AI Shocks Again: DeepMind V2A, AI BRAIN, OpenAI Nuclear AI, GPT-5 & More (June Monthly News)

The past month has brought some incredible advancements in the world of AI. A new AI can now generate videos that look even more real than reality itself. Claude 3.5 has emerged, giving OpenAI's GPT4O a serious challenge. Google DeepMind has made a breakthrough with AI Brain, unlocking the secrets of real human brains. Meanwhile, Elon Musk's XAI has introduced its Socrates model and we have updates on GPT5, along with notable moves from Apple and Meta. In the AI space. Apple's integration of ChatGPT has sparked quite a reaction from Elon Musk. A company claims to have created the body for AGI. And Google DeepMind has launched V2A. Additionally, there's a new free Sora AI rival called Cling. And Dot is quickly becoming everyone's new AI best friend for gamers. The first ever gaming AI assistant has been released. OpenAI is ambitiously planning to power superhuman AI with nuclear energy. But they're also facing significant challenges. Challenges. We'll cover all these exciting updates and more in this video. So let's dive right in. If you've been following all the latest AI news and events, then you know everyone's been going nuts waiting for OpenAI to drop their new Sora model. But while we were all patiently waiting, this Chinese company called Kuaishou just went and dropped an absolute game changer. Out of nowhere, they released this wild new AI model called Cling. And man, it is blowing everyone's minds with what it can do. So let's get into it. CLING is a video generation model developed by Kuaishou, a company you might know from their popular app Kuai. This new model is like Sora, but some are saying it might even be better in some areas. What's really cool is that Cling is open access, meaning more people can get their hands on it and see what it can do. Let me paint a picture for you. So, a prompt that says a Chinese man sits at a table and eats noodles with chopsticks. Kling takes this and generates a video that looks almost too realistic. And if you remember that infamous video of Will Smith eating noodles that looked like something straight out of a horror movie. Well, Kling's version puts that to shame. It's like night and day. Seriously. Hey Uncle Phil, come try this. Kling can generate videos up to 2 minutes long with just a single prompt in full 1080p quality at 30 frames per second, which is actually pretty impressive. It accurately simulates real world physical properties, which means the videos it creates don't just look good, they behave like real life videos too. The magic behind Cling lies in its diffusion transformer architecture. This technology helps it translate rich textual prompts into vivid, realistic scenes. Plus, Kling uses a proprietary 3D Vae that's variational auto encoder and supports various aspect ratios thanks to variable resolution training. Basically, it can handle different video dimensions and still produce high quality output. One of the standout features of this model is its advanced 3D face and body reconstruction technology. It allows it to create videos where characters show full expression and limb movements, all driven from a single full body photo. So essentially this tech is what makes the videos look so lifelike and consistent. Now this whole thing is making it pretty clear that China is seriously stepping up its game when it comes to AI development. Cling is just a glimpse of what's coming out of the country and and it's giving us a hint that they might be ahead of the curve. OpenAI has said they'll release Sora by the end of the year, but with Kling already out, they might have some catching up to do. The big question though, is whether China will make Kling available worldwide. Right now it's accessible through the Kua I app, but you need a Chinese phone number to use it now. Cling isn't the first AI video generation model from China. Back in April, Kuaishou released VDU AI, which could create 16 second videos in 1080p resolution. Kling is the next evolution of that, offering longer videos with even better quality. If you check out Kling's website, you'll see some of their demo videos, and let me tell you, they're pretty impressive. For example, there's a video of an emperor angelfish swimming in a rocky underwater habitat, a man riding a horse in the Gobi desert during sunset, and even a white cat driving a car through a busy urban street. These videos showcase the model's ability to handle complex sequences, scenes and movements while maintaining high quality. The technology behind cling involves a 3D spatiotemporal joint attention mechanism, which helps it model complex movements and generate video content with larger motions that conform to the laws of physics. So when you see a man riding a horse in the desert, the horse's movements, the dust trails, and even the sunset in the background all look just right. Cling also uses efficient training infrastructure and extreme inference optimization. This allows it to generate videos up to 2 minutes long at a smooth 30fps FPS. One example they showed was a little boy riding a bicycle through different seasons in A garden. The video maintained high consistency and detail throughout, which is no small feat for AI generated content. Another cool thing about this model is its strong concept combination ability. Like it can take different ideas and merge them into a single coherent video. Like this prompt White cat driving a car through a bustling city. Footage like that doesn't exist in real life, but Kling can create it from scratch, making it look believable. It also excels in movie quality image generation. The AI can produce videos that look like they were shot with a professional camera crew. One example they showed was a man and a woman walking under a starry sky with the stars moving in the background. The level of detail and quality in these videos is truly cinematic. What's more, Cling supports various video aspect ratios. Whether you need a square video for Instagram, a portrait for TikTok, or a landscape for YouTube, it can actually handle it all. And this flexibility is super useful for content creators who want to use the same video across different platforms. So let's talk about some more examples from the demos. One video shows a Chinese man eating noodles with chopsticks. The details are so precise that if you saw this at a lower resolution, you might not even realize it's AI generated. Another video shows a chef chopping onions in a kitchen. And the way the onions are cut and split by the knife is incredibly realistic. But the video I mentioned before, a cat driving a car through a busy street. This one is particularly impressive because it combines a lot of different elements. Cars, buildings, pedestrians, and the cat's movements, all in one seamless video. Another demo features a volcano erupting inside a coffee cup, showing off Kling's ability to create fictional scenes that look convincingly real. And one of my favorites is a Lego character visiting an art gallery. The video captures the unique way LEGO characters move, just like in the Lego movies. There's even a clever focus shift from one character to another, showing that Kling can handle depth and focus changes just like a real camera. Another standout feature is Kling's ability to simulate real world physics. One demo shows milk being poured into a cup, with the milk flowing steadily and filling the cup. Realistically, this might seem simple, but it's actually quite challenging for AI to get these physical interactions right. Cling's technology is also capable of generating videos with temporal consistency. This means the AI maintains a logical flow and coherence over longer videos. In one example, a video of a train traveling through different landscapes stays consistent for the entire two minutes, which is really impressive. Overall, Kling is showing US that China is rapidly advancing in AI video generation technology. They're not just keeping up with the best models from the us in some cases they're surpassing them. This could lead to a competitive race in AI development, with countries striving to outdo each other, which might bring both exciting advancements and potential risks. I can't wait to see how OpenAI responds to this cling AI model from Kuaishou. It definitely seems like they're going to want to get their SORA model out there sooner than originally planned to keep up with the competition. But while we're all eagerly awaiting OpenAI's response and reaction to cling, we're also getting some really interesting news from the company itself. In a surprising move, OpenAI has revived its robotics team after disbanding it three years ago. With AI driven robotics investments surging, the company is now actively hiring research engineers for this newly reassembled team. While specifics are under wraps, job listings indicate the engineers will focus on training multimodal models, developing features for partners and optimizing core models. Interestingly, OpenAI aims to integrate its tech into other companies robotic systems rather than competing directly. Over the past year, OpenAI's venture fund has invested in several humanoid robotics companies like Figure AI, whose robot leverages OpenAI's models. This collaboration hints at a promising future for AI powered robotics. Robotics has always been key to OpenAI's mission, and this revival suggests a strategic pivot to capitalize on the integration of AI and robotics. Anthropic has just launched Claude 3.5 Sonnet, a new AI model that's being compared to OpenAI's GPT4O in terms of performance. They've also introduced some exciting new features, making Claude 3.5 Sonnet more skilled at understanding humor, handling complex workflows, and interpreting charts and graphs. Alright, so what's the deal with Claude 3.5 Sonnet? Well, it's Anthropic's newest AI model and it's already generating some pretty big hype in the AI world. But let's start with the basics. Claude 3.5 Sonnet is part of Anthropic's AI model lineup. They've got this whole naming system going on Haiku for the smallest model, Sonnet for the middle one, and Opus for the top tier. It's a bit quirky, but hey, every AI company seems to have their own weird naming conventions these days. Now Anthropic is claiming that Claude 3.5 Sonic can go toe to toe with or even outperform some of the heavy hitters in the AI world. We're talking about models like OpenAI's GPT4O and Google's Gemini 1.5. That's a pretty bold statement, right? Anthropic says that 3.5 Sonnet is actually better than their previous top model, Claude 3 Opus. And get this, it's apparently twice as fast. That's a huge deal when it comes to AI performance. Now Anthropic has released some benchmark scores and I've got to say they look pretty impressive. Claude 3.5 Sonnet outscored GPT4O, Gemini 1.5 Pro, and even Meta's Llama 3400B in most of the benchmarks they tested, and this includes areas like graduate level reasoning, undergraduate level knowledge, and coding skills. But here's the thing, we always need to take these benchmark scores with a grain of salt. The AI world moves so fast that today's top performer could be old news tomorrow. Plus, companies can cherry pick the benchmarks that make them look good. So while these scores are definitely promising, we'll have to see how Claude 3.5 Sonnet performs in real world applications. Speaking of real world applications, let's talk about what this new model can actually do. According to anthropic, Claude 3.5 sonnet is much better at writing and translating code. It can handle complex multi step workflows more efficiently. And here's a cool one. It's apparently way better at interpreting charts and graphs. But there's one improvement that I find particularly interesting. Anthropic says that this new Claude is better at understanding humor and can write in a more human like way. Now that's something I'd love to see in action. An AI assistant that can actually get your jokes and make you laugh. Oh, and here's a neat little tidbit. Claude 3.5 sonnet can apparently transcribe text from images more accurately. That could be super useful for all sorts of applications, from digitizing old documents to helping with visual accessibility. Now let's talk about availability. If you're itching to try out Claude 3.5 Sonnet, you're in luck. It's already available for free on Claude AI and the Claude iOS app. If you're a subscriber to Claude Pro or their team plans, you'll get higher usage limits. And for the developers out there, you can access it through Anthropic's API, Amazon Bedrock, and Google Cloud's Vertex AI. Also, Anthropic has set up a pretty affordable pricing model for this AI through Anthropic's API. It costs $3 per million input tokens and $15 per million output tokens. This basically means every time you feed information to the AI or get results back, you're using tokens. And these prices are quite competitive in the AI market. Another cool thing is the 200k token context window. This might sound technical, but it's actually really important. It means Claude can handle much larger chunks of information at once. So if you're working on a big project that involves a lot of data, Claude can process it all without getting overwhelmed. But Anthropic isn't just improving their AI model. They're also rolling out a new feature called Artifacts. And this is pretty cool, folks. Basically, it lets you see and interact with the results of your request to Claude right in the app. So if you ask Claude to design something for you, you can now see what it looks like and even edit it right there. Think about it. If Claude writes an email for you, you can edit it directly in the Claude app instead of having to copy it to a text editor. It might seem like a small thing, but it's actually a really smart move. These AI tools need to evolve beyond just being chatbots, and features like Artifacts are a step in that direction. This Artifacts feature might be giving us a glimpse into Anthropic's long term vision for Claude. They've always said they're mainly focused on businesses, even though they've been hiring some big names from the consumer tech world. In their press release, they talked about turning Claude into a tool for companies to securely centralize their knowledge, documents and ongoing work in one shared space. That sounds less like a chatbot and more like a full fledged productivity platform, doesn't it? We might be looking at something that could compete with tools like Notion or Slack, but with Anthropic's powerful AI models at the core, that's a pretty exciting prospect. If you ask me. The pace of improvement in AI is just more mind blowing. Anthropic launched Claude 3 Opus in March, saying it was as good as GPT4 and Gemini 1.0. Then OpenAI and Google released better versions of their models. And now, just a few months later, Anthropic is back with Claude 3.5 Sonnet. Now, I know Claude doesn't get as much attention as Gemini or ChatGPT, but make no mistake, it's very much in the race. And with improvements like these, it's definitely a contender to watch. Let's talk a bit more about some of the specific improvements in Claude 3.5 Sonnet, Anthropic did an internal evaluation of what they call agentic coding. Basically, they tested how well the AI could fix bugs or add new features to an open source code base when given a description of what needed to be done. Here you're going to see Claude edit the function file to fix the bug. And now Claude's going to rerun those tests. And the tests are passing. So now if we rerun the function look, our image no longer has that white background. Thanks, Claude. Claude 3.5 Sonnet solved 64% of these problems, compared to only 38% for the previous model. That's a huge jump. Now let's address safety and privacy, because these are huge concerns when it comes to AI. Anthropic says they've put Claude 3.5 sonnet through rigorous testing and trained it to reduce misuse. They've even brought in external experts experts to evaluate the model's safety, including the UK's Artificial Intelligence Safety Institute. They've also incorporated feedback from outside experts to make sure their safety evaluations are robust and up to date. For example, they worked with child safety experts from an organization called Thorn to update their classifiers and fine tune their models. And here's some reassuring news for those concerned about data privacy. Anthropic says they don't train their generative models on user submitted data unless the user explicitly gives them permission to do so. That's a pretty strong stance on privacy in a world where data is often seen as the new gold. So what's on the horizon for Anthropic? They're not taking a break anytime soon. Later this year, they plan to roll out Claude 3.5 Haiku and Claude 3.5 Opus, completing the Claude 3.5 model family. They're also developing exciting new features like one called Memory, which will enable Claude to remember user preferences and interaction history, making the AI experience more personalized and efficient. They're also exploring new modalities and features to support more use cases for businesses, including integrations with enterprise applications. It's clear that Anthropic is gunning for the business market in a big way. Now, I know we've covered a lot of ground here, but there's one more thing I want to mention. Anthropic is really emphasizing their commitment to improving the trade off between intelligence, speed and cost. They're aiming to make substantial improvements in this area every few months. That's an ambitious goal, but if they can pull it off, it could really shake up the AI industry. It's an exciting time to be following these developments and I can't wait to see what comes next. What do you think about Claude 3.5 sonnet? Are you excited to try it and see how it compares to other AI models? Let me know in the comments below. In this video, we will be talking about one of the most incredible, incredible breakthroughs at the intersection of neuroscience and artificial intelligence. Researchers from Harvard University and Google DeepMind have achieved something really astonishing. They've created an artificial brain for a virtual rat that can control the rat's movements in an ultra realistic physics simulation. This groundbreaking work, which was published in the prestigious journal Nature, opens up huge new possibilities for understanding how real brains work, work and how they control complex behaviors. It could also lead to more advanced and adaptive robots in the future. So let's break down exactly what this Harvard DeepMind team has accomplished and why it's such a game changing achievement. Alright, so the first monumental task was constructing an accurate biomechanical model of a rat's body in a sophisticated physics simulator called Mujoko. This virtual rat had to obey the laws of physics with factors like gravity, friction and the musculoskeletal mechanics of a real rodent's body coming into play. The researchers drew from a vast dataset of high resolution motion data recorded from real rats engaging in all kinds of natural behaviors and movements. This gave them an incredibly rich source of information to build and validate their virtual rodent model. But having an anatomically accurate rat body was just the first step. The team then had to create an artificial neural network that could learn to control this virtual body's biomechanics and and replicate the diversity of movements seen in the biological data. This is where Google DeepMind's expertise in machine learning came into play in a huge way. The DeepMind researchers collaborated closely with Harvard to apply advanced deep reinforcement learning techniques to train the artificial neural network which would serve as the virtual rat's brain. Specifically, they used an approach called inverse dynamics modeling, which is how our own brains are theorized, to control complex movement. When you reach for a glass, your brain doesn't manually move each muscle. Instead, it rapidly calculates the desired trajectory and translates that into the required forces and torques to make it happen smoothly. The virtual rat's neural network was fed reference motion trajectories from the real rat data as inputs. Through deep reinforcement learning, it learned to output the precise pattern of forces that could actuate the virtual body's joints and musculature. To successfully recreate those trajectories in the physics simulation. What makes this even more remarkable is that the neural network could generalize what it learned to produce realistic rat behaviors and movement sequences that it was never explicitly trained on. Just like a biological brain, it exhibited broad generalization capabilities. Now, with this virtual rat brain successfully controlling the biomechanical model, the researchers could then probe the activations and dynamics within the neural network to gain insights into how real rat brains might control movement. Stunningly, they found that the patterns of neural activity in the virtual brain lined up incredibly well with neural recordings made from motor cortex and other brain regions in behaving rats. This suggests the deep learning algorithm discovered internal models and motor control principles similar to those employed by biological brains. One of the key properties the virtual brain exhibited was the ability to spontaneously transition between different operational regimes based on context, closely mirroring how rodent brain dynamics are observed to switch between distinct patterns for different behaviors like grooming, running, or rearing. The researchers could also analyze how the network dealt with redundancy. Since there are typically multiple ways to achieve a given movement trajectory, how did it choose the optimal solution? The virtual brain appeared to implement a minimal intervention strategy to make only the minute corrections needed, avoiding unnecessary necessary expenditure of energy or forces. Now this aligns with theories of optimal feedback control that neuroscientists believe biological motor systems evolved to implement. The virtual rat's brain was discovering these principles from scratch simply by trying to generate naturalistic movement. Another powerful insight came from analyzing how variability or noise in the neural activations mapped to variability in the virtual rat's kinematics and dynamics. There were clearly structured patterns in how neural fluctuations cause specific types of movement deviations. This kind of variability signature is essentially invisible from just looking at the overall movements, but a window into it could shed light on the neural coding strategies used by the brain. Having a fully observable and controllable virtual system made these insights possible. Alright now, while there's still much more to explore, it's clear this virtual rat brain has opened up a new paradigm for investigating motor control and broader brain function that was simply intractable before. Rather than being limited to just recording neural signals during behavior, neuroscientists can now probe and perturb an accessible model of the entire brain. Body environment, control loop in simulation It's a new frontier of what some have dubbed virtual neuroscience. This virtual rat platform provides a convenient sandbox for testing theories about how neural circuits implement specific specific computational processes like state estimation, predictive modeling, optimizing costs and rewards, and orchestrating coordinated patterns. Of movement. Even more powerfully, it allows constructing simulated neural networks with arbitrary architecture, connectivity patterns, neuron properties and learning rules, then seeing how they give rise to emergent dynamics and behavioral capabilities. It's an unprecedentedly transparent window into the neural mechanisms behind both overt actions and the COVID cognitive processes supporting them. Perfectly controlled experiments and causal manipulations become possible in a way that's vastly more difficult with biological specimens alone. There are also exciting opportunities to use these kinds of virtual brain body models to simulate neurological conditions or injuries. By introducing targeted perturbations or lesions, it could provide a powerful new way to gain insight insights into brain disorders and test putative therapies or neural prosthetics in silico before animal trials. And then even if we go beyond the neuroscientific implications, this advancement also has immense potential for revolutionizing robotic control by reverse engineering how biological intelligence emerges from distributed neural dynamics. While classical control theory has given us robots that can perform specific pre programmed routines, modern AI and deep learning has already shown an ability to generalize and respond to open ended real world environments in more flexible, intelligent ways. However, most existing robots are still incredibly clumsy and inefficient compared to even a simple animal nervous system's ability to coordinate dexterous movement using distributed sensor motor control loops deeply entangled with the physics of their embodiment. By studying how the virtual rat brain coordinates its virtual biomechanics with, roboticists may be able to abstract out the core principles and neural architectures responsible for this biological intelligence and port them into new robotic platforms. We could see robots that dynamically adapt their control strategies in response to their environments, develop realistic general movement skills, optimize force and energy expenditure like animals do, maintain robust operation despite sensor or mechanical failures, and ultimately become far more versatile and capable autonomous systems. Alright, now I want to talk about nuclear fusion, which is a hot topic lately. And while you might be wondering what nuclear fusion simulation has to do with any of this neuroscience stuff, well, while it may seem unrelated on the surface, there are deep connections in terms of the tools and approaches being used. As part of their work on studying massively complex scientific phenomena, the Researchers at Google DeepMind have developed Torax, an open source differentiable tokamak core transport simulator implemented using advanced machine learning frameworks like jax, Torax can simulate the flow of particles, heat and electrical currents inside the core of an experimental nuclear fusion reactor, which is an extraordinarily complex coupled system of non linear differential equations describing magnetohydrodynamic plasma physics. Like the virtual rat brain work a Key innovation in Torax is its tight integration with power powerful machine learning techniques to solve these kinds of incredibly high dimensional physical modeling challenges. Specifically, Torax leverages JAX to enable just in time compilation for lightning fast compute times as well as automatic differentiation to be able to compute gradients of the entire plasma simulation pipeline. This allows using gradient based optimization methods to calibrate the parameters and couple data driven machine learning surrogate models like neural networks trained on gyrokinetic turbulence simulations into the core physical calculations. So in essence, Torax combines high fidelity physics modeling with state of the art machine learning in a differentiable programming framework. An approach very similar in spirit to what enabled the virtual rat brain breakthrough. Both are exemplars of a powerful new paradigm for tackling the mind bogglingly intricate problems in computational neuroscience, biophysics, plasma physics and many other domains of complex systems analysis. In fields like materials science and chemistry, we could virtually prototype new materials by simulating their atomic and molecular dynamics using physics constrained machine learning models. This would speed up the design and discovery of novel compounds with customized properties for things like energy storage, catalysis and quantum computing. Similarly for aerospace engineering, virtual models with AI could drive breakthroughs by optimizing aircraft and propulsion system designs through realistic simulations of aerodynamics and turbulent fluid flows. Combined with learned surrogate models, the possibilities even extend to fundamental physics. Projects like the virtual muon experiments at Fermilab are using differentiable simulation and AI to analyze massive particle collision data, enabling new insights into the nature of matter and forces at the time same subatomic scale. As these virtual modeling capabilities improve, we might even be able to create digital twins or simulations of entire cities, societies, economies and ecosystems. Essentially massive multiplayer simulations to play out scenarios and policies before the real thing straight out of sci fi. Of course, as these virtual world simulations become hyper realistic, they raise profound ethical questions about preventing self awareness or or existential risks like what are the implications of virtually replicating conscious minds? Deep issues we'll need to tackle. But there's no doubt that the rise of virtual modeling and AI accelerated simulation is a pivotal shift for understanding and engineering complex systems across science, technology and even social systems. A new age driven by cutting edge simulations and AI. There are some exciting news from Google, but there's also a bit of controversy to unpack. So. So let's just get started. Google just announced a massive update with their video generation model called Veo. This model is part of their DeepMind project and now it can create video clips from a single reference image. Essentially you give it a picture and some text instructions, and it can generate a whole video clip based on that. For example, they showed off an image prompt where a woman's hands are holding a large amethyst crystal, her nails are painted purple, and she has a gold ring on her pinky finger. The new model took that and generated a video clip that looks just like the original style, but animated. As you can see, there's no loss in quality, no glitches or anything like that. It's incredibly realistic, maybe even a bit too realistic. Another example was an elderly woman wearing a straw hat and a pink jacket sitting next to a brown and white dog. Both are looking off into the distance with serene expressions. The text prompt for the video was a woman gazes lovingly at her loyal dog. It's pretty amazing how it captures the essence and details from just a single image. Now they also introduced Video FX Update, which now allows developers to create full HD videos from text prompts using veo. So whether you're an artist, content creator or developer, this tool could actually change the basics of how you produce video content. Google has also introduced a new Gemini AI button in the side panel of several Google Suite apps like like Gmail, Google Drive, Docs, Sheets and Slides. With this button, you can ask questions, write emails, and get summaries of documents and email threads. However, it's still unclear when these features will be available for everyone to use, as developers are still working on making the interfaces user friendly. Now, while this technology is super exciting, Google is facing some serious challenges with their AI search features, as we all already know. Recently they rolled out AI overviews in Google Google Search, designed to give you a quick summary of complex questions, kind of like a smarter, more detailed search result. However, there have been some hiccups. People on social media have pointed out some odd and inaccurate AI overviews. For instance, there were reports of AI overviews providing strange or misleading advice from user generated content on forums. One funny example was an overview about how many rocks someone should eat. Obviously, that's not a common or sensible question. Now, Google has been quick to address these issues. First, they rolled back the new AI Overviews feature in Search, and then they released a post saying that they've made technical improvements to better detect nonsensical queries and limit satire or humor content. They've also refined how they handle user generated content to avoid misleading information. Plus they've put additional guardrails in place for sensitive topics like news and health. It's clear Google is committed to improving these tools, but the recent mishaps have highlighted some of the challenges they face in making sure their AI is accurate and reliable. Alright, now we also have some news regarding Elon Musk's AI company xai and their latest developments with the Grok AI chatbot. If you're already familiar with Grok, you know it's a pretty impressive tool that's available online and through the X social platform. Well, it looks like XAI is gearing up to roll out two new modes for Grok, Socrates and dei. According to the X Daily News, XAI might soon introduce these two new modes. GROK already has a few modes, including Normal mode, Fun mode, and the recently announced Unhinged mode, which offers a more unpredictable and entertaining interaction style. First, let's talk about the DEI mode. This stands for diversity, equity and Inclusion. When you select this mode, Grok will be described as the Senior vice president of DE&I. Okay, even the Grok icon gets a makeover and becomes more colorful. The idea here is that Grok will handle responses with a focus on inclusivity and sensitivity. Now, there's been some chatter on social media about this mode. For example, Testing Catalog on X mentioned that they're curious if this is Elon Musk poking fun at companies like Google or if it's a genuine effort to promote DE and I. Honestly, I think it could be a bit of both. Knowing Elon, he loves to stir the pot. But it also seems like a must have addition given today's emphasis on inclusivity. According to another X post, this mode is already implemented and functioning on the standalone version of Grok, but it's not available to the public just yet. The standalone GROK is only accessible in the US for now, and if you ask Grok an incorrect question in this mode, it's designed to handle the response thoughtfully, addressing all the nuances. Now, the Socrates mode. Named after the famous philosopher, this mode isn't operational yet, but there are hints it's in the works. While details are scarce, we can imagine that Socrates mode might involve Grok taking on a more inquisitive and thoughtful personality, prompting users to think deeper and engage in more meaningful conversations. It's actually exciting to see how XAI is expanding grok's capabilities, making it more versatile and engaging for users. All right, now, if you're into making videos, podcasts, movies, or even games, you're gonna love this. Elevenlabs, already known for their AI generated human voices and music, has just launched a sound effects AI tool that lets you create custom sound effects just by typing in a prompt. So here's how it works you type in what kind of sound effect you need. Maybe it's an explosion, a creaky door, or some spooky background noise, and the AI generates up to 22 seconds of that sound for you. You get at least four downloadable audio clip options to choose from, so you can pick the one that fits your project best. Eleven Labs actually partnered with Shutterstock to build a library and train their AI model. Shutterstock is a big name in stock media and they've licensed their content to lots of AI companies like OpenAI, Meta and Google. This means the sound effects you get from ElevenLabs are based on high quality professional clips. Create any sound you can imagine. Now let's talk about the cost. The sound effects tool is free to use, but there are paid tiers if you want to use the sounds commercially. If you're on the free tier, you'll need to Give credit to 11Labs by including 11Labs DIO in your project's title. Free users get 10,000 characters per month for writing prompts for sound effects. It uses 40 characters per second of audio if you set the duration yourself, or 200 characters if you go with the default duration. There are already plenty of libraries out there for sound effects, but they can be pricey or hard to navigate. ElevenLabs says their tool is designed to generate rich, immersive soundscapes quickly, affordably and at scale. This could be a game changer for creators who need specific sounds without the hassle. Eleven Labs isn't the only one in the game though. Other AI developers like Stability AI and Meta are also working on their own text to sound generators. Stability AI has stable audio which can create music and sound effects. And Meta's audiocraft models can generate natural sounds like wind or traffic. Alright, finally, I have to tell you about this super cool invention from researchers at Cambridge University. A robotic third thumb. And it's not some weird extra finger made of flesh. It's a mechanical digit that's surprisingly easy to use. The researchers showed off this third thumb in a demo video and it looks like it can be used for all sorts of things. People were doing everyday tasks, playing games like Jenga, soldering, and even playing the guitar with it. It's pretty impressive. It's designed to be inclusive for all sorts of people. They tested it with folks aged from 3 to 96 years old. Out of 596 people, only 13 couldn't use it within the first minute. That's a 98% success rate. It didn't matter if you were young, old, left handed or right handed. Almost everyone got the hang of it quickly. So how do you control this third thumb? It's funny actually because there are no sensors or probes attached to your hand. Instead you use your toes. Moving your left big toe makes the thumb go up and down and your right toe moves it across your palm. You can make all sorts of precise movements just by wiggling your toes. The researchers believe this third thumb could be useful for a wide range of people, regardless of their age, gender, lifestyle or even whether they're tech savvy or not. It could help with daily tasks, work or just for fun. Plus, it might be a great tool for people who are missing a finger or two Exciting Stuff on the Agenda Today we're looking at the upcoming GPT5 model that's set to change the AI game. A potential team up between Apple and Meta, and some jaw dropping tests on robot dogs from China. Let's kick things off with GPT5 since that's probably why most of you are here now. It's not officially called GPT5 yet, but given the naming pattern, that's pretty probably what it'll be called. According to Mira Muradi, the next GPT model is going to be a massive breakthrough in AI. Mira, who is the chief technology officer at OpenAI, recently shared some exciting insights. First off, let's talk about the intelligence levels. If you think about the earlier models like GPT3, Murati compared them to having the intelligence level of a toddler. They could do some pretty cool stuff but were still relatively limited. Then came GPT4, which Murati described as having the intelligence of a smart high school schooler. It was more sophisticated and capable of handling a wider range of tasks. Now the exciting part is that GPT5 is expected to be at a PhD level in terms of intelligence for specific tasks. To be honest, that sounds as a huge jump, but how do we get from a high schooler to a PhD? Murati explained that the secret sauce is a combination of three neural networks, a ton of data and a massive amount of computing power. When you combine these, you get what we call data deep learning. Essentially, this allows the AI to learn and understand complex patterns in the data, which leads to improved performance over time. This combination has been driving the progress in AI and it's what will make GPT5 such a game changer. Muradi mentioned that these systems are already performing at human levels in specific tasks and this is only going to get better. Alright, let's address the big question. When is GPT5 coming out while there's no official release date yet, Muradi hinted that we can expect expect this next gen model to arrive in about 18 months. That's roughly a year and a half, which is pretty exciting considering this timeline. All the hype about reaching AGI by 2027 starts to make more sense. Whether that's a good or bad thing is up for debate, but it definitely seems possible now. Murati explained that the development involves a lot of research and testing. They don't just want to release a new model, they want to make sure it's safe and beneficial for society. This means working on both the capabilities of the AI and and the safety measures needed to manage those capabilities effectively. One of the key points, Muradi emphasized is that developing capabilities and ensuring safety go hand in hand. As the AI becomes smarter, it also becomes easier to direct and manage. Think of it like training a smart dog versus a less intelligent one. The smarter one understands and follows commands better, making it safer to interact with. This approach is crucial because with great power responsibility, OpenAI is very aware of the potential risks and is working hard to mitigate them. For example, they are developing methods to predict the capabilities of these models before they're fully trained. This helps them prepare the necessary guardrails to ensure the AI behaves as expected. Looking ahead, one of the most exciting possibilities is the development of AI agents. These are AI systems that can connect to the Internet, communicate with each other and collaborate with humans. Muradi talked about the concept of AI agents working together seamlessly, which could revolutionize how we interact with technology. These agents could handle tasks autonomously, freeing up our time for more creative and complex activities. So essentially, GPT5 is set to transform numerous industries. We're already seeing AI making waves in fields like finance, healthcare and customer service. But with the advanced capabilities of GPT5, the impact will be even more profound. Of course, with any technological advancement, there are concerns about job displacement. Murati acknowledged that while some jobs might be lost, new ones will also be created. The key is understanding how these AI systems can integrate into workflows and help people become more productive. To be honest, it's still early days and we don't fully understand the impact AI will have on jobs. But there's a huge potential for AI to augment human capabilities, making us more efficient and opening up new opportunities. Another important aspect is regulation. Murati and OpenAI are advocating for more regulation, especially for frontier models like GPT5 that have immense capabilities. They believe it's crucial to work with policymakers and regulators to ensure these technologies are used responsibly. This includes setting up guidelines for how AI should be used and implementing measures to prevent misuse. For example, they're working on watermarking and content policies to manage misinformation, especially given the rise of deepfakes and other potential issues. Alright, switching gears a bit from GPT5 remember a few weeks ago when there was talk about an OpenAI and Apple partnership? Well, recent reports suggest that Apple is also exploring options with other AI giants. According to the Wall Street Journal, the tech giant has been in discussions with Meta about potentially integrating Meta's generative AI models into Apple Intelligence, the AI suite announced at WWDC 2024. The report states that Meta, along with other companies developing cutting edge generative AI, is keen to leverage Apple's massive distribution network through its ubiquitous iPhone. This move would parallel Apple's existing App Store model, where users can choose from a vast array of third party applications. If these talks bear fruit, Apple's customers could potentially have the option to select which X Store external AI models they wish to employ in addition to Apple's proprietary AI systems. This versatility would allow users to tailor their AI experiences based on their specific needs and preferences. Meta's foray into the generative AI arena began with the release of its large language model Llama 2 in July 2023, followed by the more advanced Llama 3 in April of this year. While Llama has already gained significant traction with within the tech community, a partnership with Apple could solidify Meta's position as a dominant force in the AI landscape. However, it's important to note that these discussions are still in their early stages and there is no guarantee of a finalized deal. The Wall Street Journal cautioned that negotiations could ultimately fall through, leaving the potential partnership in limbo. Alright, now we've also got some pretty wild news from China, where they've been putting robot dogs to the test in some pretty extreme ways. Unitree Robotics, a Chinese company, recently shared footage of their robot dogs being kicked, hit with sticks and even thrown to the ground. Now, I know this might sound shocking and it really is quite intense to watch, but there's a purpose behind these harsh tests. The developers at Unitree Robotics are using these aggressive methods to improve the robot's balance and recovery algorithms. Basically, they want to make sure these robot dogs can handle strong and unexpected expected physical impacts and get back on their feet quickly. In the video, you can see the robots named Goto and B2 enduring all sorts of rough treatment. They're kicked, hit hard with wooden sticks and tossed around. But it's not all about the abuse. The video also shows these robots navigating rough terrain and even doing three consecutive flips without stopping. The impressive adaptability of these robots comes from a control algorithm based on reinforcement enforcement learning. This means they learn and improve from each action they take, getting better at maintaining their balance and handling tricky situations. Unitree Robotics is one of the first startups in the world to sell robot dogs publicly. The company was founded by Xing Shing Wang, who developed his first robot dog X Dog during his graduate studies in 2016. By September 2017, they introduced another four legged robot named Laikago, inspired by the famous space dog Laika. And recently, China's Central Television featured a report on joint military exercises between the People's Liberation army of China and Cambodia's army called Golden Dragon 2024. These exercises included practical scenarios using armed robot dogs. The robots used in these drills were modified versions of Unitree's Go to model, showing just how versatile and potentially useful these robots can be in various situations. Apple brought the heat in a major way with their new personal intelligence system called Apple Intelligence. We're talking a genuine game changer that could flip the entire AI game on its head. Now I can already hear some of you saying, but Apple is usually so behind the curve when it comes to AI, right? Let me stop you right there, because with Apple Intelligence, they've straight up leapfrogged the competition in certain areas. This smart system is packing some seriously powerful generative AI models integrated at the core across your iPhone, iPad and Mac devices. And it's not like it is providing generic one size fits all AI assistance. This thing actually taps into your personal context, habits, activities, all that good stuff to deliver intelligence that is tailored specifically for you and how you operate. This could be the first real personal AI companion that dates deeply understands you as an individual while still taking care of your privacy. Even though Elon Musk would not agree with this statement, but we will talk about that a bit later in the video. Anyway, according to Apple, with Apple Intelligence they are implementing intense security and encryption at every single step to ensure your personal data remains fully protected. And in this era of non stop data breaches and privacy fiascos, that kind of commitment is absolutely massive. Alright, now let's get into some of the cool new AI features Apple is bringing to the table with this system. One that immediately caught my eye is notification prioritization on your iPhone. We've all been there. Your phone is just relentlessly blowing up with notifications all day, every day. It gets so overwhelming that you end up missing something legit important. Well, Apple Intelligence aims to solve that by actually prioritizing and filtering your notifications notifications based on importance, minimizing distractions while ensuring you don't miss the crucial stuff. Your priority notifications appear at the top of the stack, letting you know what to pay attention to at a glance and to make scanning your notifications faster, they're summarized it is also powering brand spanking new AI driven writing tools that are integrated across Apple's ecosystem. So essentially we will have an AI writing assistant embedded into iPhone, iPad, Mac that helps us crafting emails, reports, social media posts, etc. Now here's where things get really interesting. Apple has introduced this functionality called Private Cloud Compute that allows Apple Intelligence to flex its computational muscle by seamlessly tapping into even larger, more powerful AI models running on Apple's servers. Basically the brain power of a supercomputer, but coupled with Apple's signature privacy and security protection, a brand new standard for privacy in AI and unlocks intelligence you can trust. These cloud based AI models are running on custom Apple silicon servers that were designed from the ground up with a heavy focus on security and transparency. They utilized Swift programming language which has robustness and protection baked into its core. And get this, Apple is offering complete transparency into how these systems operate, which is somehow a reassuring thing for me. But we're still just getting started. They've taken Siri, their iconic virtual assistant and integrated it even deeper into the overall system experience across Apple's products. When you chat with Siri now, you'll see this glowing light animating around the borders of your screen, creating this super immersive lifelike interaction. What does the weather look like for tomorrow at Muir Beach? Oh wait, I meant Muir Woods. The forecast is calling for clear skies in the morning near Muir Woods National Monument. Plus, thanks to major upgrades to Siri's language understanding capabilities, you can now talk to it in a much more natural conversational style. Even if you stumble over your words, rephrase things on the fly or use imperfect perfect grammar, create an event for a hike there tomorrow at 9am Hike is scheduled for 9am to 11am on June 11. I didn't have to mention mere woods again. Siri understood what I meant when I said there. Another innovative feature Apple unveiled is Image Wand. If you're sketching something on your iPad with an Apple pencil, you can just circle your rough sketch and the AI will analyze it along with any text you've written. Then it creates a cleaned up, high quality version of your image right in your notes or project. Oh, and Apple has also cooked up this new cleanup tool, leveraging their AI tech. Basically, it can automatically detect distracting objects or elements in the background of your photos and seamlessly remove them while keeping the primary subject intact. And now let me hit you with possibly the biggest bombshell of them all. Apple is integrating none other than ChatGPT directly into their ecosystem and experiences. So Siri can tap into ChatGPT's expertise when it might be helpful to for you. That's right, the pioneer. The one that started the generative AI revolution. ChatGPT will be woven into Apple's software stack in a huge way. Absolute insanity. You can also ask questions related to your documents, presentations or PDFs. You'll be able to access ChatGPT for free without creating an account. But not everyone think that's a great idea. While undoubtedly a huge move that supercharges Apple's AI efforts by leveraging ChatGPT's strengths, this controversial partnership is already drawing major backlash from none other than Elon Musk. The Tesla and SpaceX CEO who co founded OpenAI, before leaving over disagreements went on an angry rant about the news on his social media platform X. Musk stated that if Apple goes through with integrating OpenAI at the operating system level, he would consider it an unacceptable security violation and would ban Apple devices at his companies. He threatened that any visitors would have to surrender their Apple devices at the door, where they would be stored in a special Faraday cage that blocks crucial wireless signals like cellular, WI fi and Bluetooth. Musk accused Apple of being clueless about what happens once they hand your data over to OpenAI, claiming they are essentially selling you down the river with this partnership. This fiery reaction seems rooted in Musk's long standing feud with OpenAI, the company he helped establish but later departed amid clashes over its direction. OpenAI has dismissed Musk's claims as frivolous and simply him being jealous of their success without his involvement anymore. Despite Musk's theatrical threats of an Apple ban, it may be difficult for him to actually follow through, given how many major tech giants like Google, Samsung and Microsoft have already integrated advanced AI across their platforms and operating systems. Some experts suggest Apple's approach of trying to process AI locally on device as much as possible. Only engaging OpenAI after explicitly asking permission could make their implementation more privacy conscious compared to certain rivals methods. So at this point the only question is when we can actually get our hands on these incredible AI capabilities Well, developer betas for Apple's new operating systems with full Apple intelligence integration are already out for devs to start tinkering with. Then next month public betas will start rolling out so more people can take it for a spin. And finally this fall we're looking at the full stable release packed with all of these transformative AI features shipping out to everyone. I've got to say, Apple really swung for the fences and knocked it out of the park with this whole Apple Intelligence announcement. They fully embraced cutting edge generative AI and deep learning while still making user privacy, data security and personalization the absolute top priorities. But now I want to hear from you. Are you over the moon about these game changing AI integrations from Apple? Maybe you've got some skepticism or privacy worries you want to voice, or any other burning thoughts, predictions or theories you're itching to discuss. You might remember a little while back when I covered Robot Era, that ambitious startup that unveiled their crazy dexterous X hand AI robot. Well, they just one upped themselves in a major way by becoming the first company ever to have a full size humanoid robot walk on the Great Wall of China. So their latest robot, called the Xbot L conquered one of the most iconic and challenging landscapes in the world. The ancient stone pathways and steep stairs of the Great freaking Wall of China. This is a huge milestone not just for roboterra, but but for humanoid robotics as a whole. So a 5 foot 5, 165 cm tall robot, looking kind of like a slimmed down terminator, marching its way across those uneven century old stone surfaces without faltering. It was waving hello, practicing kung fu moves, climbing up and down stairs, the whole nine yards. And to be honest, the Great Wall is no walk in the park even for us humans. Those paths are cracked, the stairs are crazy steep with no railings. There are random potholes everywhere. It's an obstacle course designed to trip up any bipedal robot. But the XBOT L handled it all with surprising ease, thanks to roboterra's advanced perceptive reinforcement learning algorithms. Basically, this robot can actually perceive and make sense of its environment in real time. Using sensors and AI, it can identify obstacles, changes in terrain, and literally adapt its walking gait and balance on the fly to deal with whatever the great walk ball throws at it. According to Yue Xi, one of the co founders, their perceptive reinforcement learning tech gives the Xbot L something like human level awareness and decision making capabilities in unfamiliar environments. The robot can essentially think for itself about the best way to navigate complex areas safely and efficiently. And we're not just talking about some basic pre programmed motions here. The xbot L was climbing rugged castle stairs, dealing with slopes and inclines, avoiding obstacles, adapting to poor lighting conditions, all without any human guidance or help. It was basically pulling off the kind of adaptive, intelligent, situationally aware movement that we associate with people. Roboterra says. This was one of the biggest challenges, developing end to end AI algorithms that could translate perceived data from the robot's sensors into precise stable locomotion in the real world across an ultra complex environment like the Great Wall. Sounds like they cracked that nut though, based on this demo. What's really crazy is that this Great Wall feat comes just a couple weeks after they showed off the X hand dextrous robot I mentioned earlier. Between these two products, it's clear Robot Era is going all in on embodied AI building robots with human like physical capabilities to operate in our world. And they're not messing around either. Despite only being founded in 2023, they've already invested heavily into cutting edge AI tech like reinforcement learning, neural networks, sim to real transfer and more. Basically everything needed to create robots that can tackle the real world just like we do. Their goal seems to be keeping that competitive edge by iterating and churning out smarter, more capable humanoid robot products at a rapid pace. And they want to keep perfecting ways to transfer all that AI training from simulations into physical machines, enabling those robots to become more versatile in real life scenarios. The endgame seems to be developing ultra flexible general purpose humanoid robots for countless valuable applications across industries like manufacturing, healthcare services and more. I've got to give robots.