Massive AI News : MAJOR AI Breakthrough, New AI Agent, Gemini 2, Claude 4, Deepseek BANNED!

So one of the first things that you know, happened last week in AI was of course, deep research. I do know that I am a little bit late on this, but I still think that this is a really, really important thing because deep research is something that fundamentally, I don't want to say it changes the economy, but most people are missing what deep research kind of shows us. And it basically shows us how agents are going to be integrated into society. Now deep research, if you aren't familiar, is essentially a agent framework that allows you to do a ton of research via autonomous agents that essentially scour the Internet and do deep research. So they analyze, synthesize hundreds of online sources and create a comprehensive report in tens of minutes, which is basically something that's going to save you hours and hours on end. Now this is something that I personally think is absolutely incredible. And the reason I say that most people don't realize what this is, is that when we thought about how AI would be integrated into certain works in certain places, we would often think that, you know, AI is just going to control the computer and take certain roles. But this is an agentic framework on the back end where you have hundreds of agents scouring the Internet, synthesizing information and delivering your report. And this is something that most people haven't realized has already changed the economy. And this is something that Sam Altman actually talks about. And the fact that this is something that is going to save you hours and hours of time and as time goes on, there will be more and more things released like this that are going to slowly change how we do knowledge work. And this is where Sam Altman actually talks about this deep research thing. And this is why I say most people, you know, they might not use this on a day to day basis, but it's important to understand that the economy is already changing. Even if you didn't understand what deep research is, what it's able to do. But, but it's important to understand how crazy this actually is in terms of economic value. Just because it's top of mind. This deep research thing that we just launched, my, this is not a scientifically rigorous thing. My vibes based estimate is that does about 5% of all tasks in the economy today. One feature, 5% of all the tasks in the economy today, I think you launched a week ago. Yeah, and that's why I wanted to actually talk about deep research because I do believe that the announcement is a lot more significant than people do think. And if he's saying that, look, this one feature can do 5% of all tasks in the economy today. What does it look like in two to three years when we do have other different tools and abilities from AI agents that allow us to do so much more? So Deep Research, I will have a video coming on that shortly. I know it is a little bit late. Like I said before, I was off very, very ill and there is still so much to talk about. So of course as well with Deep Research, many people are stunned by the capabilities of this. You can see someone sweared, wow, things are moving very fast. This is coming from Tyler Cohen, the American economist who hasn't exactly been bullish recently on AI and he says, I've had it write a number of 10 page papers for me. Each of them are outstanding and I think the quality as comparable to having a good PhD level research assistant and sending that person away with a task for a week or two or maybe more. Except Deep Research does the work within five or six minutes and it does not seem to make errors due to the quality of the embedded O3 model and it seems that it can just cover about any simple topic. So with that being said, I think this is going to be something that's really, really impressive and it's something that you know, if you haven't tried it out already, definitely do try it out because it is something that can save you hours of time and allows you to deep, deep just do a deep, deep dive on something that exists already. Now I'm not going to lie, Google have had their Deep Research tool for quite some time and, and I think it's pretty crazy that like you can literally see right here, Gemini had 1.5 pro with deep research. The fact that OpenAI just literally straight up took that name from them, I don't know, I think that is absolutely insane in terms of how the AI world is. But nonetheless it is still very competitive in the AI space. And that just goes to show you. Speaking of AI, have you actually wondered how large language models like the ones we talk about consistently actually work under the hood? Well, this is where today's sponsor Brilliant comes in. Brilliant is where you learn by doing with thousands of interactive lessons in math, data analysis, programming and AI. And what I personally love about Brilliant is that they break down complex concepts into intuitive hands on problems that you can build real understanding from the ground up. Now their courses on AI and programming are particularly fascinating. You'll start building programs on day one with their built in Python editor and they have an excellent course that lets you peek under the hood of LLMs and to understand the concepts powering today's AI technology. Instead of just watching passive lectures, you're actively solving problems and building intuition, which has been proven to be six times more effective for learning. What I especially appreciate is how Brilliant helps you develop a consistent learning habit. In just a few minutes a day, you can actually level up your programming and AI skills, whether you're commuting, taking a break, or waiting for one of your AI models to train. Now, if you want to try it out, the first 30 days are completely free when you sign up with my link brilliant.org aigrid or alternatively, you can either scan this QR code on screen right now or just click the link right there. Now, if we're going to be talking about things like the economy, you may have missed Sam Altman's blog post about 2035 and there were three things that I want to quickly go over. In this video he discussed that the intelligence of an AI model roughly equals the log of resources, train and run it, and basically says that, look, these scaling laws aren't going down and it appears that you can spend crazy amounts and get continuous predictable gains. And the scaling laws that predict these are accurate over many orders of magnitude. Basically what he's stating here is that there's no obvious roadblock to this kind of intelligence at the current rate, and it seems that there's going to be continued continued investment into this space, considering that these scaling laws are very, very accurate. Also, this is something that is really important considering DeepSeq and other things, is that the cost to use level of AI falls about 10 times every 12 months and lower prices lead to much more use. So he says basically GPT4 to GPT4O, the price per token has dropped about 150 times, and Moore's Law changed the world two times every 18 months. But this is something that's pretty crazy. So for those of you who are wanting to build with AI, this is something that is really, really effective. And I think that this is something that makes a lot of sense, especially for those of you who are trying to get into AI. I definitely think now the best time, because AI is going to get cheaper and cheaper and whatever it is that you do want to build, it's going to become less and less expensive. And of course the socioeconomic value of linearly increasing intelligence is super exponential in nature. So of course, like I said before, he sees no reason for exponentially increasing investment to stop in the near future, which means a lot of the companies that you think are overvalued Maybe they might just not be. Because I think most people are forgetting the fact that whilst, yes, AI is going to bring us a new level of intelligence, that level of intelligence is then going to bring us more breakthroughs. And then from those breakthroughs we're going to get more things and more things. So AI is going to be something that compounds on itself, which is why a lot of people are predicting that this is going to be super exponential and it's going to be something that you definitely shouldn't underestimate. And there was also the largest medical AI randomized control trial. Basically they enrolled 100,000 women undergoing mammography screening and it was published today. And the use of AI led to 29% higher detection of cancer. No increase in false positives and reduced workload compared to radiologists without AI. And this person, this MD said, I've said it before and I'll say it again, not using AI, especially in medical diagnostics or in screenings like the one below, has now become completely unethical and unacceptable. MDs who reject this because it isn't standard care are liable to failing to properly care for patients. I seriously cannot wait for the day that AI is seriously adopted in the medical space. Because I do believe, I mean, I don't want to say I do believe, but from every single paper that I've seen, from every single research study that I've seen, AI has consistently outperformed humans. And I've said this, and I'll say it again, it seems like in certain cases, I'm not saying that, you know, AI is the end all and be all. I'm not saying that, AI doesn't have its mistakes, but in many cases it does show us that AI with these people, or at least AI by itself, tends to outperform humans. And you have to ask yourself, at what point does it become foolish to not use at all if it can actually be more accurate? You know, you have to ask yourselves, at what point are you going to adopt it? I personally do believe it is going to be a situation where certain countries, certain innovative countries decide to adopt it and then their healthcare starts to see an uptick and improvement. They're starting to see, you know, people living for longer as a byproduct of everything getting caught earlier and as a byproduct of the AI having all of your knowledge, all of that context, all of that screening, and it's able to just pin down everything. I mean, it's going to have your parents biological data, your grandparents biological data. It's going to Know what you eat, like how you train, how you sleep, how your job, how your mental health. All of that data is just going to make everything so more accurate, so much more accurate. And I think it's probably going to take maybe one hospital that's going to be able to do it, or maybe one small country. And then from there we're probably going to see widespread adoption after those trials are present. But I do think that AI is definitely going to be there in the future. It's just a matter of time due to the, of course, legalities of this. Because the medical field, they don't just approve anything. There have been things that take a really long time to get approved because of course, you are dealing with human lives. And if there is an issue where the AI manages to make a mistake and someone dies from it, that is going to be a very big problem. Now, of course, there was also this. This isn't crazy news, but I think this is crazy news for those of you who pay attention to robotics. So the company PI Zero, they actually open source their framework for robotics. And I think this is big news because this actually means that if they open source this framework, this actually means we're going to get a ton of other individuals who are using robots, be able to do a lot more autonomous tasks. And the reason I say that is because if you aren't familiar with the company PI zero, they made this model PI zero that you used to. You can basically generalize across different robots and it's really, really effective. Like it's the most effective autonomous robot policy that I've seen so far, like the entire model that you can use. And I'm really, really hoping that now just how AI open source is slowly creeping up to the level of closed source AI. I'm hoping that robots closed source, the open source can catch up to where the closed source is. So this is going to be something that's super interesting for me because I'm actually watching this company really, really closely because I don't know about you guys, AI is crazy, but I just think that robots in the real physical world, that is going to be absolutely mind blowing. And of course, we did also have Detroit based Borg sharing a clip of collaborative work between its AI pallet Jack and the O1 humanoid. The humanoid actually uses a humanoid modular design that actually enables it to switch between a bipedal and a wheeled base. And it's going to be launching in a quarter two of this year. I personally think this is, you know, how robots are probably meant to be Because a lot of the times whilst yes, humanoid robots are good, I do think that they probably should have a wheeled base in some format because certain specific areas are just going to be so much more effective for you to just use wheels. And I think so much engineering goes into legs and I'm not sure it's entirely worth it in certain scenarios. So for me personally, it's great to see a company taking a different approach and not focusing on the gate of robots and how realistic they walk. Of course, you know, it is going to be something that certain companies focus on, but it is really interesting to see humanoid robots that are taking a different approach. Now this is probably the thing that I'm most excited for in the coming months. And we got an update from Brett Adcock and he actually said, and this is crazy, right? Like this is crazy even for the AI space. He said that today I made the crazy decision. Well, he didn't say crazy. I'm putting words into his mouth. He said, I made the decision to leave our collaboration agreement with OpenAI. Figure made a major breakthrough on fully end to end robo AI built entirely in house. We're excited to show you in the next 30 days something no one has ever seen on a humanoid. So it's quite crazy because apparently they made a major breakthrough on a fully end to end robot. So this is going to be something that is 100% autonomous. And I'm wondering what kind of breakthrough it is that they did make. The last time that Figure actually spoke about something that was really cool coming, the video actually did kind of break the Internet. I remember uploading it and the video shot straight to a million views. So it is quite likely that what they are able to do is a significant breakthrough. Now of course, some may argue that, you know, you've of course got to hype up your own company to drum up funding. I think this company raised another round, valuing it around $39 billion. But. But this actually might be why. In fact, let me actually just bring up that article now. But yeah, you can see right here, it says figure AI is in talks to raise $1.5 billion at 15 times its last valuation. Now, like I said before, I don't know if this is because of the recent breakthrough that they spoke about, but it might be. It actually might be because if they made a breakthrough that is so significant that they think, look, we are going to have the lion's share of the humanoid robots market, they can of course basically show that demo to these investors and if they are in talks to raise that amount of money at that valuation. And these investors, they're like, okay, wow, we've seen the demo, we want to invest at that valuation. Then of course, that's how much the company is valued at. So I am wondering if the fact that they're now in talks to raise $1.5 billion at 15 times its last valuation, which would put it at $39.5 billion in valuation, that is absolutely crazy. So I am waiting for this. They did say in the next 30 days, so it will be by the, probably end of February or maybe by the start of March. I'm going to be super intrigued to see what happens there. And yeah, it's going to be pretty, pretty crazy. So I'm not sure what kind of breakthrough it is. I'm not sure what they're working on or what they've been able to do, but I'm guessing that it must be a big one. And I think if robotics does manage to catch up to the pace of AI, and this is going to be something that is absolutely incredible because we've got these two things that are very synergetic in nature that are both moving very quickly now. There was also the Replit agent that was launched on Android and iOS, which essentially allows users, in fact, let me actually play the video.

You always wondered why that app didn't exist. Now you can make it. Meet Replit Agent. Want a custom workout plan? Make an app for that. Have a great idea at 2am make an app for that. Need to run an analysis or learn with your kids? Make an app for that too. This isn't just making apps. This is your imagination unleashed. Describe what you want and watch it come to life. No coding experience right from your phone. Made an awesome app. Share it with a friend. Have a business idea. Start in minutes. Everything you need is in your hands. Now is the perfect time. Create something amazing. What will you make today? Try it free@replit.com essentially the Replit agent.

Launched on Android and iOS, which basically allows people to make apps with just text prompts and it handles all the heavy listing, the coding, the databases, the integrations, the hosting and more. And basically you can produce usable and shareable apps within seconds. And you can actually try this for free. So this is basically one of the biggest game changing things and you might think, okay, they just released this cute caller app, but I don't think so. We have to look about where this is going to be in the next five years. If right now we are currently at a stage where you can text to app. Okay. And although the apps are super basic, what happens in the next five years when these apps are fully fledged and they are really easy to use and you can do complex, different things with them? How different is that going to make the SaaS space? Because I think SaaS is going to change completely when people can create their own customized versions of their software that they don't need to buy into all the SaaS companies for. So it's pretty crazy. It's a real big change to the ecosystem. So if you are developing a SaaS, this is definitely a trend to continue. I don't remember who tweeted it, but someone tweeted that the biggest threat to your SaaS in the future is going to be the fact that users can potentially create that sass for a fraction of the cost within seconds. So I personally think that with whatever SaaS you do create, I think there's probably going to have to be some kind of data, some kind of integration to something physical or something proprietary, like, for example, you know, Uber Eats. That's something that, you know, I can't just go ahead and make a food delivery app. Of course, I need to be able to be logged into their shops and their APIs and stuff like that. So I think those kind of apps will still stand. But certain consumer apps, like very simple ones, I think those ones, I don't want to say they're going to be going away, but I just think that the marketplace is going to be changing when users can create their own custom dynamic apps, especially for certain things. So for certain things where, you know, you can upload your own data, maybe like accounting, maybe things like finance, or maybe like budget tracking, those apps, I'm not sure how those ones stay relevant. I'm guessing through, of course, certain APIs, but it's going to be super interesting to see how that changes. There was also Mistral, which has Lichat. And Lichat is basically a chatbot arena that is not chatbot arena. It's actually a chatbot that is super fast. The very best selling point right here that you're going to get from this model is, is that it is the current fastest model in terms of sheer output responses. So if you ever just wanted a model that responds super, super quickly, that has Internet access, that can create you images and do all of that stuff, this is going to be the model for you to use. And I'm pretty sure that this is free for the most part. So for those of you that are trying to save money, this is going to be something that you want to use. And if you're someone who wants essentially a really, really fast model that's really, really effective. Lichat is certainly the way to go. So it offers web search, doc processing, code interpreter, image generation, and apparently this is 10 times the speed of ChatGPT. So I think every company right now is trying to operate on different factors and I think Lichat what they're doing, since they're using, I think Cerebras inference chips or either Grok inference chips, they're basically competing on speed. And of course, if there's one thing that we know people love, it's it is of course speed. If you could get your responses absolutely instantly. I've used GROK before. It is something that's absolutely insane. And I personally don't know why these companies aren't just buying all of their chips because that kind of speed makes your work so much faster. And if I had like a 03 level chatbot that was that fast, I don't even know how quick I would be working. It's absolutely insane. Now in Google News, in terms of other companies managing to produce chatbots, we got the Gemini 2.0 flash. It says we're making the updated Gemini 2.0 flash generally available in the Gemini API, in Google AI Studio and in Vertex AI. So Google have released 2.0 flash. They've also released the very best version of Gemini 2.0 Pro. This is their best model yet for coding performance and complex prompts. This is also available in the Google AI Studio and Vertex AI. And they've also released Google Gemini 2.0 flashlight, which is the basically cost efficient model that is a small model that you can pretty much use for anything. Now I have to be honest with you guys, Google are in a crazy spot because they also released Flash thinking experimental, which is something that is a thinking model. And this one actually connects to other apps like search Maps and YouTube. So overall, Google have essentially released four new models that are pretty much, I don't want to say state of the art because I'm not sure where the benchmarks are at the moment, but these models are super, super good. And I'm hearing things about Gemini 2.0 flash flashnking experimental and the fact that these models are super quick and they're able to digest so much information that it is absolutely incredible. So with that on Google's part, I want to show you guys right here, if you put the drop down menu, you can see that you've got 2.0 flash, you've got Flash syncing experimental and then you've got thinking experimental with apps. Now if you're wondering which one of these is going to be the most useful, I personally think 2.0 flash thinking with with apps is probably going to be the most most useful because you have access to YouTube, you have access to maps and you have access to search. So for me that is what you're definitely going to want to use. But if you're look, you know, thinking about cost effective developer solutions, I would definitely be using the 2.0 flash or 2.0 flash syncing experimental because you also get the speed and the effectiveness of the reasoning model. And I know there was a lot from Google, but Google are not slowing down at all. And I think probably by next year Google might even take the lead completely with AI because they've shown us their hand in terms of what they're capable of doing. And I think this year they might take the lead with just how much they're able to do. Now another crazy thing about Google is that starting today you can actually ask Gemini to consider your chats to craft its responses. You can easily pick up where you left off or have it summarize a previous topic. You can view, edit or delete any chats you've had with Gemini and see when it is used. You can try it in Gemini Advanced. And basically what this is, they should have just said that they have infinite memory. So if you've had a past conversation with Gemini, you can use those chats as context in order to build on your newer chat. So this is going to be something that, you know, helps with the usability of the model. I know personally that having to give Gemini all the responses is something that just was a big thing for me. One of the reasons that Gemini just didn't work that well was because it didn't remember certain things. And although Gemini is a good model, despite its terrible reputation, from what I remember, Google were the first to publish the research on this stuff that is essentially about how they're able to have unlimited, and I mean truly, truly unlimited context length window. So it's going to be super interesting to see if this is actually real, if this is actually, you know, possible. I'm definitely going to be testing this over the week. I do think it would be useful to have something that has access to my YouTube, to my Gmail, to my search, to everything that's able to have full context. It's just going to be so much more useful. And this is something that, you know, apparently was going to be solved in 2025 anyways. Now, speaking of newer models, many people have speculated that Grok 3 isn't going to be that good. One of the former employees said that GROK might actually be behind 0103 mini and ChatGPT 01 Pro. So for me, it's going to be super interesting to see where this lies because there was a lot of debate around this model and this was something that I was quite surprised by, but then again, not too surprised by, considering the fact that X AI have always sort of been behind the eight ball, considering that they started their company quite a little bit of time ago. So for me personally, I'm thinking that this company maybe they're not going to catch up just yet. I'm not sure what vector they're going to be able to attack on considering the fact that that they are currently behind a bit and the AI space moves so quickly. So with that, I do wonder where that's going on. This an employee that actually stopped working at the company because of this statement. They were basically like, look, I can't believe as effective, yada yada yada. But I made a whole video on that. And the long story short is that Grok 3, there are some serious claims being thrown around. Elon Musk said it's the smartest chatbot in the world. This guy said it's not that good. He left the company. It's going to be super interesting to see what exactly happens now in terms of other companies producing reasoning models, Anthropic is actually doing a new way to reason their models. So you can see right here that it says after OpenAI released its reasoning models last fall, Google, Alibaba and others are following with their owns. And one competitor, Anthropic, has been noticeably absent from the race. It's crazy because Anthropic, even though they haven't released any reasoning model yet, they've still been rather effective at making models that with the base model and 3.6 Clawsonet was one of those ones that just blew the industry away because it's so effective and it's not a thinking model. And people still use this model over many others. So essentially what we have here, it says Anthropic is taking a slightly different approach to reasoning. It has developed a hybrid AI model that includes reasoning capabilities, which basically means that the model uses more computational resources to calculate answers to hard questions. But the model can also handle simpler tasks quickly and without the extra work by acting like a traditional large language model and the company plans to release it in the coming weeks according to a person who's used it. So this is a hybrid model that basically calculates, you know, the resources. It basically uses, you know, more computational resources for hard questions and it just, you know, does simpler questions quickly. So I hope this actually does work because you know, they're talking about like GPT4, you know, someone's spoken about GPT5 being very, very similar to this architecture. But I hope that this actually does work because I think it's going to be really bad if the AI thinks that your question isn't a difficult one and it answers it very quickly or if it thinks like a simple question is a super hard one and then overthinks about the problem, I could see that becoming an issue. So it will be interesting to see how they would solve that because I guess you could say the difficulty of each question is quite subjective. But I mean it's intelligence. These things are probably way intelligent than I am, so they probably have some kind of brain that figures out exactly what you want. But it's going to be super interesting to see how it handles that. And honestly because of Anthropic and their track record, I'm super excited to see how smart those models are going to be. You can see right here they said by setting the slide scale to zero, developers can use the Anthropic model as a run of the mill non reasoning. And basically they also as well give you the ability to control how much resources it spends on a query. And I think this is is of course because you know, developers, you know, it's probably going to cost a lot when it's, you know, released. So see right here it says developers can adjust the amount with a sliding scale measured by the number of tokens or word the model will develop or will process or generate as it tries to figure out an answer or complete a task. Talks about how OpenAI has a similar capability. And you know, of course developers have limited three settings, low, medium and high. And of course it can be hard to predict how many tokens the model will actually process at these levels. So of course it's going to be hard to query, you know, hard to figure out actually how expensive the thing will be. Now of course elephant in the room, Deep Seek. But Deep Seek's been under some flack recently. Deepseek AI model is the best workout of China, but the hype is exaggerated. Google DeepMind CEO says this is Demis ASAP. So pretty sure you've all heard about Deepseek how it ruffled the feathers of the AI industry. But many are stating that, look, this is pretty much exaggerated.

It's a very impressive model, very impressive piece of work. I think the team is probably the best, best team that I've seen come out of China. So that said, I think a lot of the claims are exaggerated and a little bit misleading. So first of all, when you report how much it costs to do a training run, they seem to report just their final training run, which is actually a fraction of the cost it normally takes to explore and train and, and do all the tests beforehand before you do your final run.

And so after listening to what Demos Hasabis says, you can clearly understand why many are stating that the Deepseek claims were quite exaggerated in nature. Now, Deepseek isn't just done there. Apparently Deepseek might be getting banned. The state of New York has actually banned Deepseek already on government devices. Government Kathy Hochul said she issued the directive on Monday, citing serious concerns about Deep Seq's apparent censorship and potential for foreign government surveillance. The AI app created by the small research lab owned by the Chinese hedge fund, of course has had praise and suspicion since it abruptly passed some of the most well known models last month. She says that public safety is my top priority and we're working aggressively to protect New Yorkers from foreign and domestic threats. New York will continue fighting to combat cyber threats, ensuring the privacy of our data and safeguard against state sponsored censorship. So it's pretty crazy that, you know, the government is looking to ban this. It's quite like the same with TikTok. Now, of course you can see Deepseek ban. A new proposed law could see people in the US face significant fines or even jail time for using the Deep Seek app. The bill filed by the Republican Senator Josh Hawley aims to prohibit United States persons from advancing AI capabilities within the People's Republic of China and for other persons. The proposed law would prevent the importation of technology or intellectual property developed in China, with anyone violating the restrictions facing up to 20 years in prison. Individuals could also be fined up to $1 million to wealth of businesses. It could reach up to $100 million. You know how insane that is? I mean, do you guys think that people are actually worried about Deep Seek being some kind of, you know, Chinese intelligence thing? Or do you think they're actually just worried about the fact that the Chinese AI might just surpass US's? I mean, it's quite the conundrum. And Elon Musk has kind of weighed in on this situation, talking about some very interesting things.

There are a lot of very smart, very motivated engineers in China. So you should expect that China will come up with many great things. And they have come up with many great things. Again, look at that. And if one looks at the long history of China, also an ancient nation that has invented and developed many things, and for most of actually human history, China has been the most powerful nation on earth. So, you know, I think particularly from, say, with the possession of people in China, they. They simply view China's ascendancy as resuming their normal sort of place in history, which is the most powerful country on earth. And so you can expect that they will do many great things, Deep Seat being one of them. But that is simply a result of the immense amount of talent in China. That is very impressive. But is it some sort of complete revolution in AI? No, it is not. Xai and others will soon be releasing models that are better than deepsea.

So what do you guys think? Do you guys think this is the end? Or do you guys think that this is something that is just beginning? Do you think China's about to overtake America? That being said, hopefully you guys enjoyed the video and I'll see you guys in the next one.