do it because they also have this random generation mode here where it just gives you random subjects, scenes and styles and you could even prompt for them. So maybe subject wise, I would want a gaming setup, let's say. And then for the scene we'll do underground dungeon. And for the style, I'm just going to randomly generate something. In this case this Japanese style. I like it. So I'm just going to run this combo and see if I can get a gaming set setup that looks like an underground dungeon in a Japanese style. Oh, hell yeah. Oh, yes. Please look at this. As you can see, it wrote the entire prompt for you. No need to engineer the heck out of this. You just define the three different variables or randomly generate them. I'll just do one more test. So I'm just going to get some random suggestions here. And yeah, this actually works. Let's use the T Rex as a subject here for a final demo. All right. This is kind of super unique, isn't it? And a very intuitive way to generate stuff. And one thing that popped into my mind is that as this is completely free and easy to use now, this might be a stretch, but it could potentially be a fun application to try out with your loved ones over the holidays. I mean, seriously, you could have each family member just pick one of these variables and then you'll get some surprising and funny results and teach them about AI capabilities in the process. If you do that though, please report back in the comments on how that went. I'd be super curious to hear it. Okay, onto the next one, which is just a brief follow up on last week midjourney canvas announcement that's fully available to everybody. And we gave this thing a quick spin with the images of the AI artists on our team and hates. And as I talked about within the mood board you can create entire worlds and maintain a certain theme within here. So it becomes really easy to generate images in the exact same style. Like this epic penguin right here. Look at this. Maybe it's just me, but for me this print worthy stuff that I put up in my apartment and as mentioned last week, this mood board feature really is there to help you in creating entire worlds. Not just sets of images, characters, scenes, stories. I like this direction that my journey is taking here and I'm looking forward to seeing seeing more from them when it comes to actually telling stories. Alright, next up, we have two major releases in the land of AI video. First of all, it's VO2 that is not really accessible to everybody, but it has been announced that some people have access to it and that's what we'll start talking about. And then I'll also briefly touch on Pika 2 which brought a new feature to the table here. So VO2, this one is really easy to sum up. It's Google's brand new AI video generator and it's definitely an S tier one. Now where this exactly ranks will release in our monthly video generator tier list updates that will release for free. Well, if we somehow manage to get our hands on this, that is if anybody from Google is watching this, it would be appreciated. Igor@myadvantage.com but without access, what we can do right now is look at some of the prompts that they shared in here with the results and actually compare them to some of the competitors. So I figured why not just throw the same prompts into Sora and compare these results. Now arguably Sora might not be the very best video generators. A lot of people make the point that the Chinese Hanyu 1 video generator is actually the best ones. I think it really depends on the use case and we're in the middle of testing and comparing these on many different categories. But what I can tell you is that the interface in SORA is the best and it's from OpenAI, so it's a worthy competitor to make these comparisons with. So first of all, we'll look at this shot of the sun rising slowly behind a perfectly plated breakfast scene with gold maple syrup pouring in slow motion over a stack of fluffy pancakes. So this is the interesting thing about VEO here. It can create sequences, not just single shots. That's the thing we haven't seen seen yet if I'm not mistaken. Okay, so I quickly put in two More prompts that will generate here in Sora while we compare these results. Alright, so clearly on one of these we have these pancakes come out of nowhere. And I think in both cases the fluid maple syrup is just not even close to realistic, whereas the scene from VEO seems to be far superior. Plus there's obviously the fact that it generates a sequence and not just a shot. But yeah, the maple syrup here looks deceptively real, whereas the this is not even close. Now sure, it would probably be fair to generate like 60 different versions of this, but I think the fact that on the first two the syrup is that far off is a pretty good tell. So now let's look at the next shot, which is this one. Fluffy cockapoo, I think that's how it's pronounced. Perched atop a vibrant pink flamingo float in a sun drenched LA swimming pool. Okay, what about this? Cockapoo ensora? Oh, look at that. That's actually really good. Okay, his eyes twitching a bit, but that's fine. Ah, this one's just so cute. Look at that. Okay, the float is a bit weird. I just noticed the double flamingo there. So yeah, you're pulling ahead here, but I would say that this one actually looks even more cinematic. Again, this is just a quick first look. You bet that these were heavily cherry picked amongst maybe 100 generations. So you gotta cut SORA some slack here. But actually I really like these Sora generations here. You really gotta be nitpicky on the first one to find something that is very off with this one. Let's do one more that these models are usually good at animated scenes. So this is the VO2 version of it and this is the Sora version of it. And look at that. Completely different aesthetic. Of course that will vary from prompt to prompt. And as I said, all the video generators are good at this. I think SORA actually does a fantastic job here if you dub over this. And I think that most people would be able to tell that this is even AI. So yeah, once we get access to this, we'll do in depth comparison and tests across a variety of use cases with original prompts. But right now my verdict would be that Google did some serious catching up on all fronts. Their imagen free model is in the front the front line of what's possible with AI imaging. VO2 is among the leading pack of video generators from everything we can see here and their upgrades to Gemini Flash with by the way, the rumors of a brand new experimental Gemini Pro currently again topping the leaderboards of LMSwise Arena. Once they release that, they're amongst the top players in the market. Amongst imaging video small models, large models, they're the best with context windows and video upload. And they're cooking up interesting things left and right, like Google's Notebook LM their Learn models, or experimental interfaces like Whisk. It took a while, but at this point they're not just catching up, they're setting the pace in some of these categories and I'm looking forward to seeing more from them. But talking about video generators, as I talked about, there's really two releases. The second big one is Pika 2.0 that is now accessible. I'll keep this segment rather short because what we did is we ran our standard benchmarking image to video prompts on it right here on screen. We'll put up a comparison between some of the other models and and Pika 2.0 so you can make up your own mind on where this ranks. My overall verdict is that some of these are really good. They get funky just like every video generator, but there's segments of it that are not just usable, but actually really good. Looking over all of this, I gotta say it's gonna be really hard to rank some of these because the level is becoming this high. I suppose it really depends on these harder categories. Something like close up of a face is not that easy to pull off. People are really good on picking up on all the nuances there. And also I personally always like this car benchmark because there's just a certain realistic feel feel that is hard to pull off of a video generator. Here you can see Pika turning this into more drift than anything else. Overall, my feeling is this is very good. Maybe a tier, but probably not S tier. Something like VO2 and Hanyuan video or Minimax are just incredibly good while this is good. But it doesn't end there because Pika is always the fun player here. And they added this little feature called Ingredients where you can give it a picture and then add a second ingredient to it. In this case, the team added a picture of me and said Igor flying away into the sky. And the clip you're about to see is what Pika Ingredients did here. So as you can see, this is super fun feature that is worth trying again. It's sort of just silly. But the last set of silly features they released actually went Giga viral across all social media accounts. Independent of the fact that they're AI they were just really interesting and cool and now they added even more. And there you go that's another week of AI news you can use. I want to highlight one more clip here, and that is from the AMA with Sam Altman. He was asked a question about when will the context length of various models expand? And he answered this, even getting to the, like, 10 million tokens of very fast and accurate context, which I expect measured in, like, months, something like that. So Sam essentially confirmed that we're going to have models with a context limit of up to 10 million tokens in months. So I think that means sometimes in 2025. And that will literally superpower every single thing we talked here about. You're going to be able to search the web, you're going to be able to interact with it via voice. It will see what you're doing on your computer, have your personal context, it might even store your interactions from past few days. And it will even have the ability to think about things and then perform actions just like the new 01 through the API. Slowly but surely, this is all coming together to the Sci Fi vision that a lot of people have when they get into AI, and then they realize that we're still at this chatbot level. But slowly but surely, all these announcements, they're leading us into a future that is a bit more hands off, a bit more autonomous. And here on the channel, it's a pleasure to bring you along for this ride. Even though the holidays are coming up, we have some uploads lined up for next week. And with that being said, I hope you have a wonderful day.