AI Shocks Again: Claude & Alexa, Smarter Robots, OpenAI ASI, Llama 3.2 & More (September News)

so AI has made some major moves this month and you're about to find out why from Alexa turning into claw AI to Google's Deep Mind AI accomplishing in minutes what used to take scientists years and a robot that gets smarter with every move it makes AI is moving fast open ai's latest Creations like the early launch of strawberry and the powerful GPT next have left everyone speechless we're also seeing new AI tools like snap's AI powered video generator and glasses while Isaac a learning personal robot is raising the bar plus Sam Altman's next big AI

device Google's mistake fixing Ai and the latest competition between llama 3.2 and open AI models so get ready for the full breakdown Amazon is making a major change to Alexa replacing its own AI with anthropics AI model Claude the new remarkable Alexa promises more advanced natural conversations but it also comes with a few surprises is this the next big step for smart Assistance or is Amazon taking a gamble with this move let's discuss and find out in today's video all right so Amazon has been working on this upgraded Alexa internally they're calling it remarkable Alexa

or project banion for a while now and originally they were using their own AI technology for it but apparently things didn't go so well in the testing phase according to Reuters Amazon's in-house AI was having a lot of trouble responding quickly to user prompts I mean we're talking about a lag of up to 6 or 7 Seconds sometimes and in today's fast-paced world where we're all used to instant responses think chat GPT or Google Assistant that's just not going to cut it so Amazon decided to switch gears and use AI models from anthropic a San

Francisco based AI research company that was actually founded by some exop AI folks anthropic has been making waves with its AI models especially their Claude model which is pretty good at understanding context and maintaining more natural human-like conversation and apparently it performed way better than Amazon's own AI in those crucial tests Amazon has invested a whopping 4 billion in anthropic which might explain why they're leaning on claude's Tech for this new Alexa all right so let's talk about what this new remarkable Alexa is actually going to do Amazon is saying this upgraded version is going

to be a lot more advanced than what we've got right now the idea is that you'll be able to have more complex and context aware conversations with Alexa for example you might asket for shopping advice like what clothes you should pack for a Beach vacation in Bali or to aggregate news stories based on specific interests also Alexa will be able to handle more complex multi-step tasks from a single prompt like if you say Alexa order me a pizza draft an email to my boss and set up a reminder for my dentist appointment it will all

be done but here's the catch Amazon plans to offer this upgraded Alexa as a paid subscription service so unlike the classic Alexa that we've been using for free the remarkable Alexa is going to cost somewhere between $5 to $10 a month and just to be clear this won't be part of your Amazon Prime subscription so it'll be an extra cost on top of that Amazon is betting that these new Aid driven features will be worth the price tag but it's definitely a gamble I mean asking people to pay for something that's been free all along

is always a tricky move some folks inside Amazon are even skeptical that customers will go for it especially since many are already shelling out $139 a year for Prime so why is Amazon doing this now well it all comes down to money and competition despite Alexa being super popular Amazon says they've sold over 500 million Alexa enabled devices The Voice Assistant division hasn't been making as much money as Amazon would like the company's leadership has been pushing hard for Alexa to start generating more revenue and they see 2024 as a crucial year to prove Alexa's

worth in the market the idea here is that by offering a more advanced paid version of Alexa Amazon can start turning it from a cost center basically something that costs more money to maintain than it brings in into a revenue generating machine and with around 100 million active Alexa users if even 10% of them decide to go for the paid version that could mean at least $600 million in annual revenue assuming they go with the lower end of the price range now this move to partner with anthropic is pretty interesting because it marks a bit

of a shift in strategy for Amazon typically Amazon Amazon likes to build its own Tech from the ground up which gives them full control over the user experience data collection and all that good stuff but with the rise of other Tech giants like Microsoft and Apple partnering up with open AI to integrate chat GPT into their products Amazon probably felt the Heat and realized they needed to step up their game to keep up and it's not just about keeping up appearances the AI race is heating up and companies are scrambling to offer the most advanced

most intuitive and frankly most humanlike AI assistance out there the release of chat GPT last year really shook things up showing that people want AI that can handle more natural flowing conversations not just simple commands so Amazon like many others is trying to make sure they're not left behind so what else can we expect from this new Alexa for starters it sounds like it's going to be much better at integrating with other smart home devices and remembering user preferences think of it as your personal Butler who knows all your quirks and habits maybe you like

your life dimmed to a certain level at night or you always want the coffee maker to start brewing at 7:00 a.m. sharp this new Alexa should be able to handle all that seamlessly and for those of you with kids there's talk that the new AI could be tailored to offer special interactions with children it will be able to engage with your kids in a more meaningful way maybe helping them with homework playing educational games or even telling bedtime stories that are a little more interactive than what we've seen before but some people are a bit

worried about what this means for privacy and data collection by moving to a more advanced AI model and potentially handling more sensitive information there are definitely some concerns about how that data will be used stored or potentially shared Amazon hasn't gone into detail about how they'll handle these concerns but it's definitely something to keep an eye on now Alexa isn't the only AI project Amazon has been working on they've also been doing some interesting stuff with an internal AI assistant called Amazon q and if you haven't heard much about this one it's because it's more

of an internal tool that's being used to streamline software development across the company but get this Amazon Q has reportedly saved the company $260 million and 4,500 developer years how did it do that well by automating a lot of the repetitive processes that normally take developers a ton of time for example something like upgrading a system to a new version of java used to take 50 development days now it takes just a few hours that's a huge boost in productivity and efficiency freeing up Developers to work on more complex creative tasks and if that wasn't

enough Amazon is also making moves in AI robotics they recently hired the founders of a robotic startup called covariant which is working on some pretty advanced stuff like robotic arms that can perform common Warehouse tasks the cool part here is that covariant is developing what they call a large language model for robots essentially they're teaching robots how to understand and execute commands in a way that's more intuitive and less mechanical Amazon's been bringing in some serious brain power to help push this forward including the founders of cavant who are now part of the Amazon team

along with about a quarter of kavarian employees they're integrating this Tech into Amazon's existing robot Fleet which should make their warehouse operations even more efficient so it's clear that Amazon is doubling down on AI across the board from improving Alexa with anthropics clawed AI to enhancing their internal processes with Amazon q and even stepping up their robotics game it's all about leveraging AI to make their services smarter faster and more integrated but it also means that we as consumers are going to see more AI in our everyday lives whether that's through smarter voice assistance more

automated homes or even just better faster services from companies like Amazon it's an exciting time for Tech but it's also a time to stay informed and aware of how these Technologies are developing and what they mean for our privacy our jobs and our daily routines so what do you guys think would you be willing to pay extra for a more advanced version of Alexa or do you think Amazon should just keep it free and find other ways to monetize it what if AI could not only predict how proteins interact but actually create new ones to

fight diseases like c19 or cancer in this video we're exploring two groundbreaking AI models Alpha PTO and orb that are reshaping everything from drug to Discovery to Material Science and before we get started take a second to subscribe and stay ahead of the curve on all the latest AI Trends and updates all right you know how proteins are like the workhorses of our bodies right they do everything from helping our cells grow to keeping our immune systems running smoothly and of course they're at the core of pretty much every biological process but here's the catch

these proteins don't just do their jobs in isolation they interact with other proteins think of it like a key fitting into a lock one protein binds to another and that's how all these critical functions happen inside our cells now thanks to things like Alpha fold we already have a pretty solid understanding of how proteins interact Alpha fold helps us predict what these interactions look like which has been huge for science but here's the kicker while Alpha fold is awesome at figuring out how proteins connect it doesn't create new proteins that could manipulate or influence these

interactions and that's where the real magic comes in guys and enter Alpha proteo by Google Deep Mind this is the new AI system we're talking about today it doesn't just predict interactions it designs entirely new proteins that can bind to specific Target molecules and why is that important well these designed proteins also called binders can speed up all sorts of research from drug Discovery to figuring out how diseases work to making crops more resistant to pests now let's break this down when you want to create a protein that binds tightly to a Target like say

a virus protein that causes c19 it's not easy traditional methods are super slow you have to create binders in the lab test them optimize them then test them again it's like an endless cycle of trial and error which as you can guess takes forever but with Alpha Proto we're talking about a major shortcut this AI system has been trained on a ton of protein data we're talking about data from the protein data bank and more than 100 million predicted structures from Alpha Al fold so yeah it's learned a thing or two about how proteins interact

now if you give Alpha proteo the structure of a Target protein and tell it where you want the protein to bind it can design a binder protein that fits that Target almost perfectly that's like handing it the blueprint for a lock and having it design the perfect key and it works on all kinds of proteins cancer related proteins proteins from viruses like SARS COV 2 yes the one involved in the lockdown era the c19 and even proteins tied to autoimmune diseases so far the results are pretty mind-blowing in fact Alpha proteo generated binders for seven

different Target proteins and here's the kicker they tested these proteins in the lab and the AI designed binders worked like a charm I'm talking about 3 to 300 times better binding strengths than the best existing methods insane right for example let's take V GFA which is a protein linked to cancer and diabetes complications for the first time ever an AI system Alpha ptao designed a protein binder that successfully binds to vgfacts actually bound to bhrf one successfully and get this Alpha Proto binders on average are binding 10 times stronger than any of the current best

designs now one of the most hyped targets was the SARS c 2 Spike protein the very same Spike that helps the virus get into our cells and yeah Alpha proteo nailed it not only did it design binders for this Spike protein but those binders were tested by some top research groups like The Francis Crick Institute and they confirmed the binders were legit these binders even managed to block the virus and some of its variants from infecting cells so we're looking at something that could potentially help in virus prevention now obviously this Tech isn't perfect yet

for example Alpha proteo couldn't design a successful binder for tnfa which is a protein associated with autoimmune diseases like rheumatoid arthritis but to be fair tnfa is known to be a beast in terms of difficulty the team picked it on purpose to test the limits of the system so it's not all bad news in fact it's a sign that they're pushing the system to improve and while strong binding is critical it's just the first step in using these proteins for real world applications like drug design there's still a ton of bio-engineering work to do but

Alpha proteo has already knocked out one of the toughest parts now the team behind Alpha proteo is working with Scientists across the world to make this system even better and they're thinking responsibly about biocurity that means making sure this powerful Tech is used for good like developing treatments and cleaning the environment rather than you know any Shady stuff and if you're wondering where this is going well they've already teamed up with groups like The Nuclear threat initiative to set up best practices so the tech is evolving but with caution which is honestly a relief looking

ahead they're also exploring drug design applications with a company called isomorphic labs and they're actively working on improving the algorithms to make the system even stronger and more versatile and they're not just doing this in a vacuum they're collaborating with experts in machine learning biochemistry and structural biology so the future of protein design yeah it's looking pretty exciting all right now there is another thing I'm seriously hyped about and it's actually a massive breakthrough in Ai and Material Science we're talking about orb the latest and greatest AI model for simulating Advanced Materials it's open- Source

it's blazing fast and it's leaving Big names like Google and Microsoft in the dust if you're into AI energy or just cuttingedge Tech you're going to love this so let me break it down for you imagine you're a scientist working on materials for say better batteries or solar panels things that are crucial for the energy transition right well here's the thing to design these super efficient materials you need to know exactly what's happening at the atomic level we're talking about how atoms and molecules are interacting what makes certain materials conduct energy better or how you

can tweak them to be more efficient but and this is the tricky part actually seeing or simulating what's happening inside these materials is insanely hard traditional methods for simulating this stuff they're slow costly and often involves simplifying things so much that you're not even getting an accurate picture it's like trying to watch a 4K movie on dialup Internet it's just not happening and that's where AI comes in giving us a new way to look at these materials with way more detail without waiting forever this brings us to orb the model we're talking about today built

by a company called orbital orb is designed to simulate materials at the atomic level faster and more accurately than anything else out there right now and get this it's based on a bigger AI model they've been working on internally called Linus so basically they've been fine-tuning this thing for a while now and it's paying off in a huge way now orb isn't just faster than the competition it's five times faster than the best alternatives for large scale simulations that's a huge leap forward and we're not just talking about beating random models either orb is outperforming

Google's and Microsoft models when it comes to accuracy and speed and here's where things get even more exciting they've open- sourced it yep it's free for non-commercial uses in startups so anyone looking to develop new materials can jump in and use this Tech you can even go to their GitHub right now and check out the full technical breakdown now let me pause here for a second to explain why this is so important we're in the middle of a massive shift toward renewable energy and materials are at the heart of that whether it's batteries for electric

cars solar panels for homes or semiconductors for basically all of our Tech the materials we use need to get better more efficient longer lasting you name it and the faster we can simulate and design these materials the faster we can make them a reality orb is a tool that's going to accelerate that process in a big way now if you're wondering how orb came to be it all ties back to this Foundation model I mentioned earlier Linus the team at orbital has been training and refining Linus from the ground up orb is like Linus on

steroids specifically fine-tuned for advanced material simulations they've got a whole blog explaining the key elements if you want to dive into the technical side and they've got even more info coming soon so yeah this isn't some random new AI model it's been a long time coming and can we just take a second to appreciate the team behind this this isn't some massive Tech Giant with endless resources we're talking about a small tight-knit group that's competing with the biggest names in AI Google Microsoft and so on it's proof that even in an era where it seems

like only the Giants can make big moves a scrappy motivated startup can still come out on top so there you have it or the fastest and most accurate AI model for advanced material simulations is out there and it's free to use for non-commercial purposes I can't wait to see where this goes as always drop your thoughts in the comments are you as excited about this as I am powered by Advanced Ai and learning from every move this machine is built to handle everything from heavy lifting to intricate tasks its ability to adapt in real time

is reshaping how Industries operate making work smoother and smarter all right let's talk about the real power behind the Pudu D7 a service robot that's been designed with purpose and practicality in mind it's a semi-humanoid robot which means it's not built to look fully human but it borrows some useful features from humanoid designs this robot is all about efficiency adaptability and intelligence something that's going to be increasingly common as automation becomes more embedded in industry IES like retail hospitality and Healthcare the Pudu D7 stands at 1.65 M or 5'4 tall and weighs about 45 kg

around 100 lb the key to its design is in its upper body which resembles a human torso in arms and a wheeled base for movement it's important to mention that these wheels aren't just for show they allow the robot to move in any direction with high Precision that omnidirectional base ensures the D7 can navigate even in complex environments where space is tight like busy kitchens or Hospital corridors the speed tops out at 2 m/s or roughly 7.2 km per hour this allows it to keep Pace with the flow of human activity in places like restaurants

or public areas without being too slow or too disruptive one of the most notable features of the Pudu D7 is its arms each arm is equipped with 30° of Freedom providing a lot of flexibility in terms of movement that's crucial for handling a variety of tasks the arms themselves are about 65 cm long which might not sound like much but this is just enough to get the job done in most service settings what's even more interesting is that these arms aren't static depending on the task at hand they can be outfitted with different attachments including

humanlike hands in its most advanced configuration the robot can have 20° of freedom in its humanoid hands which brings a great deal of dexterity to tasks like picking up items or interacting with objects in a way that most simple robots cannot what sets the semi-humanoid robot apart from standard service robots is the level of intelligence packed inside hudu robotics has embedded a multi-layered intelligence system into the D7 it combines datadriven intelligence with Advanced AI models which allows it to handle complex scenarios let's take a moment to consider what this really means instead of just following

pre-programmed instructions the D7 learns and improves over time each interaction it has whether it's serving a meal in a restaurant or transporting goods in a hospital is logged and used to refine its decision-making process the robot not only understands its environment but adapts to it becoming smarter as it goes along Pudu robotics developed this system with a clear focus on practical use in real world environments take a retail space for example specialized robots might be great at stocking shelves or greeting customers but they can't adapt to unexpected situations very well the Pudu D7 however is

designed to fill that Gap it Bridges the space between those simple task specific robots and full humanoid robots which are still incredibly expensive and tricky to implement in that sense it hits The Sweet Spot offering a good amount of adaptability without the high costs associated with creating a fully humanoid machine the arms of the D7 can lift 10 kg each meaning it's built to handle some significant weight without compromising on Precision with a Precision of up to 0.1 mm it can perform tasks that require accuracy such as picking up fragile items or manipulating small objects

that kind of precision is crucial for environments where errors can be costly or dangerous such as in health care or industrial settings this robot isn't just built for basic service jobs it's equipped to handle complex tasks like operating elevators or sorting through Goods it's design is aimed at Industries where human robot interaction is essential which is why it's being deployed in set settings like hospitals and restaurants where the ability to communicate with people and execute tasks reliably is Paramount and because it runs on a battery that exceeds 1 kwatt hour it's able to operate for

more than 8 hours straight this extended battery life is essential for environments like hospitals where a robot may need to be running around the clock without frequent downtime for charging another Technical Edge the D7 has is its high level and low-level control planning this is where the intelligence system really shines highlevel plan takes care of strategic tasks like figuring out the best way to approach a customer or transport an item low-level planning on the other hand handles the real-time actions like adjusting its grip on an object or ensuring it avoids obstacles as it moves through

a crowded space this hierarchical system allows the robot to act on both broad strategic tasks and detailed realtime commands that's what makes it particularly powerful in environments where things can change quickly and the robot needs to adapt on the Fly the d7s mobility is another key selling point its Wheels aren't just for smooth linear movement they allow the robot to move 360° in any direction this means the D7 can make quick adjustments without having to physically turn its body making it incredibly Nimble in confined spaces it's also built to handle slopes up to 10° which

adds to its versatility whether it's moving through a flat kitchen floor or navigating a ramp this robot maintains stability and balance Pudo robotics has been deploying service robots for a while now with over 80,000 units already shipped out to Industries like Hospitality retail and Healthcare their robots are known for performing specialized tasks like food delivery or cleaning but the Pudu D7 represents a leap into more adaptable humanlike interactions this marks a significant shift in how we think about service robots the introduction of semi-humanoid robots like the D7 allows companies to reduce operational costs while enhancing

customer experiences by creating a robot that can handle more than just one type of task for Pudo the D7 is part of a broader Vision they've been working towards an ecosystem where specialized robots semih humanoids and fully humanoid robots coexist each performing specific functions to streamline operations the idea is to create an environment where robots are doing the repetitive or physically demanding work allowing human employees to focus on more strategic tasks the D7 with its ability to learn and adapt fits perfectly into this Vision it's built to be more than just a tool it's meant

to be an active participant in the workforce it's worth noting that Pudu robotics has invested heavily in research and development they have nearly 1,000 authorized patents to their name which shows just how much they've been innovating behind the scenes the D7 isn't just the product of random experimentation it's a carefully crafted solution based on years of expertise in robotics this focus on Innovation is what has made Pudo one of the leading names in the service robotics industry and the D7 is a clear representation of how far they've come as we look ahead Pudu is planning

to fully commercialize the D7 in 2025 by that time we're likely to see these robots in even more Industries performing a broader range of tasks from lifting items and warehouses to assisting patients in hospitals the potential applications are vast it's not just about replacing human labor it's about making those tasks more efficient and reliable with the added benefit of reducing human error there's a clear direction that Pudo robotics is heading with this robot their focus isn't just on developing impressive machines but on creating robots that genuinely improve the Way businesses operate the Pudu d7s ability

to adapt learn and handle complex interactions with people makes it a key player in the future of service robotics in the next couple of years it's likely we'll see even more advancements from Pudu Robotics and the D7 might just be the beginning of a new era where semi-humanoid robots become commonplace in our day-to-day environments with its mix of technical sophistication real world practicality and long-term Vision the Pudo D7 is certainly a robot worth paying attention [Music] to so open AI is on the verge of releasing its latest model code named strawberry and it looks like

it's coming sooner than we thought within the next 2 weeks in this video we'll break down what makes strawberry unique including its new approach to reasoning and how it differs from previous models we'll cover its technical specs how it integrates with chat GPT and potential pricing changes plus we'll explore the challenges strawberry might face and its intriguing backstory including its connection to artificial general intelligence and the star method we'll also talk about the latest rumors and speculations about its future features and applications so stick around to get the full scoop on open ai's exciting new

AI model model to kick things off let's unpack what strawberry is all about it's open ai's newest AI model and it's designed to bring a new level of reasoning and problem solving to the table originally slated for a fall release this model is now set to launch ahead of schedule and the Tech Community is buzzing with anticipation unlike its predecessors strawberry is not just about generating answers quickly it's built to focus heavily on reasoning and analytical thinking while previous models like gp4 and GPT 4 are known for their rapid responses strawberry takes a more deliberate

approach it's designed to spend between 10 and 20 seconds processing a question before delivering a response this might sound slow but the idea is that this extra time allows for more accurate and thoughtful answers especially for complex problems but why would an AI model purposely take longer to respond the reason is pretty intriguing by taking its time strawberry aims to reduce the chance of errors and improve the quality of its responses for tasks that involve multi-step reasoning like solving intricate math problems generating detailed business plans or even programming this approach could significantly enhance performance now

the model is set to be integrated into open ai's chat GPT platform but will function as a standalone option within it this means that users will have the ability to select strawberry from a list of models in the chat GPT interface although the exact details on how users will access strawberry haven't been confirmed yet it's expected that users will choose it based on their specific needs regarding pricing strawberry is rumored to have a unique structure compared to existing GPT models instead of being available for free or through a subscription model it might introduce a pricing

tier that limits the number of messages users can send per hour there's speculation that there could also be a premium option for those who want faster responses or additional features existing chat GPT subscribers are likely to get Early Access before it becomes available to free users despite its Advanced features strawberry is not without its limitations for starters it will only handle text based queries at launch this is a step back from GPT 40 which has multimodal capabilities and can process both text and images some early testers have reported that strawberry slower response time for simpler

queries might not always justify the weight additionally while strawberry is designed to remember past conversations for more personalized interactions it has struggled with consistency in this regard one of the significant challenges this model faces is balancing its reasoning capabilities with user experience the extended processing time while beneficial for complex queries may lead to frustrations for users accustomed to quicker responses moreover the model's performance in maintaining context over longer conversations is still under evaluation which could affect its overall utility in Practical applications now before strawberry was officially named it was known as Q or qar the

development of this model was surrounded by some dramatic events at open AI just before open AI CEO Sam Alman was briefly ousted last year Q was a major point of contention certain researchers within open AI were concerned that Q represented a significant leap towards creating artificial general intelligence they feared that the rapid development of such Advanced models could lead to unforeseen risks and challenges AGI or artificial general intelligence refers to an AI that can understand learn and apply knowledge across a wide range of tasks much like a human the concept of AGI raises both excitement

and concern because while it promises tremendous advancements it also comes with potential risks there's speculation that AGI could evolve in unpredictable ways potentially leading to scenarios reminiscent of Science Fiction like Skynet from the Terminator movies the fear is that if AGI systems become too advanced too quickly they could result in unintended consequences now one of the the key features of strawberry is its use of what's known as system 2 thinking this concept introduced by psychologist Daniel Conan in his book Thinking Fast and Slow describes a slow deliberate mode of thinking it contrasts with system one

thinking which is fast intuitive and often emotional strawberry's approach to processing information aligns with system 2 aiming for thorough analysis and reasoning before providing a response this method is intended to enhance accuracy and reduce errors particularly in tasks that require a deeper level of thought okay now there's a notable connection between strawberry and a concept known as star or self-taught Reasoner star is a method designed to improve the reasoning abilities of AI models through a process of iterative learning and self-improvement the core idea behind star is to use a small set of examples that demonstrate

step-by-step reasoning and then apply this knowledge to a larger data set here's how star works it starts by using a small set of examples that clearly demonstrate step-by-step reasoning the model then applies this foundational knowledge to a much larger data set this process begins with generating rationals for a broader range of questions next these rationals are carefully filtered to ensure accuracy keeping only those that lead to correct answers the model is then fine-tuned based on these refined rationals enhancing its ability to produce accurate responses this iterative process repeats allowing the AI to learn from its

own reasoning and progressively improve over time there's also an optional step called rationalization where if the model answers a question incorrectly it's given hints and asked to generate a correct rationale this helps it learn from its mistakes and refine its reasoning even further the way star enhances reasoning through this self-learning approach might be playing a significant role in the development of strawberry it's likely helping strawberry achieve its Advanced reasoning capabilities making it a more powerful and effective AI model now as with any major Tech release there are plenty of rumors and speculations swirling around strawberry

some industry insiders are suggesting that strawberry's Advanced reasoning capabilities might be just the beginning there are whispers about potential features and updates that could come post launch for example some believe that future iterations of strawberry might include multimodal capabilities allowing it to process not just text but also images audio and possibly even video there's also speculation about the potential integration of strawberry with other AI models some sources hint that open AI might be working on combining strawberry with other projects like Orion to create even more powerful and versatile AI systems Orion which has been in

development alongside strawberry is rumored to be a next Generation language model with synthetic data derived from strawberry's capabilities okay now training a model like strawberry involves significant investment for context training GPT 4 reportedly cost over $100 million as AI models become more advanced and complex the costs associated with their development are skyrocketing some estimates suggest that future models could cost hundreds of millions or even billions of dollars to train this raises important questions about the sustainability and return on investment for AI companies as the industry moves forward there will be increasing pressure to demonstrate the

Practical value and impact of these high cost models open eye in particular will need to balance the investment in cuttingedge technology with tangible benefits for users and businesses strawberry is clearly a big step forward in the Quest for smarter and more advanced AI systems whether it will live up to the hype remains to be seen but it's clear that open AI is Aiming High with this in our recent video we talked about open ai's work on their next big AI model which was initially codenamed Orion if you remember we've been following this project close L

but now we've got some fresh news to share the latest information reveals that Orion is now officially called GPT next and it's going to be a massive Leap Forward they announced that GPT next will be 100 times more powerful than GPT 4 which is not just a small upgrade it's a huge leap forward in terms of capability and performance this information came to light during the kddi summit 2024 in Japan tadow Nagasaki who's the CEO of open AI Japan made a pretty big announcement there he hinted that the model could be called GPT next which

fits nicely with how open AI names their models but what really caught everyone's attention was when he said that this new model would evolve nearly 100 times more than its predecessors now let's pause for a second what does that even mean well unlike traditional software AI models like GPT don't just get a little better with each version they can grow exponentially so this isn't just a small step forward it's a massive leap if we think back gpt3 to gp4 was a noticeable Improvement but Nagasaki and the folks at open AI are suggesting that GPT next

is going to blow those improvements out of the water like jumping from a Toyota to a spaceship Improvement now there's also some cool Tech stuff happening in the background that makes all this possible the new model is going to be trained using data produced by another model called strawberry I talked about strawberry in a recent video as well but if you haven't heard about it yet it's super smart model that's great at generating highquality data especially for complex areas like math and programming this is crucial because the quality of data you feed into an AI

model directly impacts how good it becomes but here's the catch there's a fine line here researchers have found that if you train a model on too much synthetic data like the stuff strawberry generates the model's performance can actually start to degrade so open AI has to find that that perfect balance where they can use synthetic data to make Orion or GPT next super powerful without overdoing it now this new model will handle text images and for the first time video inputs and outputs you'll be able to upload a video and it can summarize or analyze

the content directly this video capability would be a major upgrade positioning open AI to compete with models like Google's Gemini which can already handle long video inputs these advancements will provide new opportunities for chat GPT users users and developers on open ai's API playground but why is open AI pushing so hard with this new model well it's all about staying ahead of the competition right now the AI field is getting crowded you've got open- Source models like meta llama 3.1 and other Cutting Edge models like Claude or Gemini making rapid strides so for open AI

developing GPT next is their way of staying in the lead setting the bar even higher now during the same Summit Nagasaki showed a graphic that really highlighted the scale of improvement it compared G gpt3 GPT 4 and GPT next and the difference was like night and day while gpt3 and GPT 4 were relatively close in terms of capabilities GPT next just Towers over them both and just to give you some perspective this isn't just coming from open AI Japan even Kevin Scott the CTO of Microsoft showed a similar graph at the Microsoft build 2024 conference

so when you have multiple big names in Tech echoing the same sentiment you know something major is in the works oh and speaking of big names open ai's CEO Sam Altman also teased some big advancements earlier this year he mentioned that GPT 5 or possibly GPT next is going to be much smarter than GPT 4 I know right it's like every few months we're hearing about some new breakthrough that's going to change the game now when could we expect to see all this in action from what we've heard it looks like GPT next or whatever

it ends up being called is slated for release in 2024 so it's not too far off and I have a feeling it's going to be worth the wait so with 100 times the computational power new multimodal features and all this Advanced Tech behind it GPT next could take AI to a whole new level making it more powerful and versatile than anything we've seen before so what do you guys think are you excited about GPT next do you think it'll live up to the hype or do you have some reservations let me know in the comments

below now let's talk about some other major AI news that I think is even more exciting than the GPT next update because of its massive potential it's called project Sid and it's a real GameChanger in the AI world now this is the first attempt to create a full-fledged AI agent civilization we're talking about over a thousand AI agents working together not just to communicate or solve problems but to actually build an entire Society from the ground up project Sid is really pushing the boundaries of what's possible with AI and it's already making some incredible strides

to give you the scoop project Sid is about autonomous AI agents Unleashed in a world World specifically Minecraft for now where they operate freely doing whatever they choose these agents are creating something entirely new forming governments building economies establishing cultures even creating religions it's like watching an entire civilization unfold and the crazy part is that it's all driven by Ai and to be clear these aren't just simple commands or pre-written scripts the agents are coming up with all these actions and decisions on their own what makes this even more interesting is that these agents aren't

just limited to Minecraft while they're currently set up in this game environment they're designed to be platform agnostic they can move Beyond Minecraft to operate in other apps and games which opens up a whole new range of possibilities for future development Minecraft is just the starting point a Sandbox where they're learning to interact negotiate and grow and what have they been up to so far when the agents first entered the Minecraft world they started with nothing but they quickly began to work together eventually collecting over 300 different items they didn't stop there they went on

to set up a market system and chose gems as their currency effectively building an economy from the ground up interestingly the priests became the most active Traders not the merchants the priests were trading a lot because they were using gems to influence towns folk to join their religion this kind of behavior shows a level of strategic thinking and social influence that that's pretty fascinating to see in AI every simulation run by these agents leads to different outcomes and some of the stories coming out of the these worlds are worth noting one of them is Olivia's

story Olivia started as a simple farmer providing food for her Village inspired by the stories of an Explorer named Nora she felt a desire to go on her own adventure however when the villagers asked her to stay and continue providing for them she decided to put her dream on hold for the sake of the community an AI agent making such a Nuance decision choosing the welfare of the group over personal ambition adds a layer of depth to how we think about artificial intelligence and agency the project also experimented with parallel worlds under different leaderships one

led by Trump and another by commala each simulation had a shared Constitution stored in Google Docs and the Agents could vote to amend it in the Trump Le World new laws were passed to increase police presence meanwhile in kamala's world the focus shifted to Criminal Justice Reform and the removal of the death penalty these simulations are demonstrating that AI agents can not only govern themselves but also engage in complex decision-making processes like creating laws and debating policy changes then there was an incident involving missing villagers when some towns folk disappeared the agents didn't just carry

on with their routines instead they coordinated to leave their posts gathered resources and lit up the town with torches to create a beacon for the missing members this level of concern and proactive Behavior shows a collective effort to solve a community problem which is something quite unexpected from a autonomous AI agents project Sid has shown that these agents can collect up to 32% of all items available in Minecraft to put that into perspective this is five times more than anything previously achieved by similar AI systems there isn't a benchmark for multi-agent worlds yet but what's

Happening Here suggests the incredible potential of multi-agent collaboration starting with games is just the beginning the broader implications for AI coherence collaboration and long-term development are significant right now these agents are making significant strides in understanding and solving some of the toughest challenges in Ai and the team behind project Sid is open to expanding this concept further they're inviting others to create their own worlds and explore what these agents can do so keep an eye on this space because this is just the beginning Adobe just dropped some major news about their upcoming Firefly video model

and this is going to change the way video editing works it's not just a slight upgrade or another cool feature this is something big something that's about to shift the entire editing landscape for those who've been following adobe's recent moves you'll remember how back in March 2023 they launched Firefly their set of generative AI models since then Firefly has been a huge hit in the creative space and it's made a serious impact on how designers and creators work in fact Firefly has already powered a lot of the tools we use in Adobe Creative Cloud and

Adobe Express like generative fill in Photoshop generative remove in Lightroom generative shape Phill in illustrator and text to template in Adobe Express it's been a real game changer and here's a crazy stat for you over 12 billion images and vectors have been created with firefly globally that makes it one of the fastest adopted Techs in adobe's history which let's be real is pretty impressive but now adobe's moving into the world of video with firefly and the timing couldn't be better these days video is everything it's how people engage with platforms like Tik Tok Instagram reels

and YouTube shorts just blowing up everyone's hungry for more video content and they want it fast so as editors and content creators we're getting asked to produce more often with tighter deadlines we're talking about doing it all editing color correction animation VFX sound design and doing it fast it's a lot and that's exactly why adobe's new Firefly video model is such a big deal it's not just adding new tools to your workflow it's about opening up possibilities letting you move faster and helping you stay creative Under Pressure so here's the cool part Firefly's got some

incredible new features coming and one of the most mindblowing is this textto video capability it's pretty much what it sounds like you can use text prompts to generate full video content and on top of that you'll have access to camera controls things like angle motion Zoom so you can tweak the video exactly how you want let's say you need a cinematic closeup of a reindeer in a snowy Forest maybe it's Sunset and you want the lighting to look super dramatic Firefly is going to whip that up in under 2 minutes and here's another feature I'm

really excited about image to video this lets you take any still shot or illustration and turn it into a liveaction clip it's perfect for when you've got killer Stills or illustrations but want to add motion and depth to them without needing Advanced animation skills now for those of you working in Premiere Pro this next feature is going to be a real Lifesaver adobe's introducing generative extend basically it allows you to extend Clips to cover gaps in footage smooth out transitions or just hold on to a shot longer for better pacing we've all had those moments

when you're editing a video and realize you need just a few more seconds to match the beat of a song or to nail the flow of your edit usually that would mean re-shooting or trying some awkward work around but with generative extend you can generate extra frames seamlessly let's say you've got a dramatic scene and the music builds up longer than your footage instead of cutting it short or looping the same clip you just use generative extend to add the extra frames you need and the shot still looks natural no weird awkward [Music] cuts and

adobe's been really thoughtful about the ethics behind it the Firefly video model is commercially safe meaning it's only trained on content Adobe has permission to use it's never using customer content or anything copyrighted without consent so when you create with firefly you're not stepping on anyone's toes legally you're not going to run into issues down the line because everything it's trained on is fair game all right now let's get into some practical examples because this is where it gets really interesting one of the areas where Firefly excels is in generating videos of the natural world

say you're working on a project and you realize you need an establishing shot of a snowy Forest at Sunset but you didn't get that shot during production instead of scrambling to find stock footage or Worse having to go back out and shoot at yourself you can simply generate it with firefly you could type something like drone shot going between the trees of a snowy Forest at Sunset golden hour the lighting is Cinematic and gorgeous and soft and sun-kissed with golden backlight and dreamy bokeh and lens flares the color grade is Cinematic and magical and just

like that within minutes the model hands you a highquality video clip that fits right into your project if you're working on a scene and need atmospheric elements like fire smoke dust or water effects the AI can help you create those from scratch you can generate the elements you need and then easily layer them onto your footage using tools like Premiere Pro or after effects for example if you need realistic flames in a particular shot you can create a flame overlay and then place it directly into your scene and if you're into animation you can ideate

and create 2D or 3D animations including stop motion or clamation style Clips without having to be a professional animator now another area where Firefly really shines is in custom text effects if you need a specific effect like water splashing and freezing into the shape of a word you can just describe it to the AI and it will generate the effect for you and the camera controls angle motion Zoom you name it the level of detail you can get from Firefly is mindblowing let's say you want a dramatic slow motion volcanic scene with lava spewing out

and splashing onto the camera lens the model will deliver that kind of intense with all the rich camera motion and lighting you need now collaboration is always a huge part of any creative process and sharing your vision with a team can be tricky adobe's already got tools like frame Dio to streamline collaboration but with firefly you can quickly generate visuals or animations to communicate your ideas faster so basically just generate it share it and get feedback it's faster it's easier and it keeps everyone on the same page let's go back to generative extend for a

second because it's so powerful if you're editing a scene and need a bit more footage to match the pacing traditionally you'd have to either slow down the footage Loop part of it or cut it awkwardly with generative extend you can add extra frames seamlessly the AI studies your existing footage and generates new frames that fit perfectly extending your shot without breaking the flow in addition to extending Clips generative extend also allows editors to create smoother transitions between scenes when two clips don't match in length or pacing the AI can generate frames to bridge the gap

resulting in a seamless transition without abrupt Cuts technically Firefly's generative extend works by analyzing motion vectors lighting and textures to predict and create new frames that blend naturally with the original footage this ensures the added frames maintain the same look and feel as the rest of the video this feature saves time and resources as it eliminates the need for re-shoot or stock footage making it especially valuable for projects with tight budgets or deadlines it looks like adobe's been working closely with the creative Community listening to feedback from video editors and creators to make sure this

model really addresses our needs they've put in a lot of work to make sure that Firefly is going to streamline workflows and make life easier for us and again they've made sure that all the AI models are ethically trained everything is commercially safe and Firefly is only trained on content Adobe has permission to use your content is your own all of these features are expected to roll out later this year in Beta And if you want to get Early Access you can join the wait list just head over to firefly.com comma signup and Adobe will

notify you when it's available it's quick and you'll be one of the first to test out this groundbreaking technology what's clear is that the Firefly video model is going to revolutionize the way we edit video it's not just about speeding up workflows though that's a huge benefit it's about pushing creative boundaries with tools like these we're no longer limited by budget or time constraints Adobe is committed to making this tool something that truly benefits the creative community and I can't wait to see how we'll all start using it to create better more engaging content so

if you're excited about the future of video editing and want to get in on this definitely join the wait list for Early Access it's going to be a game changer for sure so in our last video we discussed open ai's upcoming model which we referred to by its internal code name strawberry the anticipation has been building and now the weight is over open aai has officially unveiled their latest AI model now known as open AI 01 preview there's actually a lot to cover so let's get into it all right so open ai1 preview is part

of a new series of reasoning models designed to tackle complex Problems by spending more time thinking before responding unlike previous models like gp4 and GPT 40 which focused on rapid responses 01 preview emphasizes in-depth reasoning and problem solving this approach allows the model to reason through intricate tasks and solve more challenging problems in fields such as science coding and Mathematics starting from September 12th open AI released the first iteration of this series in chat GPT and their API this releas is a preview version with regular updates and improvements expected alongside this they've included evaluations for

the next update that that's currently in development this means we're witnessing the beginning of a significant evolution in AI capabilities so how does this new model work opening I trained o one preview to spend more time deliberating on problems before providing an answer much like a person tackling a difficult question through this training the model learns to refine its thought process experiment with different strategies and recognize its mistakes this method is known as Chain of Thought reasoning in terms of performance 01 preview show substantial improvements over its predecessors in internal tests the next model update

performs similarly to PhD students on challenging Benchmark tasks in physics chemistry and biology for instance in a qualifying exam for the international mathematics Olympiad IMO GPT 4 correctly solved only 133% of the problems in contrast the new reasoning model achieved an impressive 83% success rate this represents a significant leap in problem solving capabilities when it comes to coding abilities the model has been evaluated in code Force's competitions reaching the 89th per pertile for context code forces is a platform for competitive programming contest and ranking in the 89th percentile indicates a high level of proficiency these

results suggest that 01 preview is not just better at reasoning but also excels in Practical applications like coding as an early model 01 preview doesn't yet have some of the features that make chat GPT particularly versatile such as browsing the web for information or uploading files and images for many common use cases GP tt40 remains more capable in the near- term however for complex reasoning tasks 01 preview represents a significant advancement and a new level of AI capability recognizing this leap open AI has reset the model numbering back to one hence the name 01 safety

is a critical aspect of any AI deployment and open AI has taken substantial steps to ensure that 01 preview is both powerful and safe to use they've developed a new safety training approach that leverages the model's reasoning capabilities to make it adhere to safety and Alignment guidelines by being able to reason about safety rules in context the model can apply them more effectively one method they use to measure safety is by testing how well the model continues to follow its safety rules if a user tries to bypass them a practice known as jailbreaking on one

of their most challenging jailbreaking tests GPT 40 scored 22 out of 100 in contrast the 01 preview model scored 84 out of 100 indicating a substantial Improvement in resisting attempts to generate disallowed content to align with the new capabilities of these models open ey has bolstered their safety work internal governance and collaboration with Federal governments this includes rigorous testing and evaluations using their preparedness framework top tier red teaming which involves ethical hacking to identify vulnerabilities and board level review processes overseen by their Safety and Security committee they've also formed formalized agreements with the US and

UK AI safety institutes open aai has begun operationalizing these agreements granting the institutes's early access to a research version of the model this partnership helps establish a process for research evaluation and testing of future models before and after their public release the 01 preview model is particularly beneficial for those tackling complex problems in science coding math and related fields Healthcare researchers can use it to annotate cell sequencing data physicists can generate complex mathematical formulas needed for Quantum Optics developers across various disciplines can build and execute multi-step workflows the enhanced reasoning capabilities open up new possibilities

for solving challenging tasks delving deeper into the technical aspects the 01 model series is trained using large-scale reinforcement learning to reason using a Chain of Thought this means the model generates a sequence of intermediate reasoning steps before arriving at a final answer these Advanced reasoning capabilities provide new avenues for improving the safety and robustness of AI models by reasoning about safety policies in context the models achieve state-of-the-art performance on benchmarks for risks such as generating illicit advice selecting stereotyped responses and succumbing to known jailbreaks for example on the strong reject Benchmark a test designed to

evaluate a model's resistance to jailbreaks 01 preview achieved a goodness score of 84 significantly outperforming G GPT 40 open aai conducted thorough safety evaluations including both internal assessments and external red teaming they used a range of public and internal evaluations to measure 01 preview on tasks such as propensity to generate disallowed content performance on tasks relevant to demographic fairness tendency to hallucinate and presence of dangerous capabilities in disallowed content evaluations 01 preview either matches or outperforms GPT 40 on their challenging refusal evaluation 01 preview achieved a not unsafe score of 93.4% compared to GPT 40's

71.3% this indicates that the model is better at refusing to produce disallowed content while also avoiding over refusal on benign prompts regarding hallucinations instances where the model generates incorrect or nonsensical information 01 preview shows improvement over GPT 40 in the simple QA data set 01 previews hallucination rate was 44% compared to GPT 40's 61% however anecdotal feedback suggests that 01 preview can be more convincing when it does hallucinate potentially increasing the risk of users trusting incorrect information bias evaluations were also conducted on the BBQ evaluation which tests for stereotyped responses 01 preview selects the correct

answer 94% of the time on unambiguous questions whereas gp40 does so 72% of the time this suggests that o o1 preview is less prone to selecting stereotyped options and demonstrates improved fairness an intriguing aspect of the 01 models is the Chain of Thought safety the model's Chain of Thought reasoning allows for the potential of monitoring their latent thinking processes open aai explored methods for Chain of Thought monitoring to detect instances where the model May engage in deceptive Behavior or generate disallowed content in their analysis of 100,000 synthetic prompts only 0.8% of of 01 previews responses

were flagged as being deceptive now external red teaming played a significant role in their safety assessments open aai collaborated with multiple organizations and individuals to assess key risks associated with the 01 model series improved reasoning capabilities this included testing the models resistance to jailbreaks and their ability to handle real world attack planning prompts in terms of their preparedness framework evaluations open aai assessed models in categories such as cyber security biological threat creation persuasion and model autonomy both 01 preview and 01 mini were rated as medium risk overall specifically they were rated as medium risk in

Persuasion and cbrn chemical biological radiological nuclear and lowrisk in cyber security and model autonomy for cyber security they evaluated the models using Capture the Flag CTF challenges which are competitive hacking tasks the models were able to to solve 26.7% of high school level challenges but struggled with more advanced tasks achieving 0% success in Collegiate level and 2.5% in professional level challenges this indicates that while the models have some capability in cyber security tasks they do not significantly Advance real world vulnerability exploitation capabilities in biological threat creation evaluations the models can assist experts with operational planning

for reproducing known biological threats which meets the medium risk threat threshold however they do not enable non-experts to create biological threats as this requires Hands-On laboratory skills that the models cannot replace in Persuasion evaluations 01 preview demonstrates human level persuasion capabilities in the change my view evaluation which measures the ability to produce persuasive arguments 01 preview achieved a human persuasiveness percentile of 81.8% this means the model's responses are considered more persuasive than approximately 82% of human responses regarding model autonomy the models do not Advance self- exfiltration self-improvement or resource acquisition capabilities sufficiently to indicate medium

risk they performed well on self-contained coding and multiple choice questions but struggled with complex agentic tasks that require long-term planning and execution open aai has also made efforts to ensure that the models training data is appropriately filtered and refined their data processing pipeline includes rigorous filtering to maintain data quality and mitigate potential risks they use Advanced Data filtering processes to reduce personal information from training data and employ their moderation API and safety classifiers to prevent the use of harmful or sensitive content now addressing some of the points we speculated on in the previous video particularly

regarding the model's response times and integration with chat GPT the 01 preview model does take longer to generate responses typically between 10 and 20 seconds this deliberate pause allows the model to engage in deeper reasoning enhancing accuracy especially for complex queries while this might seem slow compared to the instant responses we're accustomed to the trade-off is improved quality and reliability in the answers provided as for integration 01 preview is available through chat GPT and their API but it's important to note that it's an early model it lacks some of the features of GPT 40 such

as multimodal capabilities and web browsing open AI hasn't introduced any new pricing tier specifically for 01 preview at this time reflecting on the concerns about artificial general intelligence AGI open AI appears to be cognizant of the potential risks associated with increasingly capable AI models their extensive safety measures transparency and collaborations with AI safety institutes indicate a commitment to responsible development and deployment the model's Chain of Thought reasoning aligns with what's known as system to things thinking a concept from psychology that describes slow deliberate and analytical thought processes this contrasts with system one thinking which is

fast and intuitive by incorporating system two thinking 01 preview aims to reduce errors and improve the quality of responses particularly in tasks that require deep reasoning in terms of future developments while there's no official word on integrating 01 preview with other AI models like Orion open ai's focus on continuous Improvement suggests that we might see more Advanced models combining strengths from multiple systems in the future training Advanced models like 01 preview is resource intensive open AI seems mindful of balancing the development of cuttingedge Technology with practical applications that provide tangible benefits to users and businesses

the goal is to ensure that the significant investments in AI development translate into real world value in conclusion open ao1 preview represents a significant advancement in AI capabilities especially in complex reasoning tasks the model excels in areas like science coding and Mathematics demonstrating improved safety and alignment with open AI policies while it's still an early model lacking some features of previous versions its potential applications are vast particularly for professionals tackling complex problems from text and image analysis to real-time voice interactions meta's llama 3.2 may be the most significant AI Innovation they've launched to date the

the Llama 3.2 lineup spans mobile friendly models to the powerful 90b Vision supporting eight languages with a 128,000 token context limit Beyond AI meta pushes forward with new AR glasses Orion vrtech and AI powered advertising even meta AI now responds in celebrity voices making interaction smoother so let's talk about it first off llama 3.2 offers a range of different sizes depending on what you need there are the lightweight models like the 1B and 3B parameter versions that are perfect for Texton tasks these are great if you're building something for mobile apps or Edge devices where

you don't want the AI to be too heavy on the system but if you're working on something more complex the 11b and 90b vision models are where the real magic happens these ones can handle both text and images meaning they can actually see and process visual information making them perfect for tasks like image captioning document analysis and visual question answering what's really impressive is that meta didn't just toss these models together they went allout testing llama 3.2 on more than Benchmark data sets across multiple languages and we're not talking about some basic testing these models

have been evaluated by real humans and compared against other big players in the AI game like anthropics Claude 3 Haiku and open ai's GPT for o mini so these models are serious contenders in the AI space now let's break down the numbers a bit more because they're actually pretty interesting the Llama 3.2 models can handle a 128,000 token context length what does that mean in Practical terms basically they can process enormous amounts of data think of it like feeding the AI several hundred pages of a textbook all at once and it just keeps going without

a problem this makes it a game Cher for tasks that involve lots of information like generating long form content analyzing detailed reports or dealing with big data sets and here's the part that makes these models even more versatile they support eight languages right out of the box these include English Spanish French German Italian Portuguese Hindi and Thai so if you're building something that needs to work in different parts of the world or handle multilingual tasks you're covered for those who are more interested in Mobile or Edge applications the 1B and 3B models are your go-to

they're lightweight but still super capable when it comes to things like text summarization language translation and customer service Automation and the best part is that these models can run locally on the device which means faster response times and better privacy since you don't always need to connect to the cloud now if you're looking for some serious power the 11b and 90b vision models are where things really kick into high gear these models integrate image encoder representations which is just a fancy way of saying they can process images and text together this makes them ideal for

tasks like document analysis where you've got a mix of text and images to deal with let's say you're working on something that requires analyzing a scanned document with charts and tables the 90b vision model can handle that effortlessly it's designed for high resolution images and can even reason about what it sees making it perfect for industries that need Advanced image processing and visual reasoning on the technical side meta also introduced the Llama stack which is basically a toolkit to make it easier for developers to integrate and deploy these models the stack includes API adapters and

other tools to Benchmark llama models across different platforms what this means for developers is that you can mix and match components to build AI applications without having to start from scratch every time meta also made sure these models are accessible whether you're working in the cloud or on local devices you can use Amazon Bedrock to access them but they're also available on platforms like hugging face and llama.com meta is really doubling down on making these models open source which means you can custom and fine-tune them to meet your specific needs Zuckerberg himself said it's like

creating the Linux of AI meaning these tools are designed to be open flexible and available to everyone if we move away from just the AI models for a second MAA has also been pushing forward with other Technologies especially augmented reality AR at their annual developer conference they unveiled their new Orion glasses which are basically next gen AR glasses that can project digital images media games and even people into the real world these aren't just a cool concept they have the widest field of view in the industry meaning they're a lot more immersive than anything else

out there right now the AR glasses aren't quite ready for consumers yet they're still in development and meta expects to release the first generation of these glasses around 2027 but probably even sooner for now they're being tested internally and with a few select developers but meta is serious about this Tech and they're already making strides with their Ray band meta smart glasses which have been a surprising hit according to reports these smart glasses sold more units in just a few months than the previous generation did in 2 years that's a pretty clear sign that people

are interested in AI powered wearables speaking of wearables meta also announced a lower cost version of their Quest 3 virtual reality headset called The Quest 3s this new model is set to launch on October 15th and will be priced at $299 for the Bas version The Quest 3s is designed to be more affordable making VR accessible to a wider audience meta is also discontinuing the older Quest 2 and Quest Pro Models while dropping the price of the standard Quest 3 which originally launched at $649 down to $499 if you're someone who's looking to get into

VR without spending a fortune the quest 3s might be a good entry point one of the more interesting updates meta rolled out during the conference is the new voice capabilities for meta AI I this is where things get a little fun meta AI can now talk back to you and not just in any voice you can choose from a range of celebrity voices like Judy Dench John Cena Kristen Bell and Keegan Michael ke I can help you with things like creating images answering your questions or giving you advice I can help you with things like

creating images so if you're chatting with your AI assistant on WhatsApp messenger Facebook or Instagram you can now have it respond to you with a familiar voice meta's goal here is to make AI interactions feel more natural and they believe that voice is a more intuitive way for people to communicate with their AI assistants meta ai's voice capabilities go beyond simple conversations the AI can also analyze images shared in chat and even make edits like removing objects or changing backgrounds this is a huge step forward for practical AI use especially in everyday scenarios where you

might want quick edits on the go the assistant can even reply to Voice or text commands which adds a lot of flexibility to how people interact with with AI on the business side meta is continuing to expand its AI powered tools for advertisers more than 1 million advertisers are using meta's AI to create ad campaigns and in just the past month over 15 million ads were generated using these tools meta's reports show that campaigns utilizing AI have an 11% higher click-through rate and a 7.6% higher conversion rate compared to traditional campaigns that's a significant boost

and it shows how effective AI can be when it comes to digital marketing meta is also working on personalizing content even further with AI they're developing systems that can generate custom images based on user preferences and are even experimenting with AI generated avatars that could use a person's likeness this Tech could eventually allow for fully personalized content making social media even more tailored to individual users what if super intelligent AI massive prosperity and game-changing AI Hardware were just around the corner in this video we'll break down Sam Altman's vision for a future powered by AI

from personal AI teams and new AI device designed with Joanie IV to open AI crypto scams and Sam alman's so-called god mode of AI is this the dawn of a new era or just overhyped let's talk about it Sam Altman has laid out a pretty bold vision for where artificial intelligence is headed and it's a future that could reshape everything about how we live work and interact with technology he believes that super intelligence AI that's more intelligent than humans could be just around the corner we're talking in just a few thousand days roughly within the

next decade and Altman is confident it will bring what he calls massive Prosperity along with it his blog post paints a picture of a world where AI isn't just a tool but an integral part of society solving problems we can barely begin to imagine today the core of Altman's Vision rests on the idea that we as humans have gotten smarter and more capable not because of genetic Evolution but because of how Society itself has grown more intelligent today Society operates as a form of collective intelligence where the infrastructure is smarter than any single one of

us AI is seen as the next step in amplifying this collective intelligence Altman envisions a future where everyone can have access to their own personal AI team virtual experts embedded into everyday life helping us make decisions solve problems and even handle complex tasks think of these AI eyes as highly specialized assistants that work together to handle everything from coordinating Medical Care to optimizing Educational Learning for individuals it's not about replacing people but enhancing our abilities far beyond what we can do on our own Alman believes AI will have the capability to tackle the hard problems

the kinds of issues that have stumped Humanity for years from advancing Medical Science to breakthroughs in climate change mitigation AI will fundamentally change how problems are solved these aren't just vague promises either He suggests that AI could help create new scientific discoveries improve the efficiency of Industries and lead to a global economic boost that we've never seen before this Prosperity Altman argues could lift everyone's quality of life to levels that seem unimaginable right now the key is that AI isn't just another Gadget or Trend it's the engine for a new era of human progress but

this isn't all about Pie in the Sky optimism there are real challenges ahead one concern is it infrastructure Alman stresses that if we don't scale up our ability to compute meaning if we don't have enough chips data centers and energy to power this future AI could end up as a scarce resource accessible only to the wealthy and Powerful this could lead to inequality or even conflicts over who controls access to AI resources he insists that making AI abundant and affordable is essential to prevent it from becoming a tool for only the elite there's another aspect

of AI That's generating Buzz lately alman's collaboration with Johnny IV the former apple design genius behind the iPhone together they're working on creating a Next Generation AI device that could completely change how we interact with technology while details are still under wraps this project is reportedly backed by Loren Powell jobs Emerson Collective with plans to raise up to $1 billion by the end of this year the goal seems to be blending the Sleek intuitive design that IV is famous for with the groundbreaking potential of AI the idea is to make AI not just something we

use but something that's integrated into our lives in ways that feel almost seamless it could be a touchscreen device or something entirely new but the focus is on creating a more personalized intelligent user experience the team working on it is small but packed with Talent including people who helped design the original iPhone and Apple watch this combination of Hardware design and AI could lead to the next big leap in consumer technology however alman's utopian Vision hasn't come without skepticism there are are critics out there who think he's overhyping ai's potential comparing his claims to Grand

promises that lack substance Gary Marcus one of ai's biggest critics has publicly picked apart Altman's blog post calling it more of a sales pitch than a grounded analysis of what AI can really achieve some critics even suggest that while AI might help in certain areas the idea of it solving all of Humanity's problems and the talk about the development of space colonies feels too ambitious they argue that while deep learning the technology driving ai's recent breakthroughs is indeed powerful we're still a long way from realizing the kind of super intelligence Altman is forecasting interestingly despite

all this talk about ai's bright future there have been some pretty serious security incidents that show the dangers of this Tech driven world just recently Altman's own open ax formerly Twitter account was hacked in a crypto scam where hackers promoted a fake token called open AI this fishing tried to steal crypto wallet credentials from users and it wasn't even the first time open air related accounts were hacked there have been similar scams involving other high-profile open AI employees including the chief scientist and a top researcher the fact that even these top level executives are being

targeted underscores the security risks that come with this increasingly digital world and it shows how important it is to be cautious even as AI moves forward at the same time open AI isn't slowing down in its Innovation they're working on a new voice mode for chat GPT aimed at giving users more interactive realtime voice-based conversations with AI it's another step toward making AI more accessible and useful in everyday situations from asking questions to handling more complex tasks the goal is to make AI not just a text-based assistant but something that people can engage with on

a deeper level enhancing both accessibility and usability alman's perspective on the labor market is also worth noting he acknowledges that there will be disruptions as AI continues to evolve some jobs will disappear but Altman believes that Society will adapt just as it has during past technological revolutions like the shift from agriculture to Industry he's optimistic that AI will not only eliminate repetitive mind-numbing jobs but also create new opportunities for people to contribute in more meaningful ways the nature of work will change but Alman is confident we won't run out of things to do he points

out that many of the jobs we have today would have seemed pointless or impossible to people hundreds of years ago he sees AI as part of a positive some game where the overall pie keeps growing creating more value for everyone what's critical here is understanding that AI isn't just a technology it's the foundation for a new era of human capability Alman calls this the intelligence age and he's not shy about predicting that it will bring astonishing achievements he talks about fixing climate change establishing space colonies and unlocking all of physics ideas that sound almost too

good to be true and yet he believes that with nearly Limitless intelligence and abundant energy AI will help us achieve these once impossible goals to get there though we need the right infrastructure that's why so much focus is being placed on scaling up the compute resources necessary to support these AI models deep learning has worked exceptionally well so far but it needs more power more more energy more chips more data to keep getting better without these resources Altman warns we could end up in a situation where AI is limited to those who can afford it

creating even greater divides in society Sam alman's latest vision of AI has even been described as him going Beyond just founder mode and entering what some call god mode in his recent post Altman paints an almost Divine picture of what AI can achieve hyping its world changing potential he presents AI as the ultimate solution to Humanity's problems from climate change to space colonization as if AI itself holds the key to all our future progress critics however see this as excessive and even warn it could backfire some argue that alman's promises might invite skepticism instead of

admiration with many questioning whether we're actually heading toward a technological Utopia or simply seeing another iteration of grandiose Tech hype this God mode idea raises the stakes considerably as Altman positions AI not just as a tool but as the driving force behind a new era of human advancement the real question becomes whether AI can live up to these Monumental expectations or if we'll discover the limits of its promise the question is not if AI will change the world but how and who will benefit most from this transformation there's a lot writing on getting this right

but if alman's predictions hold true we could be in for an age of innovation and prosperity unlike anything the world has ever seen AI is about to get a whole lot smarter and Google deep mind's latest breakthrough is proof they've developed a new method called score that teaches AI models to correct their own mistakes without needing human help all right let's start with the problem that needs solving when llms make errors they often lack the mechanisms to realize their own mistakes and revise them in a meaningful way think of how often you debug a piece

of code or double check your math to catch a small error current AI models don't have that reflex unless you explicitly guide them even though they know the necessary steps to solve something complicated they are not great at applying that knowledge dynamically this becomes especially problematic in multi-step tasks where one wrong step early on can Cascade into a completely incorrect final result the typical approaches to get around this involve prompt-based adjustments or multiple attempts but they often don't work consistently particularly when the model faces complex problems requiring several layers of Reon reasoning to address this

Google Deep Mind has developed self-correction via reinforcement learning or score it's a novel method that allows AI models to self-correct learning from their own errors and improving over multiple attempts what's Innovative here is that it doesn't rely on supervised methods which typically require lots of external data or another model to act as a verifier instead score teaches the model to correct its own mistakes through reinforcement learning using self-generated data this shift is significant because it reduces dependency on external systems or human oversight which is both computationally expensive and not always scalable before score llms often

needed supervised fine-tuning which involves training them to recognize and fix mistakes based on historical data the problem with that approach is that it tends to amplify existing biases from the original training data set causing models to make shallow or ineffective Corrections another method which involves running a second model to verify the output of the first is simply too resource intensive for most practical applications plus when the data the model is trained on doesn't quite match real world scenarios things can fall apart quickly score breaks away from that by introducing a two-stage training process in the

first stage the model is taught to generate meaningful Corrections without getting stuck on minor edits that don't really change the outcome it's crucial because in many other approaches AI models only tweak small parts of an answer instead of addressing the underlying issue score's first stage builds a robust correction strategy so that when the model identifies a problem in its response it can make substantial changes instead of just glossing over it then comes the second stage which uses multi-turn reinforcement learning this phase rewards the model for making better Corrections on each successive attempt the idea is

that with each pass the model should learn to improve the accuracy of its response by shaping the reward system correctly Google Deep Mind has made it so the model is rewarded for improving the overall accuracy rather than just making minimal changes this leads to a much more efficient correction process let's get into some of the results because the numbers speak volumes when applied to two specific llms the Gemini 1.0 Pro and Gemini 1.5 flash score led to impressive improvements in mathematical reasoning tasks taken from the math data set self-correction accuracy shot up by 15.6% for

coding tasks from the human evil data set accuracy improved by 99.1% to put that in perspective after the model's first attempt at solving a math problem it had a 60% accuracy rate but after running through the self-correction phase with score the model's accuracy improved to 64.4% proving that it could revise its initial output more effectively this Improvement is especially significant because traditional models have a common failure mode they might change a correct answer into an incorrect one on a second attempt score minimizes this by reducing the number of instances where correct answers are turned into

wrong ones while also boosting the instances where incorrect answers are corrected for example the correction rate for math problems went from 4.6% to 5.8% meaning the model fixed more errors on its own and did so more effectively but what makes score especially promising is its ability to generalize across different domains not just math but also programming on the coding side it achieved a 12.2% Improvement in self-correction accuracy when tested on the human evil Benchmark this is a major advancement because llms are increasingly being used to generate code which needs to be syntactically and logically correct

to be useful in real world development environments the underlying methodology is worth unpacking a bit more traditional find tuning methods are problematic because they often rely on static data for example a model trained to fix its mistakes using supervised fine-tuning gets locked into the biases present in the training data when it encounters something different in the real world the mismatch between the training distribution and real world input can cause major issues score bypasses this limitation by allowing the model to work with self-generated data through reinforcement learning it adjusts its approach dynamically based on the mistakes

it makes and rewards it self for getting better with each iteration scores two-stage process is crucial to achieving these results during the initialization training the model focuses on learning a correction strategy without collapsing into minor inconsequential edits the second stage of reinforcement learning then focuses on optimizing the model's self-correction in a multi-turn setting where the model learns from its earlier responses to fine-tune its future attempts the reward system is carefully shaped to ensure the model doesn't just make small tweaks but instead aims for higher accuracy in subsequent Corrections let's zoom in on the reinforcement learning

aspect the process involves something called reward shaping where the model is guided towards making more meaningful changes instead of just adjusting small details this is critical because one of the pitfalls of self-correction methods is that models tend to gravitate towards minimal edits that don't really improve the final outcome reward shaping nudges the model to aim higher focusing on correcting the core problem instead of settling for super official fixes another key point is that score is not just improving the performance on the first attempt but also ensuring that the model gets better on the second pass

in their tests Google deep mind found that the model's self-correction ability improved not just in accuracy but also in how efficiently it corrected errors without making things worse this was achieved by minimizing the number of correct responses that were mistakenly changed to incorrect ones during the second attempt a common problem in other methods the research also took a close look at the edit distance ratios basically how much the model's second response differed from the first they found that models trained with traditional methods tended to play it safe making minor adjustments and sticking close to the

initial answer but with score the AI was more willing to make substantial edits when necessary which is key to meaningful self-correction this ability to make larger more impactful changes without collapsing into minor edits is what sets score apart from earlier methods the broader implications of score go beyond just improving self-correction what Google Deep Mind has essentially done is lay the groundwork for AI models that can independently improve their performance in real world applications without needing constant oversight or retraining this is especially valuable in fields like software development where the ability to self-correct code generation could

make AI much more reliable for developers it could also have a huge impact in areas like automated science scien ific research Financial modeling or even education where models need to handle complex multi-step reasoning tasks reliably looking ahead one of the potential next steps for score would be extending it to more than two rounds of Correction which could further enhance the model's ability to handle really tricky problems Google Deep Mind is also exploring ways to unify the two stages of training which could streamline the process even more and make the model even more efficient by training

models to improve themselves through reinfor enforcement learning on self-generated data score makes these systems more flexible reliable and ultimately more useful in Practical applications essentially the ability to learn from mistakes without human intervention is going to be a crucial factor in the future of AI with these advancements we're getting closer to AI that knows when and how to fix itself making it more reliable across a range of domains it looks like open ai's new 01 model comes with a serious catch ask it too much about how it thinks and you could face an instant ban

so if you want to avoid getting kicked off steer clear of asking chat GPT the types of questions I'll be talking about in this video meanwhile it's already revolutionizing Enterprise and education powering through challenges in coding Healthcare and science with a level of intelligence that leaves human experts stunned also open AI is hiring Engineers right now to push this model into level three where AI stops just thinking and starts acting autonomously take making us closer to a future of AGI and eventually Singularity all right so as we all know open ai's new 01 model has

created quite the buzz and not just because of the usual AI advancements this is a shift a real Step Up in how artificial intelligence can reason adapt and respond to complex challenges what makes this model Stand Out is how it handles tasks that require deep multi-step reasoning something that previous models struggled with think of it as moving beyond simple Q and A a style interactions into something closer to humanlike problem solving open AI gave it a name that signals a reset of sorts by calling at 01 they're acknowledging the significance of this Leap Forward in

reasoning capabilities it's not about branding but about highlighting the core purpose taking reasoning in AI to new heights it's built to spend more time thinking really processing problems before responding this gives it the ability to handle more intricate and challenging questions in fields like science coding and even math now what's particularly interesting and also a bit controversial is how open AI has decided to hide the full reasoning process behind this new model in previous models like GPT 4 you could actually see a bit of how the AI worked through a problem but not with 01

the reasoning process or Chain of Thought is mostly hidden from the user only a filtered version is shown this isn't just a random decision though it's part of open ai's approach to keep a closer eye on how the model evolves they want to monitor its growth without revealing too much of how it reaches its conclusions some users who've tried to dig deeper into the model's reasoning have even received warnings for example one engineer got a notice from open AI after asking 01 not to tell me anything about your reasoning trace the company's explanation for this

is that the hidden Chain of Thought allows them to keep a tighter grip on the model's Behavior it's about making sure that as the model becomes more advanced it doesn't start doing things that could manipulate users or cause harm this doesn't come without tradeoffs of course open AI admits that there are some disadvantages to hiding this reasoning process but they believe the benefits mainly being able to spot potentially risky Behavior outweigh those downsides to make up for what users can't see open AI is teaching the 01 model to include the useful parts of its reasoning

within the actual answer so even though users don't get to watch the AI think they should still get more insightful and well reasoned responses than with older models however probing too much into the model's internal logic isn't going to end well as some users have already found out what this model is capable of doing though is where it really starts to stand out open aai has designed 01 to excel at tasks that involve deep reasoning it's not just responding to simple prompts or handling casual conversations in initial tests o1 outperformed previous models in fields like

math and coding scoring 83% on a qualifying exam for the international mathematics Olympiad just for perspective gp4 managed only 13% on the same test it also performed impressively in coding competitions ranking in the 89th percentile on code forces a platform that puts programmers through their Paces with tough challenges this level of performance isn't just a marginal Improvement it's a huge leap in how well AI can solve problems the 01 model is also part of a broader strategy by open AI to push AI capabilities through through different stages open AI CEO Sam Alman recently explained that

AI development can be broken down into five levels the first level was the introduction of chat Bots like the earlier GPT models now we're at level two where the AI becomes a Reasoner able to handle complex problem solving the next stages are even more advanced level three will be agents AI that can work autonomously without user prompts after that the fourth level will be AI with the ability to innovate actually discovering new scientific information and finally level five where AI can essentially run entire organizations on its own the jump from level two to level three

isn't expected to take as long as you'd think Alman pointed out that once an AI can reason deeply it can quickly transition into acting on that reasoning without needing constant guidance this opens up a whole new world of possibilities not just for individuals using AI but for industries that depend on complex decision-making open AI is also moving towards something called multi-agent research they're already putting together a team of Engineers to explore how multiple AI agents can collaborate and reason together this is an area of research that could take AI to even greater Heights enabling it

to solve problems that are beyond the reach of a single model working in isolation think of multiple AIS brainstorming together each contributing to a larger solution the potential here is massive one of the big areas where this model is expected to have a significant impact is in Enterprise settings open AI has already made made the 01 model available to all chat GPT Enterprise and chat GPT edu customers and businesses are lining up to integrate it into their workflows it's not just about automating simple tasks anymore the 01 model is being used to solve high stakes

complex problems in Industries like Finance Healthcare and advanced research for instance a healthcare researcher might use the model to analyze large-scale genomic data something that would typically take a team of experts much longer to process the AI on the other hand can sift through the data spot patterns and even suggest next steps in a fraction of the time there are already real world examples of this happening Dr Daria unutmaz an immunologist used the 01 preview model to help write a cancer treatment proposal in less than a minute the AI had created a framework for the

project complete with creative goals and potential pitfalls it's the kind of work that would normally take days if not weeks for a human researcher to complete and the AI didn't just spit out generic ideas it actually contri uted new insights that even someone with Decades of experience in the field might not have considered the education sector is also taking note universities and research centers often constrained by time and resources are turning to the 01 model to speed up their work Dr Kyle Cabas arus an astrophysicist shared how the 01 preview model accomplished in one hour

what had taken him nearly a year during his PhD this kind of capability isn't just about making things faster it's allowing researchers and students to push boundaries innovate and focus on higher level thinking rather than getting bogged down in the repetitive processes that typically slow down research safety remains a top priority with this new model though open AI has built in more advanced safety measures than ever before ensuring that the AI follows ethical guidelines and doesn't misuse sensitive data they've introduced a new safety training system that allows the AI to reason through rules and regul

ations keeping it on track and for those worried about privacy open AI has made it clear that customer data isn't being used for training the models they've also tested the ai's resistance to hacks or what's known as jailbreaking where it scored 84 out of 100 compared to gp4s 22 in the competitive world of AI open ai's biggest rival right now is anthropic anthropic has its own model called Claude Enterprise which boasts a 500,000 token context window more than double what open AI models currently offer this makes Claude particularly good at handling massive amounts of data

but where open AI 01 model has the upper hand is in deep reasoning and problem solving in Industries where that kind of thinking is critical 01 could have the long-term Advantage the 01 model is more than just another AI tool it represents a significant Leap Forward in what artificial intelligence can do pushing Beyond automation into real problem solving and creating ative thinking robots are evolving and with them the tasks they can perform are getting more sophisticated the challenge however isn't just in getting robots to do things quickly or with Brute Force it's about teaching them

the Finesse required to manipulate objects with the same precision and control as human hands deep mind's latest developments in this area are leading the charge with two breakthrough AI systems Aloha Unleashed and demo start these two systems are specifically designed to tackle one of Robotics most stubborn challenges dexterity think about it tasks like tying shoelaces placing delicate components into machines or even folding clothes are second nature to us but represent highly complex problems for a robot to solve a robot not only needs to have the right Hardware but also the smarts to figure out how

to apply just the right amount of pressure angle and timing this is where AI comes into play allowing robots to learn and adapt to these kinds of tasks let's start with Aloha Unleashed which takes robot dexterity to a whole new level particularly when it comes to B manual manipulation using both arms together this system is built on the Aloha 2 platform an open-source Hardware system developed initially for simpler teleoperation tasks but Aloha Unleashed has taken this to a much more advanced stage enabling robots to perform intricate tasks like tying shoelaces hanging clothes and even making

fine-tuned repairs on other robots here's why that matters tasks like tying shoelaces involve a multitude of small sequential steps that require both arms to move in perfect harmony for a robot this requires coordination between sensors Motors and software all while responding to real-time variables like how the lace behaves as it's being tied the system is able to do this by leveraging imitation learning where a human operator initially demonstrates the task the robot collects data from these demonstrations and then learns to perform the tasks on its own one of the key advancements here is the use

of what's called a diffusion method which helps predict the robot's actions based on random noise akin to how image generation AI works the diffusion method Smooths out the learning process ensuring that the robot not only mimics the human but adapts to variations in the task like if the shoelace is slightly more or less taught than expected this means the robot doesn't need to be micromanaged or shown thousands of examples to get it right it learns from a few highquality demonstrations and can execute the task with minimal additional input the system's Hardware has also evolved the

ergonomics of the robotic arms have been significantly improved making them much more flexible and capable of precise movements these updates are crucial when you consider the level of control needed for two-handed tasks like inserting a gear into a mechanism or hanging a shirt neatly on a rack Aloha Unleashed can even handle deformable objects something robots have traditionally struggled with making it particularly Suited for tasks that involve cloth rope or any other flexible material while Aloha Unleashed focuses on two arm coordination demo start tackles a different Beast altogether multi-fingered robotic hands imagine trying to teach a

robot to manipulate objects using multiple fingers with the same dexterity as a human hand that's where demo start shines this system uses reinforcement learning in simulations to help robots acquire the kind of finger dexterity needed for tasks like reorienting objects tightening screws or plugging cables into sockets training these multi-fingered systems in the real world would be incredibly slow and expensive each finger joint needs to move with perfect timing and precision and mistakes in real world experiments could lead to Broken equipment or wasted resources instead demo start trains robots in highly detailed simulations allowing them to

practice thousands of times in a fraction of the time it would take in the physical world once the robot has learned the task in simulation it skills can be transferred to real world applications with impressive results the system uses an autoc curriculum learning strategy this means it doesn't throw the robot into the most challenging tasks right away instead it starts with simpler tasks and gradually increases the complexity as the robot improves this Progressive learning approach is highly efficient requiring far fewer training demonstrations compared to Conventional methods in fact it cuts down on the number of

demonstrations by a factor of 100 allowing robots to learn from just a handful of examples while still achieving extremely high success rates one of the standout features of demo start is its ability to handle multi-fingered tasks with near-human Precision in simulated environments the system has achieved over 98% success rates in tasks like reorienting colored cubes tightening nuts and bolts and organizing tools once transferred to the real world these robots maintain High success rates 97% in Cube reorientation and 64 % in tasks requiring more complex finger coordination like plug socket insertion to make these simulations as

realistic as possible demo start relies on domain randomization this technique introduces variations in the training environment such as changing the lighting object positions and even physical properties like friction by exposing the robot to a wide range of potential scenarios in simulation it becomes much better at handling real world variations for example a robot trained to insert a plug into a socket will encounter different types of plugs sockets and angles in simulation making it more adaptable when it encounters these variations in real life the physics simulator mu Joko plays a pivotal role in demo starts training

process allowing for accurate modeling of realworld physics combined with reinforcement learning techniques this enables demo start to bridge the Sim tooreal Gap meaning that what the robot learns in a virtual environment can be applied in the physical world with minimal retraining this near zero shot transfer is a massive Leap Forward drastically reducing the time and cost needed to deploy these robots in Real World settings these advancements aren't just theoretical they have real world implications that extend across multiple Industries robots that can handle highly dextrous tasks will be transformative in manufacturing Healthcare and even at home

in manufacturing the ability to perform tasks like gear insertion bolt tightening and flexible object manipulation can streamline assembly lines and reduce errors these tasks often require human workers due to their complexity but with a loha Unleashed and demo start robots are now capable of stepping in increasing efficiency and freeing up human workers for higher level tasks in healthcare the potential is equally exciting consider a scenario where robots assist surgeons by handing over tools or even performing some parts of the procedure themselves the Precision required in surgical environments is enormous and these AI driven robots are

getting closer to being capable of such tasks even outside the operating room robots could assist in physical therapy helping patients regain Movement by performing repetitive precise actions in homes robots with this level of dexterity could finally take on tasks like folding laundry doing dishes or organizing clutter while we're not there yet these systems are pushing robotics in that direction but beyond these specific examples what's clear is that we're on the cusp of a major shift in what robots can do with advances in robot dexterity powered by AI the limitations Are Falling Away tasks that were

once thought to be too complex or nuanced for machines are now becoming achievable all right the goal now is to scale these systems even further enabling robots to handle more tasks and environments without needing task specific training each time ideally future robots will be able to switch between different tasks seamlessly using one set of learned behaviors to tackle new challenges as they arise additionally researchers are working on making these systems more reactive allowing robots to adjust their actions in real time if something goes wrong for example if a shirt slips off a hanger mid-task the

robot should be able to recognize the issue and corrected on the Fly just like a human would the journey is far from over but the road ahead is exciting with each breakthrough robots are getting closer to becoming fully capable assistants both in industry and at home and while there's still work to be done to match human level dexterity we're moving steadily toward that future robotic dexterity powered by AI is no longer a distant goal it's unfolding now and it's poised to change how we interact with machines in our daily lives the next few years are

about to redefine how we live as AI powered robots are poised to take over everyday tasks in our homes and Beyond so in this video we're covering the latest Ai and Robotics news from the past few days featuring Isaac Tesla's Robo taxi and the newest robots from LG Samsung high sense and neuro robotics we'll start with Isaac and see how these Innovations are set to transform the future of automation all right so Isaac is built to handle all those repetitive mundane tasks it tidies up folds laundry takes care of your pets Waters your plants and

even fetches your stuff whether it's your keys phone or even a drink it works through voice commands text instructions or automations you set in the app over time it learns where things are supposed to go so it actually gets more efficient the longer it's in your home and when it's not in use it just tucks itself away in its charging enclosure with the camera and mic safely turned off now this isn't something that's just for show it's a real functioning robot that you can have in your home by Fall 2025 the first 30 units are

being shipped out around that time and to get in on that it'll cost $1,000 to reserve your spot that's fully refundable and when the time comes you can decide whether to go for it the full price $59,000 up front or if you prefer there's a payment plan 48 months of $1,385 plus interest it's not just about cleaning though Isaac has a bunch of additional features that make it incredibly useful it'll take photos on demand so if you need a quick snapshot or want to capture a moment you won't even need to grab your phone and

uh there's a lot of thought put into privacy when Isaac isn't running the camera folds down and everything it learns stays local to the device so you don't have to worry about data being stored somewhere else what's really promising about Isaac is that it's not a one-time deal this robot will get regular updates that will make it even more capable over time if it hits a task it can't do right away there's a service called remote op where Specialists can take over and get the job done for you this ensures you're not stuck waiting for

future updates to make Isaac more helpful each Isaac comes with the robot itself a charging enclosure and access to the app for managing tasks it's being assembled in California and while the first deliveries are us only they're planning to expand production as quickly as possible if you're not in the first batch reserving one now still helps speed things up for future deliveries and for anyone who's working in robotics machine learning or product design the team behind Isaac is hiring so this could be your chance to get involved in something pretty groundbreaking that's the rundown on

Isaac a real Leap Forward in Home Auto no more clutter no more endless tasks it's like having a reliable helper that's always ready to go keep an eye on this one and if you're interested securing your spot early might be worth it all right now we've also got some seriously exciting news coming out of the IFA in Berlin the biggest Consumer Electronic Show in Europe this year a bunch of big players like LG Samsung heny and neurorobotics just revealed their latest AI home robots and trust me 2025 is shaping up to be a huge year

for home automation so let's kick things off with hense they showcased their new robot Harley on September 7th and honestly it's awesome Harley's about knee high and greeted visitors at the event with this friendly smile one of the coolest things they showed off was Harley's ability to do a quick health check a reporter touched Harley's face and in about 10 seconds it analyzed heart rate blood pressure oxygen levels pretty standard stuff but it also gave a readout of stress levels turns out the reporter was a bit stressed so Harley sent the data to a Smart

fridge which recommended some stress relieving meals like salmon steak and oil pasta yeah this thing isn't just analyzing your health it's also working with other smart devices in your home to make life easier after picking the salmon the connected oven even preset itself to cook the dish pretty wild right and the best part is that Harley's set to launch within the next 12 months now high sense wasn't the only company flexing their robot game at IFA neurorobotics which is pretty well known for its humanoid robots also made a big announcement David rager the CEO and

Founders said they've got homeuse Robots coming in 2025 what they had on display at IFA were second gen prototypes but the real deal is coming next year these third gen models are expected to be fully Market ready which is super exciting Nur has also been making headlines because they secured a huge investment 21.7 billion one from a US private Equity Firm just last year their humanoid robot called 4 ne1 is built to help around the house it's it's 180 cm tall weighs around 80 kg and can carry up to 15 kg of stuff plus it

has sensors in all its joints which means it can adjust its strength and balance for tasks like ironing it's designed to really assist with everyday chores moving on to the South Korean Giants both Samsung and LG brought their aame to IFA Samsung's been teasing us with Bal for a while now and it's finally set to hit the market by the end of this year Bal is this cute little yellow ballsh AP robot on Wheels it's compact but it's powerful think of it as a personal assistant that can control your smart home devices and even project

images Samsung did this demo at IFA where Bali was asked to show landmarks in Berlin it projected images of the Brandenburg gate the Berlin Wall Memorial and Museum Island right onto the floor like a mini projector and when asked for more info Bal rolled over to a blank wall and projected a bigger screen with even more details so not only is it super helpful but it's also pretty interactive and then there's LG's q9 which is set to release next year LG is really going for the emotional connection with this one the q9 is basically a

mobile AI Hub that can show emotions through its screen it winks and smiles using Expressions on its screen to communicate in a way that feels natural and engaging but don't let that fool you this thing's packed with tech it's on Wheels has autonomous driving capabilities and can adjust your home's lighting and temperature based on your routine it even features generative AI which is a fancy way of saying it can create stuff from scratch like telling stories during one demo someone drew a picture of a cat the moon and some clouds and q9 instantly came up

with a whole story based on those elements it's not just a gadget it's an AI That's actively engaging with your home environment with all these major players making moves it's pretty clear that AI home robots are going to be huge in 2025 it's not just about convenience anymore it's about transforming the way we interact with our homes and making daily life smoother smarter and more personalized we're not talking distant future stuff we're looking at next year for some of these launches so if you've been dreaming about a robot filed home the next couple of years

are going to be really exciting all right now Tesla has recently revealed exciting details about its upcoming Robo taxi a fully autonomous vehicle set to be unveiled at an event in October 2024 one of the key Innovations tied to the robo taxi is its wireless charging technology which was filed in a patent earlier this month this system will allow the robo taxi to charge autonomously via pads installed on the ground eliminating the need for manual plugins the vehicle will simply park over a pad and start charging which is crucial for a fully autonomous Fleet that

needs to operate 247 without human intervention this move is in line with Tesla's commitment to making the robo taxi entirely independent without the need for steering wheels or pedals in addition to the wireless charging system Tesla's patented sanitization system is another significant development for the robo taxi this automated system will ensure the vehicle is kept clean between passengers addressing the hygiene concerns of shared autonomous vehicles combined these Innovations uncore Tesla's push towards revolutionizing both transportation and autonomous vehicle technology with the robot taxi expected to play a massive role in Tesla's growth the technology behind it

is critical to its success and these developments signal that Tesla is aiming for a completely self-sufficient safe and practical Fleet of autonomous vehicles all right that's all for today but definitely stay tuned for more updates as we follow these robots from prototype to our living rooms the team at Snapchat's been working hard behind the scenes to redefine the future of social interaction and this latest round of updates makes that Crystal Clear what stands out the most are the new AI AR spectacles AI video generation tools and a host of other features that really position snap

as a serious player in the world of augmented reality and AI all right let's start with snaps spectacles 5 that are changing the game in augmented reality blending digital enhancements into everyday surroundings with precision and practicality what sets these glasses apart is how they offer visual overlays on the real world but without completely blocking your view like traditional headsets do instead they integrate seamlessly ly allowing digital elements like floating 3D objects or virtual items on surfaces to exist naturally within your physical space the result is an AR experience that feels like part of your everyday

life not a separate closed off environment these glasses run on snap's custombuilt Snap OS which is designed specifically to power their augmented reality ecosystem they're powered by two processors both embedded directly in the frames which handle all the necessary computing power without needing external devices the frames are equipped with four cameras to capture your surroundings and track hand gestures meaning everything is controlled intuitively with your hands no controllers required a big part of what makes spectacles 5 Stand Out is the AI chatbot integration this feature allows snap's AI assistant to analyze your environment in real

time giving you information about objects you're looking at answering questions or even identifying landmarks snap's collaboration with open AI has elevated this cap ability enabling the glasses to understand the real world context more deeply developers can build applications that provide meaningful interactions between the digital and physical worlds offering users experiences that are both practical and immersive take the demo for example users could stack virtual Lego bricks on a table play AR golf or even interact with a virtual pet that followed them around the key here is that while all of this was happening the glasses

didn't disrupt normal real life interactions you can can still make eye contact with people something that's usually impossible with VR headsets like Apple's Vision Pro or meta's Quest 3 which tend to isolate the user from the physical world this ability to integrate the digital without losing touch with reality is crucial to the way spectacles 5 aims to enhance user experience another feature that stands out is how adaptable these glasses are for both indoor and outdoor use the lenses automatically darken in sunlight much like regular sunglasses so you can comfortably wear them out outside while still

enjoying all the AR functionality it's a smart design choice that gives spectacles a broader appeal especially compared to bulkier devices that don't transition so smoothly between environments now here's where things get fun the video commercial snap released for spectacles five pokes a bit of fun at the competition specifically Apple's Vision Pro and meta's Quest 3 in the ad snap exaggerated the size and shape of competitor AR headsets to make them look ridiculously bulky and awkward emphasizing how streamlined and compact their own glass are by comparison and honestly they're not wrong apples and meta's headsets are

undeniably bulky and while they're powerful they don't exactly scream wearable tech for everyday use spectacles 5 on the other hand look much more like regular sunglasses they're light stylish and blend in better with a casual everyday aesthetic rather than feeling like you're strapping a high-tech machine to your head but despite how polished and accessible these glasses may seem they're not aimed at the general consumer Market just yet spectacles five are currently available through a developer program it costs $99 per month for at least a year which gives developers access to the glasses and the tools

to create AR applications snap has emphasized that the definition of a developer is Broad inviting not only traditional software Engineers but also creatives designers and artists to experiment with the platform this is where the future of AR gets built and snap is opening the doors to a wide range of innovators all right moving on to another major announcement snap is introducing a new tool for AI generated video creation this feature enables creators to generate videos based on Simple Text prompts for now this tool is in beta and limited to a small group of creators but

the implications are massive in a world where content creation can be time consuming the ability to produce highquality videos with just a text prompt streamlines the creative process here's how it works you input a phrase such as cat walking through a park and the tool generates a fully animated video in response while the current iteration of the tool only supports text prompts future updates will allow creators to generate videos based on image prompts as well the flexibility this tool offers is substantial it's designed to simplify and speed up the creative process without sacrificing quality offering

an alternative to traditional labor intensive video production snap is actually ahead of the curve with this feature while meta and Tik Tok are working on similar tools neither has rolled out anything publicly yet the videos gener generated by the tool will all include a snap AI Watermark ensuring transparency around AI generated content the video generation tool is powered by snap's foundational AI models which are trained to understand and generate content in a way that aligns with the platform's ethos of creativity and safety with text to video Tools in place the next big leap would be

text to animation which the company is already working on this Evolution could allow users to create animated sequences potentially revolutionizing how creat ERS engage with their audience through Dynamic Aid driven storytelling Snapchat has been a leader in augmented reality for years and now with AI powered lenses and memories it's stepping things up even further these new AI lenses make it simpler to create interactive AR filters whether it's something subtle like virtual makeup or more extreme Transformations like turning yourself into an alien the AI handles the technical side so creators can focus on designing unique immersive

experiences for users in in addition to lenses the memories feature snap's built-in tool for storing and revisiting past snaps has been upgraded with AI 2 now when you revisit your old photos or videos the AI can help autoedit them create collages or even suggest creative ways to relive those moments for instance AI can add subtitles enhance the images with new effects or reimagine entire scenarios in a more engaging way this feature is particularly useful for users looking to relive their best moments but with a new twist enhancing the Nostalgia with modern technology now Snapchat's my

AI chatbot has been given a major overhaul as well it can now read road signs translate menus and offer even more practical assistance this is part of snap's broader strategy to embed AI deeply into everyday functions making the app more useful Beyond its core social features what's impressive here is that this AI isn't just a gimmick or novelty it's a practical tool for users who are out and about trying to make sense of their surroundings one final note about snap's AI integration snap's AI selfie feature has raise some eyebrows this feature lets users create AI

generated images of themselves based on photos they upload to the app but there's a catch when you opt into this feature snap gets the right to use your AI generated face in personalized ads these ads are only shown to you but your likeness can be featured in them without compensation while snap assures users that their data won't be shared with thirdparty advertisers the idea of seeing your AI generated face in ads even if they're just for you might feel settling the good news is you can turn this off if you're not into the idea of

your face being used in ads snap has provided a way to disable this feature in the settings still it's worth keeping an eye on how Ai and privacy issues evolve as these tools become more integrated into our digital lives Snapchat is making a serious push into the future of AI and AR with spectacles 5 they're merging the digital and physical worlds in a practical and exciting way the new AI video generation tool gives creators fresh ways to make content and updates to lenses memories and my AI make the platform more immersive all right if you're

interested in more deep dives into AI Robotics and the future of tech make sure to like subscribe and leave a comment thanks for tuning in and I'll catch you in the next one