AI Weekly News Rundown Marc 23 to March 30 2025: 🧠 LLM Embeddings Explained: A Visual and Intuitive Guide 🔬Infomorphic Neurons Bring AI Closer to Brain-Like Learning 🔍Microsoft Adds Deep Research to AI Code Editors 👁️Qwen’s QVQ-Max
31:36 Share

Welcome to a new deep dive from the podcast, AI Unraveled, created and produced by Etienne Newman. He's a senior software engineer and passionate soccer dad from Canada. And if you're enjoying these explorations of AI, please do take a moment to like and subscribe to the podcast on Apple. It really does help us out.

So today we're diving into a whole bunch of recent happenings in the world of AI. We've got news covering AI's role in journalism, you know, producing it and sometimes kind of struggling with it, ethics and marketing with AI, and some really cool advancements in understanding how these complex AI models work, you know, what's going on under the hood. As always, we aim to break down those key insights and see what they mean for you. Okay. Yeah, let's get into it.

Bloomberg, they're a big name in financial news, right? And they recently tried out AI generated summaries. I guess the thinking was probably like efficiency, getting info out there quickly. Yeah, that was definitely the goal. But the user experience, well, it wasn't exactly smooth sailing at first. Feedback highlighted some accuracy problems and the writing.

It just felt kind of awkward, not very human. I guess it shows that generating reliable and readable journalism, it's not as simple as we might think for AI. Makes you think about what makes writing good journalism so tough for AI. It's not just spitting out facts, right? It's interpretation, it's context, even like a narrative flow that matters. Exactly. AI is great at processing tons of data, but making it into a compelling, accurate and readable story that still needs human expertise.

The errors, they weren't just small things. They really affected the trustworthiness of those summaries. It shows how important human oversight is to keep that journalistic integrity intact.

OK, let's switch gears to fashion now. H&M wanted to use those AI generated models, digital clones basically. What was the main reaction to that? It definitely kind of blew up. People were concerned about authenticity. Like what does it even mean for a brand to represent itself with something that isn't human? Brings up questions about representation and how consumers see it, especially in an industry that's all about human ideals. Right.

Right. Fashion thrives on aspiration, that connection to human figures. So AI models can feel fake. People expect transparency. H&M's move sparked a bigger debate about using AI to create these potentially unrealistic beauty standards. It makes you wonder about the psychological impact seeing AI generated perfection versus real human models with their imperfections in those ads. Absolutely. It's a tricky area. With ethics,

and what consumers want constantly changing. Let's get a bit more technical. There's been progress in figuring out how LLMs, those large language models, understand language. I heard about this new visual guide that's trying to explain embeddings to make them easier to understand. Yeah, that's a big step towards making this key AI concept more accessible. It's an interactive guide, and it focuses on LLM embeddings. Think of it like numerical representations, vectors that capture what words and phrases mean.

They're the foundation of how LLMs process language and connect the dots between different pieces of text. So instead of just being like abstract math,

This guide uses visuals to help us actually see how those words and ideas are related in the way the LLM understands them. Exactly. Imagine a map of meaning where words with similar meanings are closer together in a, well, a very high dimensional space. This guide uses visuals and examples to illustrate it so we can grasp how LLMs understand context and do things like text similarity or analogy. This kind of clear explanation is crucial. It helps more people understand AI. Love it.

of it. The more we understand the basics, the better equipped we are to deal with this tech, right? Speaking of learning, there's that research into infomorphic neurons trying to get AI to learn more like a real brain. What's the core idea there? It's cutting edge stuff.

Infomorphic neurons are a new kind of artificial neuron designed to mimic how our brains process and transmit information way more closely than traditional AI neural networks. The goal is to create AI that learns faster and adapts better, more like us. Now this is where it gets really cool. So if we can build AI that learns more like a brain,

What could that mean down the line? The possibilities are huge. By copying how biological learning works, we could have AI systems that use way less energy and data. They could be more adaptable and less prone to breaking down in new situations. Imagine breakthroughs in robots, medical diagnosis, even understanding what intelligence itself is. It could change how we design our neural networks entirely. Wow. That is a deep thought. We're shifting gears a bit. Oh,

OpenAI seems to be attracting a ton of investment with another big funding round in the works. Yep. Reports say OpenAI is close to a whopping $40 billion in funding. That's a lot of faith in their potential, and it'll definitely fuel their research and development in this fast-moving AI world. And they're expecting some seriously impressive revenue growth, too. We're talking some big financial numbers, huh?

What's the outlook? OpenAI thinks their revenue will triple this year, hitting $12.7 billion. And looking even further, they're projecting, get this, a potential $125 billion by 2029. But to hit those targets, they need to keep investing heavily.

In 2024, they had some big losses because, you know, building and training these advanced AI models and the infrastructure it takes, that doesn't come cheap. Things like their participation in that Stargate AI infrastructure joint venture, that's a big cost. Right. So even though the potential revenue is huge, the initial costs of leading in AI development are massive too. This funding seems key for them to stay competitive and keep pushing the boundaries of what AI can do. Exactly. It lets them attract top talent.

Invest in the best infrastructure and pursue that groundbreaking research, keeping them at the forefront of the global AI scene. OK, let's talk about Anthropic now. They've been doing some fascinating work trying to figure out how their AI model, Claude, actually thinks. It feels like a step towards making these systems less mysterious. It does. It feels like black box AI is, well, kind of scary.

Understanding how these advanced AI models work is crucial for us to trust them and to make sure they're developed responsibly. Got it. Anthropic has been using some really sophisticated techniques to peek into Claude's decision making. One amazing finding is that Claude seems to use a universal language of thought.

thought, something that goes beyond specific human languages like English or Chinese. It suggests a deeper understanding than just processing words. Whoa, hold on. A shared way of representing concepts even across different languages. That's wild. It suggests there's a deeper cognitive structure at play, not just language processing. They've also found cool strategies Claude uses for creative text, like it plans ahead to include rhymes in poetry.

And it has this built-in tendency to avoid guessing unless it's really confident, which is a key factor in preventing those hallucinations AI sometimes has. - So by understanding how Claude's mind works, we can, well, not only understand what it can and can't do, but maybe even guide its development in a way that matches our values. - Exactly. This transparency is vital for trust, for regulation, and ultimately for making sure AI develops ethically and responsibly.

Let's move from, well, AI brains to AI tools. Microsoft just launched this deep research feature for their AI-powered code editors. How's that changing things for developers? That's like a productivity boost.

This new feature basically lets developers do research, look at documentation, even find and use code examples right there on their coding environment. No more switching back and forth between apps. That's the key. It streamlines things. Instead of jumping between apps and websites to find info, developers have everything they need right there in context. It could really speed up development and make tough problems easier to solve. Anything that keeps developers focused and productive is a win, right?

Let's talk visual AI now. Alibaba's Quinn team unveiled their QVQ Max model. It's designed to make AI better at understanding and reasoning with visual information. What's special about it? QVQ Max is meant to go beyond just recognizing objects and images. It's built to interpret whole scenes, you know, figure out relationships between different things and actually reason based on what it sees. It's a step towards a deeper understanding of visual data. So not just there's a cat, more like the cat

The cat's on a map, which is in a room. Exactly. And it has this cool adjustable thinking mechanism. Research shows that giving it more steps to process that visual info makes it way more accurate, like analyzing blueprints, solving geometry problems, even giving feedback on sketches. Alibaba even mentioned they want to build a full visual agent someday.

one that can interact with devices and play games. It could be huge for things like self-driving cars, advanced medical imaging, even AI assistants that actually understand the world around them, right? Absolutely. It's a big step towards true artificial general intelligence with strong visual perception. Now, all this progress in AI, it comes with some growing pains, too. OpenAI recently had to limit chat GPT usage, right?

Why was that? Well, it seems like ChatGPT got so popular, so many people were using it, that their systems just couldn't keep up. There were even reports of their GPUs, those graphics processing units, the chips that power these models, literally overheating. It just shows how much computing power is needed to run these AI services.

The demand is pushing the hardware to its limits, and that impacts users directly. Remember those temporary limits and image generation being delayed for free users. Yeah, limiting image generation to paid users at first wasn't even enough to ease the strain. They're working on improving efficiency, hoping to lift those restrictions soon. Free users should eventually get image generation too, but maybe with usage caps. It shows that both the AI algorithms A and D of the hardware need to keep getting better.

Absolutely. We need more scalable and energy efficient hardware to meet the demand. Switching back to Anthropic, they've developed this cool tool they call an AI microscope.

What does it do? It's a tool that helps researchers look at and play with the reasoning process of LLMs, models like CLAWD. Basically, it lets them see how these models reach their conclusions. And the stuff they're finding out is amazing. They've seen that CLAWD does this multi-step reasoning, activating different parts of its neural network in sequence. When it creates poetry, it actually plans ahead to make sure rhymes fit in. And for math, it seems to use parallel processing like our brains do. Isn't that cool?

There's even a separate Google study that found similar brain activity patterns between AI models in humans during conversations, though the way they process is still very different, of course. So this AI microscope is showing us how these models work, something we couldn't see before. It has to be super valuable for building AI that's more reliable and trustworthy. Totally agree. By understanding these models better, tools like this make things more transparent, build trust, and ultimately help us steer AI development in a safer, more responsible direction.

OK, let's switch gears a bit. WhatsApp is now the default for calls and messages on iPhones. That's a big change in the mobile world. It is. Looks like it started with EU regulations. Yeah. But Apple has now rolled this out globally with iOS 18.2. So for iPhone users who update, WhatsApp takes over from Apple's iMessage and FaceTime as the go-to app. The interesting part is how this might change the whole mobile landscape.

Think about user privacy, how apps integrate with the operating system, and even the competition between these tech giants. It's a big move for both Apple and Meta, Facebook's parent company. Yeah, something to watch for sure. Yeah. Yeah.

Back to those resource strains, it sounds like ChatGPT's image generation feature is putting a lot of stress on their systems again. Yep, those melting GPUs. The demand for image creation is huge, really highlighting how computationally intensive these generative AI tasks are. They're still working on more efficient hardware and algorithms. Reinforces the fact that as AI gets more powerful and versatile, the infrastructure has to keep up. Now, in education, there's this Harvard professor who's created an AI replica of himself to be a personalized tutor.

That's a pretty unique approach. It's an amazing experiment. This AI replica can interact with students 24/7, giving each one personalized learning support. It's like scaling up the professor's availability. So students can essentially get one-on-one tutoring from a virtual version of their professor whenever they need it.

That could really change how accessible education is. Yeah, it could democratize personalized learning and change traditional teaching. But like we've seen with AI in other areas, it also raises ethical questions. What about human interaction in education? Could we become too reliant on AI in a field that values human connection? Definitely. Lots of potential benefits. But we need to think about the downsides, too.

Now, a more serious topic. Reports say North Korea's new drones likely use AI for finding targets and carrying out strikes. That's a big escalation. It's very worrying. If true, it's a significant step in autonomous weapons systems.

AI choosing and attacking targets on its own, that raises huge ethical and security issues for the whole world. It'll definitely ramp up those discussions about regulating AI in warfare. It's a stark reminder that AI can be used for good or bad, and that we really need to tackle these ethical and security challenges head on.

On a different note, open source developers are starting to fight back against AI web crawlers that take their code without permission. What's driving this? There's this growing movement among developers to stop AI crawlers from grabbing their code from repositories without asking. They want to protect their work, control how it's used, and make sure they get credit.

So it's a clash between AI needing lots of data and the rules of intellectual property in the open source community. Exactly. This resistance shows the tension around data for AI training. It could lead to new rules or even laws about how AI can use public data.

Now, in a major business move, Elon Musk's XAI, his AI company, just bought the social media platform X, formerly Twitter, for a whopping $45 billion. That's a huge investment. That's massive. Seems like a strategic play. The goal is to integrate XAI's AI, like their chatbot Grok, with X's huge user base. They want to build an everything app,

kind of like WeChat in China, by bringing in AI-powered features. The potential here is pretty exciting. X has the users and the reach, while XAI has the cutting-edge AI. This could really boost XAI in the AI race and turn X into a much more AI-focused platform, which is what Musk has been talking about. It definitely changes the landscape for AI social platforms. And it creates a really interesting player in this everything app space.

OK, let's talk about something a bit more creative. OpenAI's image generation tool had a, shall we say, a Studio Ghibli moment recently. What happened there? People figured out how to use OpenAI's tool to create images in the style of Studio Ghibli, the famous Japanese animation studio. It went viral with people making all sorts of scenes and characters with that instantly recognizable Ghibli look.

I bet it looked amazing. But I imagine that might raise some copyright issues, especially with Hayao Miyazaki, Ghibli's co-founder, being skeptical of AI generated art. Right. Artistic style isn't copyrighted, but OpenAI's tool mimicking Ghibli so well raised the question. Was it trained on Ghibli films without permission?

Legal experts are debating whether it's fair use or infringement. It shows the tricky legal ground when AI can replicate the look of artists and studios so well. So even if it's not copying exact images, the ability to copy a style creates a whole new set of copyright questions could lead to new laws or interpretations of existing ones. Yeah, definitely something to keep an eye on as this tech keeps improving. Now, let's dive a bit deeper into those open AI financial projections. The numbers we talked about were pretty

were pretty mind blowing. They were tripling revenue to twelve point seven billion dollars in twenty twenty five. That's huge growth and then exceeding one hundred and twenty five billion dollars by twenty twenty nine. That's just crazy. Looks like chat GPT pros doing well, plus more people using their API, those enterprise tools and those team plans. But as we mentioned before, even with all that revenue, they expect to be in the red until twenty twenty nine. Why? Because they're investing so much in things like that advanced computing hardware, training their AI models and expanding their whole infrastructure.

It's super expensive to be at the forefront of AI. It's like running a major cloud computing operation, but with the added cost of pushing the very boundaries of AI. High risk, high reward. If their predictions are right, the payoff will be massive.

Speaking of image generation, Ideagram just launched their 3.0 model. How does it compare? Ideagram 3.0 is a big step up in AI image generation. It's getting praise for creating photo realistic images, coming up with creative designs, sticking to certain artistic styles, and being faster than before. It's available to everyone on their website and iOS app now too. What's better about it compared to the others? One big improvement is how it handles text within images and those complex graphic design elements.

You can create layouts, include logos, handle typography, stuff that's been a challenge for other models. They even did tests and beat models from Google, Flux and ReCraft in those areas. They also added this style references feature. You can upload images to influence the style and they have a big library of preset styles too. And it's all free. That's amazing. Sounds like Ideagram 3.0 is setting a new bar.

with more realism and versatility for everyone, from designers to casual users. Totally. Now let's talk about AI in cars. BMW and Alibaba are teaming up to put advanced AI into BMWs in China. What are they planning? It's all about putting cutting-edge AI into those new BMWs made in China.

They're working together to create a custom AI engine to boost their Intelligent Personal Assistant, or IPA. Better voice recognition, more context awareness, that kind of thing. It'll be powered by Alibaba's Quinn AI model and is set to launch in their new class models in 2026. The cool thing is how they're focusing on real-time services through voice commands. Imagine finding restaurants, checking parking, getting traffic updates, all hands-free.

BMW is also planning two new AI agents, CarGenius for car diagnostics and Travel Companion for personalized trip recommendations. They're even adding gesture and eye tracking for a more natural experience. They want to make the car more intuitive, more like a partner that understands you, not just a machine. This could redefine in-car tech.

with a big car company and an AI giant like Alibaba working together. - For sure. Now, Alibaba also has this new multi-sensory AI model, QEN 2.5 Omni 7B. It's built to run smoothly on mobile devices, right? What are the key things about it? - It's all about handling text, images, audio, and video on your phone or laptop, really boosting those multimodal capabilities.

Get this. Both Apple and BMW are apparently going to use Alibaba's models in their products in China, and they made QN 2.5 on the 7B open source so anyone can use it. The model uses this thinker talker system to handle all those different data types in real time, focusing on natural sounding text and speech. It's actually outperforming some specialized audio models and tests.

Alibaba says it runs efficiently on mobile hardware, so think about real-time audio descriptions for people who can't see well, for example. Super versatile. Could really change how we use our phones and more. And making it open source is huge. Could lead to lots of new apps and ideas.

Now, Bill Gates made a bold prediction. AI will replace a lot of doctors and teachers within a decade. What are your thoughts on that? It's a strong statement, definitely. Gates thinks AI will get so good at complex decisions and personalized learning that we'll need fewer humans in those roles. It's about a huge shift in jobs and how society works. We really need to think about how we'll adapt to this kind of change.

What will it mean for schools, for jobs, for our whole economy? Exactly. We have to prepare for a future where AI plays a much bigger role in fields that have always been about human skills and interaction. Back to ChatGPT, it has a powerful new image generation feature now, right? Yeah, OpenAI has brought GPT-4.0's image generation right into ChatGPT. You can create detailed images right there in your chat.

It's become this versatile creative tool doing both text and images seamlessly. The image quality is reportedly really good, making ChatGPT a serious competitor in the AI image generation game. It can handle prompts with lots of objects and even crazy abstract ideas. You can refine images with just words and the AI remembers what you said before.

It's still got some issues with text and images and those super complex scenes, and free users are still waiting because of that high demand. Still, it's a big step forward for ChatGPT becoming this all-in-one creative assistant. More competition for those standalone image tools.

Now, for something a bit more concerning, Kim Jong-un, North Korea's leader, showed off some new military tech, including AI-powered suicide drones. That's pretty scary stuff. It's definitely a worrying development. They showed a bigger reconnaissance drone, too, along with these suicide drones, which are designed to hit specific targets and operate on their own. It shows North Korea is really into AI for their military now, could destabilize things in the region. These kinds of autonomous weapons raise huge ethical and security questions for everyone.

everyone. It'll definitely intensify the debate about regulating AI in warfare. Okay, back to Alibaba's open source model, QEN 2.5 Omni 7B.

We talked about its multimodal abilities, but it's also designed to make building AI agents cheaper. Right. That's a key goal. By making it open source and optimizing it for phones and laptops, they're making it easier and cheaper for developers to build AI agents. Think about consumer electronics, cars, all sorts of applications. It can process different data types quickly and efficiently, which is key for building truly versatile AI assistance.

Open sourcing it is a big move, could let lots of developers and companies use this tech without the cost of proprietary models that could lead to tons of innovation. Now, Google has also made some waves with Gemini 2.5 Pro, their latest AI model. What's new in this one? Gemini 2.5 Pro is a big upgrade. It uses a mixture of experts architecture, which is great for tests that need a lot of context like complex reasoning, math, coding and logic.

It's reportedly doing better than GPT-4 and Claude in some key benchmarks. It's available for developers via API and also powers that Gemini advanced subscription. Sounds like it's aimed at those tougher enterprise level tasks. What makes it so much better? One of the coolest things is it can reason and double check its own work, which is crucial for things like software development.

It also has a huge context window of one million tokens, so it can process tons of information at once. And it's breaking records in AI benchmarks, showing how capable it is. Sounds like Google's serious about making Gemini 2.5 Pro the go-to AI for complex work and problem solving.

Now OpenAI has finally launched image generation in ChatGPT for paid users. Yep, if you have Plus, Pro, or Team, you can generate those images right in your chats using GPT-4.0. Free users have to wait a bit longer. This version can handle prompts with multiple objects and those creative, out-there ideas. You can refine images with just words, making the whole process more interactive. Multimodal capabilities right there in ChatGPT.

It's a more versatile creative tool, but it's also stepping up the competition with those dedicated image platforms.

Robby Barbaro: Microsoft is also expanding their AI tools with new AI agents for Copilot and researchers and data analysts. What can you tell us about them? Amy Quinton: They're adding two new specialists to Copilot for Microsoft 365. First, there's Researcher, which is all about complex searches, pulling info from the web, company documents, all sorts of sources. It can then compile reports and generate insights. Then there's Analyst, which is basically a virtual data scientist helping users clean, visualize, and analyze data.

Researcher uses OpenAI's deep research model, while analysts uses their O3 mini reasoning model. Both should be out in April. It's a sign that we're getting more specialized AI tools built for specific jobs. We might see a lot more of these changing how we work in fields like finance, research, you name it. Totally. In the legal world, Anthropic just won a round in a copyright case with music publishers. What was the ruling?

A judge in the US said no to a preliminary injunction requested by Universal Music Group and some other big music publishers. They claimed Anthropic's Claude model was infringing by generating song lyrics. But the judge said they didn't prove they'd suffer real harm at this stage, so no injunction. The case itself isn't over, though. This is a big win for AI developers, at least for now. The judge's decision suggests that using copyrighted stuff to train AI might not be seen as automatic harm and might not always need up

upfront permission could change how AI copyright works in the future. It's a landmark case for sure, one that people in both the AI and music world will be watching very closely. OK, switching gears again, Google's quantum AI chief, Hartmut Nevin, made a pretty optimistic prediction about commercial quantum computing. He said we could see big breakthroughs in commercial quantum computing within five years, which is sooner than most people think.

He credits progress in things like error correction, better simulations, and material science. If he's right, how could that change things? It could be revolutionary. Quantum computing could go from theory to a real powerful tool. Think about drug discovery, material science, cryptography, even training those complex AI models could change how we compute entirely.

Big changes coming. Apple's making a big bet on NVIDIA's AI hardware too, right? Yeah, they're reportedly spending $1 billion on a bunch of NVIDIA GB300 NVL72 servers. Around 250 of them specifically designed for the heavy lifting needed for generative AI and LLMs. They're working with Dell and Supermicrocomputer to build a huge server cluster for their AI work. That's interesting because Apple usually focuses on their own Apple Silicon for AI, talking about the privacy and security advantages.

It seems they're prioritizing power and scalability now, even if it means using someone else's hardware. These Nvidia servers are beasts, perfect for AI work. It shows how serious Apple is about AI, but it does raise questions about how they'll maintain their focus on user privacy with this external hardware.

Still, it puts them in a better position to compete with those other tech giants who are investing heavily in AI. They want to be a leader in this space, no doubt. NVIDIA has been showing off their vision for the future, too, and it looks like robots are a big part of it. At their GTC 2025 event, they painted a picture of a future driven by AI and robots. Jensen Huang, their CEO, talked about using their new Blackwell chips and AI foundation models to power the next generation of humanoid robots.

We saw demos from companies like Agility Robotics, Disney, and even Boston Dynamics, all using NVIDIA's Isaac platform for their robots. So they're not just providing the AI brains. They want to provide the whole infrastructure for this robot revolution. Exactly. They want to be the foundation of this new AI-powered robot economy. It's a glimpse into what they think is the next big thing. Intelligent robots in the real world. That's a bold vision.

DeepSeq, another AI company, quietly released a big upgrade to their AI model. How does it compare to the big names? Their new model, DeepSeq V30324, is meant to rival GPT-4 and CLAWD. Early reports say it's much better at things like reasoning, coding, and translation. And get this, it does all that with fewer parameters than some of its rivals.

DeepSeek's also switched to the open source MIT license, aiming to be a leader in open source AI, kind of like Meta with their Lama models. Interesting. What's the feedback from developers? Developers who've tried it say it's great for coding, creating complex code with few errors. Some say it sounds a bit more robotic now, but with its open source license and those improvements, DeepSeek is a solid alternative to those Western models, especially for those who need powerful, lightweight, and multilingual AI. Lots of competition in the AI model space, and it's changing all the time.

Character.ai just added something parents will be interested in, parental controls. Yep. Their new parental insights tools let parents see how their kids are interacting with the AI bots on the platform. They can see which bots their kids are talking to, how often, and for how long. Importantly, they can't see the actual conversation, so it's about balance, giving parents some control while protecting kids' privacy. Good move.

Parents need some reassurance as AI becomes more common for kids. Finding that balance is key. It's a model for other platforms designed for young people. As more kids use AI, we'll likely see more of these AI supervision features to address those parental concerns and make things safer. Now for something different.

A startup called Earth AI is using AI to find valuable minerals. Talk about thinking outside the box. Instead of traditional surveys and digging, they use AI to analyze geological data and satellite images to find places where minerals like copper, lithium and rare earth elements might be. Yep.

These geological AI models help them find potential sources of those minerals, which are vital for clean energy tech like electric cars and renewables. This approach could be much better for the environment and cheaper than traditional methods. It's a cool mix of climate tech, geoscience, and AI. It's an innovative and hopeful way to find the resources we need in a more sustainable way.

There's also been a breakthrough in medical diagnostics. A new AI is incredibly accurate at detecting a specific cancer.

Researchers developed this model called ECG-MLP. It can detect endometrial cancer from tissue images with 99.26% accuracy, which is better than human specialists and other automated tools. Wow, that's amazingly accurate. How does it do that? It uses special attention mechanisms to find tiny patterns and signs of cancer cells that humans might miss. They even tested it on other cancers like colorectal, breast, and oral cancers, and it did really well there too. This could be huge.

Imagine early, accurate cancer detection for all kinds of cancers. It could save so many lives and make expert-level screening available to everyone. Absolutely. It's potentially a game-changer.

Now, back to the tech giants. Apple's facing a lawsuit about delays in those promised AI features. Yeah, there's a class action lawsuit against them for false advertising. People are saying those AI upgrades and features for iOS and Siri were promised but haven't arrived on time. It shows the pressure these companies are under to deliver on their AI promises.

AI is a big selling point now, so not meeting those deadlines could hurt them badly. This case could set a precedent for how companies talk about and are regulated on AI features. It's a reminder that hype isn't enough. In the fast-paced world of AI, you've got to deliver.

Google just released some really interesting real-time video AI features for Gemini. What can you tell us about them? They've launched real-time video and screen sharing features. Think about live translation, transcription, video summarization, and help during video calls.

They're introducing Gemini Astra, an AI assistant that analyzes live video and audio, responding to what it sees and hears. So it's like an AI assistant right there in your video call, understanding what's happening. Exactly. Imagine instant translation during international calls or getting info about what's on your screen. It could change how we collaborate remotely, handle customer service, and make things more accessible for people with disabilities.

But like always, we have to think about the ethical and privacy implications of using video data like this. Powerful tools, but we need to use them responsibly. Now, Cape Town in South Africa is considering using AI for traffic lights. They're testing an AI traffic management system that could cut vehicle stops by 30%.

It uses real-time data on traffic, accidents, and pedestrian activity to adjust the timing of lights. That could transform how people get around in Cape Town. Think about smoother traffic, shorter commutes, even cleaner.