This Free AI Is Smarter Than Most Humans

The first episode of AI news you can use in 2025, we had a little Christmas break. Not many releases. Wanted to give the team a little break too. But now we're back at it, covering everything that came out over the course of the past two weeks. And we'll be starting with a look at the various reasoning models that came out. I mean, heck, now we have Chinese competition 201 and also Google released a reasoning model, but then later in the video we'll be exploring a way to create custom podcasts with your very own voice in a super intuitive way. This should be a fun episode of AI News. You can use the show that looks at all the AI releases of the past week and we filter it for just the stuff that you can use right now. Let's get into it. All right, so to kick things off, I want to talk about these various reasoning models and some new open source competitors that came out over the course of the past two weeks. And there's a few releases in this category, but I'm going to compress this into one segment. Namely, we're going to talk about DeepSeek v3, the new open source king, in terms of LLMs, leading all the leaderboards and even beating models, models like sonnet and GPT4O while being fully open source. Then we have Google Gemini's deep research. This is Derek on 01. And similarly we have a new Alibaba model called QVQ. This is the Chinese counterpart to 01, with the slight twist that they focus on reasoning over visual inputs, not just text inputs. And the reason I want to talk about these together is because they obviously are all trying to catch up with what OpenAI released with 01. And now the O3 preview that we got in December. And as you might know, all ChatGPT plus subscribers now have access to 01. And on their new Pro tier you get 01 Pro. But the problem with demoing these things, testing these things, is that, yeah, they're really good at reasoning over specialized problems. And most people don't have complex coding questions or specific scientific inquiries as a part of their routine. And therefore these models are not as useful to them as they might initially seem. And now that we're getting a plethora of models, I had an idea of how we could tackle this as a community and how we could solve this question of what? What could I use this for? What are other people using this for? And that idea is running another public challenge over the course of January. So the stream for that is already scheduled. I'M going to be hosting that this coming Monday, but you can already check out the challenge and I'll be doing these once a month. In other words, I will be asking people to submit various use cases for These models like 01, QVQ72B or Gemini Deep Research and then the most useful ones will get prices and the entire community gets to see what people are actually using this stuff for. I think overall that's just a great idea. And look, I wish I could go in here and give you the specific examples of how to use this or if QVQ is better than O1 or how Gemini Deep Research relates to the O1 product or Zero1Pro. But at the end of the day a lot of the outputs are just very similar in a lot of the tasks that I'm using this day to day too. And it seems kind of meaningless to just throw random logic problems at it here in the video. That's good for benchmarking, but what I care about here on this channel is how to use this technology in your life. With that being said, there are a few details about these releases that I'll quickly summarize to you. Namely that this Chinese model, QVQ72B Preview, is a reasoning model that focuses on the vision aspect. So this is something OpenAI hasn't done yet. As you might know, this is something that's actually included in the O1 product as you can give it images and it can interact with them. But the difference is that this model is actually freely usable. You just have two attributes that has been built with qen, but you get the full model including the weights and they give you the license to use this for whatever projects you want to build with it. And then we have Geminized Deep Research that we tested here for you. Again, the conclusion is that the deep differences between these models, they're just very use case specific and we'll have to crowdsource this within the community. As I said with the challenge to really learn what people are using this for, I can tell you on basic coding tasks that I've been doing with 01 and 01 Pro and this is purely based on my intuition and usage of these tools, all of these competitors don't even come close to 01, especially 01 Pro. When I have coding problems, I throw it into 01 Pro and it just figures it out. It works like a charm. When I tested some of those on Google's Deep Research or ran it through Sonnet 3.5 or Deepseek v3, the consistency of the results just isn't this high. Is the 01 Pro worth the money? No, probably not for most people, absolutely not. But is it a better model than anything that we've seen from the competition here? I would go out on a limb and say yeah, it does feel like that, but even if we accept that as a fact, honestly, 01 completely suffices for pretty much any task you will be throwing at this. 01 Pro is sort of a luxury product. So in summary, 01 is fantastic and I would recommend you start using it and testing it with some prompts where you think that you might not even need it. 01 Pro is simply put, a luxury good that is probably unnecessary. And Gemini Deep Research and QVQ are just slightly weaker versions of this. Again, I'm not speaking from benchmarks, I'm just speaking based on what I saw, how I use these tools and what I've seen expressed within our community. And that leaves us with the last model here that is brand new and that is Deepseek V3. Now we have a fully open source model that performs better than GPT4O, better than Sonnet 3.5 on some benchmarks with the big difference that this is not just fully open source, but they also provide you with this website chat.deepseek.com where you can find freely use this so you get 4.0 like quality for free. Plus you could use this model for anything you want. And you also have this option here to turn on the reasoning capabilities, turning this into sort of a watered down version of 01. Now on this one I can very confidently say that it is actually way worse than 01. This doesn't even come close. Gemini Deep Research sort of does I would say. But yeah, those are all paid products. This is completely free. You can just go to this website and use the reasoning model, use the base language model that just works at this point. All these models just do great. But you probably know that if you're consuming this video right now and there you go, that's all the news on the latest model releases that happened over Christmas in the last week of 2024. Again, check out the upcoming stream and the challenge we'll be running to see what this community I've used these videos is actually using these models for because they are very useful but always for very specific situations and I can't wait to hear from maybe even you on how you have used these to enhance your life lately. I've seen a lot of interest in running AI locally, whether it's for privacy, security or just knowing that you always have the AI available, even if something is wrong with OpenAI servers like during the recent Sora launch or the outages in December. But to take advantage of the latest local LLMs, you do need a proper setup. And if you're looking to upgrade your rig or are in the market for a new laptop, let me introduce you to the Ryzen Pro from amd, the sponsor of today's video. Now the great thing here is that the AMD Ryzen Pro checks off all the boxes when it comes to AI readiness. It is optimized to run AI locally. It has enterprise grade security, top of the line graphics and provides all day battery life. Predictably, my favorite thing about this is the built in AI readiness. They're truly designed from the ground up with AI as the end goal. Here they've got a combination of a dedicated npu, GPU and CPU that lets you run AI powered apps directly on your PC without needing to offload stuff to other devices or the cloud. But even apart from the AI aspects, the Ryzen Pro processor is fantastic at anything else you might want from a processor like security or battery life. There's a built in AMD Secure processor that verifies code before it runs it, helping ensure that your data stays safe. And that just comes along with many other security features like the AMD Memory Guard. Ryzen Pro processors can give you up to 29 hours of battery life in a laptop. That is a lot of time you can work or play all day and then some. Plus they also make life easier for IT teams that are all on the same processor with features designed to minimize disruption and reduce support costs. So whether you want to use AI applications for yourself or as a team, the Ryzen Pro helps you keep things running smoothly. And here's the best part. Check this out. They're actually letting you try out this processor inside of a laptop for free. Yeah, you heard that right. They will lend you a laptop for free. So you can see if this is something for you. No strings attached here. So if you're ready to rethink your workplace with the AMD Ryzen Pro, hit up the free loaner offer from the link in the video description below. A big thank you to AMD for sponsoring today's video. And now let's move on to the next piece of AI news that you can use. Okay, this is just a super quick feature, but midjourney shared some of these SREF codes. If you're not familiar, you can think of these as a very Specific style that is captured inside of this preset that you can access by using one of these SREF codes. In the end, these are literally the six most used codes of this year. So these are the styles that people have been using the most inside of midjourney. If you like any single one of these six, you can simply include this at the end of your mid journey prompts that is behind a paid wall. But I thought this was a wonderful thing for them to share with the public because these SREF codes are truly powerful. We have a set that we use for the AI Advantage branding and if you ever want to create graphics that look consistent, this is a great way to do it. And here you can simply use one of these codes. And now we're at least aware of the fact that these are the popular codes that people have been using all year. Okay, onto the next story, which is one that I'm quite passionate about myself because I really like Notebook Alignment. They're extending it with a new feature. We're not going to be focusing on that because I don't think it's that big of a deal, but it's basically this interactive mode they're adding so you could participate in the podcast that NotebookLM will generate for you. If you're not familiar, it's app where you add a bunch of sources and then it summarizes them in a quite advanced manner and turns that into a human like sound. In podcast now you can join a conversation, you turn on the mic and you talk to the two podcast hosts. So you can think of it as continuing a ChatGPT conversation. You're just doing it with a voice interface so quickly. Wanted to point that out, but it perfectly transitions into the main story that I wanted to highlight here, which is 11 labs shipping like crazy. And I don't know at this point I would like to make a little disclaimer that hey, elevenlabs never sponsored this channel whatsoever, but they keep shipping so many features and the quality of these products is so high that I just love featuring them. And that's what we do in this show. Just to reiterate something that I've often said and it's very important to me, every single thing that we've done at the Advantage that has ever been sponsored, we clearly communicated that's a thing that is extremely important to me and you can hold me accountable to that. With that being said, here's the new projects feature, which is essentially NotebookLM but with way more customizations. So you can create these podcasts, but you get control over the dialogue, you get control of the voices. You can use your own custom voices. Let's have a brief look at this together here. Basically, if I log into my 11Labs account and I go into the application, this thing has been extending. I remember a year ago, there was basically just a voice creation. Now you can do things like this. And this is the feature I'm referring to. They call it Gen fm. It comes under this orange icon that says create a podcast. And I'm just going to take a random Wikipedia article here, the article on New Year's, and I'm going to put this in as the input URL. And in my headphones, I should hear these voices. A single rose can be my garden. And then, Jessica, if you spend your whole life waiting for the storm, you'll never. So how about we change this up? We keep crisp. But, Jessica, we're going to switch out for my very own custom voice, which sounds like this. As we are liberated from our own fear, our presence automatically liberates others. Okay, and one little pro tip when using some of these custom voices, we played around with all of these is you want to actually go to the Turbo model. And if you don't need multilingual, just go with Turbo V2. That just has the highest quality and makes way more of a difference than some of these quality differences that are hidden behind the more expensive plans. Standard will do it for now, and I'll just simply say generate, and it will look at this article and come up with a script and do the voices by itself. Now, here's the kicker. Once this is done, I will be able to actually edit the text, which is something that you can't do in NotebookLM. They have this customized feature where you can prompt and change the whole thing, but you don't get a text editor interface where you really get to dial in what the podcast hosts say. And of course, you don't get the ability to use a custom voice like my own voice that I trained into this. And there you go. Took about three minutes. Let's have a listen. Today we're exploring how cultures around the world mark the passage of time in ways you might never have imagined. Well, that's certainly an intriguing way to kick off our discussion. I'm curious to hear more about these diverse traditions. In this case, you might want to play with the stability here. This is something that Notebook alum does really well. It just legitimately sounds human because you get the intonation, they do the little pauses, you can get a little bit of that with these sliders. Well, that's certainly an intriguing way to kick off our discussion. I'm curious to hear more about these diverse traditions. As you can see, the pacing there sounds more natural than before. And that's basically it. You can go ahead and you can edit this transcript and then once you're happy with this, you could go ahead and convert this into a project. In this case, I'm just going to shorten it so it happens faster. Say convert, Convert project. And then it's automatically going to render our audio file for us. So that was pretty fast. Let's have a quick listen here. And in some areas, Easter was considered the start of the new year. You can imagine how confusing that must have been. Ah, yeah, I can see how that would lead to some serious calendar chaos. So when did other parts of the world start celebrating? On January 1st. Pretty good. And that's my voice right there. So definitely something you can do inside of NotebookLM. And now you know how to do it too. Now I wanted to highlight this because this is an interesting use of multiple tools that we already have, but they just combine them in a unique way. And that's what a lot of useful AI tools are like. They're not reinventing the wheel, they're just combining different, different building blocks in an unexpected manner. And I think when we look back at 2024, NotebookLM was the app that surprised everybody by how sticky and useful it was. People just love that product and they keep returning to it and now you can have a custom version of it. Also, by the way, I uploaded a video recently where I showed you how to build an even more customized version of a meaning summarizer that is similar to this, sort of like a custom notebook lm, but like really custom, even more than this. And just different people are going to have different needs. But if you want to learn how to do something like this yourself self hosted without any technical knowledge required, you might want to check out that video. If you want a small amount of control, 11 labs it is. And if you don't need the control at all and you just want an audio file of two people talking about a topic, just use Notebook alone. All right, next up, very briefly, this just popped up on my radar and I wanted to introduce this as it shows a new category of tool that I think we're going to see a lot over the coming year. As you can see, a lot of these tools and techniques are starting to merge and that is really the underlying trend here. Because you think about 01, it's something that people discovered for prompting. It's chain of thought prompting, just think step by step. And then they just embedded that into the model on a deeper level by actually training it with those examples and that prompting approach in mind. Now, surely that's just a gross oversimplification, but the basic idea is, hey, you had a large language model, then you had a prompting method, and when you merge those, you get something better. Same thing with level labs that we just looked at. Hey, you have a custom voice, you have LLM that generates text. The LLM can also generate podcasts. What if you use the custom voices that to voice the podcasts, you just get something brand new. And because all of these AI tools are built on Transformers, they're really good at transforming one thing into another. And I think a big theme of this year is going to be multiple transformations being combined. And that's exactly what this GitHub repo right here called mm audio does. It's basically a video to audio synthesizer. And I know most people won't be using this. I just want to open your minds to the possibility of these transformations happening every direction. Everybody has seen text to text now, right? It's just basic chatgpt. Pretty much everybody's also aware of all the text to video progress and all the text to image or text to audio. And right here is an open source version of something that you might refer to as video to audio. Plus it also syncs it at the same time. So you give it a video file from Sora here, for example, and it comes up with what it should sound like. Okay, yes. I don't know about those AI videos, but I think you see the point. And again, I'm not bringing this up because you will be using this in your next project. I mean, you certainly can. It's out there. The link is in the description below. But I just want you to be aware of the fact that you can transform any form of media into any other form now. And people will be combining this and really what all these agents that we keep touching on again and again are just very specific ways to transform one piece of information into another. And then obviously there's a lot of orchestration that happens there, but just like here, it generates the audio and. And then it, so to say, orchestrates it in a way where it syncs with the video. Agents also just transform one form of input into an output and they just need to coordinate for it to be productive. But understanding that in 2025, every piece of media can be somehow turned and transformed into something else might just open your eyes to solving problems or to understanding some of these tools more in depth. And that just might give you an advantage of a person who maybe used ChatGPT for 30 seconds and called it a day. With that being said, I'm genuinely excited about going into today 2025 together with you, exploring this landscape, covering everything that matters. And as per usual, we'll be mixing in tutorials because this channel is all about the use cases and education around them, and I'm just grateful to have you along for the ride. With that being said, check out the challenge that I pointed out. I can't wait to see what this community is using 01 and the other reasoning models for. And with that, I hope you'll have a wonderful day.