AI Daily News March 23 2025: 🔬New AI’s near-perfect cancer detection 🔮Apple Explores Integrating AI-Enabled Cameras into Future Apple Watch Models 💥DeepSeek Releases Major Upgrade: V3-0324
17:29 Share

Welcome to AI Unraveled, the podcast dedicated to exploring the ever-evolving world of artificial intelligence. Created and produced by Etten Newman, a senior software engineer and passionate soccer dad from Canada. If you find these deep dives valuable, please like and subscribe on Apple.

So today we're taking a look at some of the recent AI breakthroughs. From March 23rd, 2025. That's right. And we're focusing on extracting the most fascinating and intriguing insights. That could impact your world. We've got everything from revolutionary advancements in healthcare. To potential game changers in tech you use every single day. And we'll even be looking at how AI could be reshaping the cities we live in. Okay, let's jump right in. Sounds good. First up,

AI achieving some really remarkable feats in health care. Particularly cancer detection. Yeah, this is amazing. There's this new AI model called ECGMLP. ECGMLP. And it's demonstrated a truly unbelievable accuracy in detecting endometrial cancer.

We're talking 99.26% accuracy. From microscopic tissue images. That's incredible. Yeah. Analyzing images at a cellular level. Right. And outperforming human experts. Yeah, by a significant margin. So how much better is this than cardiology?

current diagnostic methods? Well, current methods for endometrial cancer by human pathologists have an accuracy that ranges between 78 to 81%. Oh, okay. So we're looking at almost a 20 percentage point jump. Wow. Which is a really significant potential paradigm shift in how we approach cancer screening. So if this

AI proves successful. Right. What could it mean for the average person? Well, for you and me, higher accuracy could lead to earlier detection. Right. Which, as we know, is absolutely critical for a successful treatment of cancer. Of course. And ultimately could save lives. Wow.

Further, imagine democratizing access to expert level screening globally. In regions where there may be a shortage of specialized pathologists, AI could really step in and bridge that gap. That's amazing. Yeah. Making these advanced diagnostics available to so many more people. Exactly. Now, ECGMLP didn't just show promise with endometrial cancer. That's correct. There were other cancers too, right? Yes.

The researchers also tested it on a number of other cancers. Okay. Including colorectal cancer, breast cancer, and oral cancers. And how did it do? It was really impressive. Okay. It had accuracy rates of 98.57% for colorectal. Wow. 98.20% for breast cancer. Okay. And 97.34% for oral.

That's amazing. So it indicates a very strong versatility in the model's ability to actually identify cancerous cells. Across different types of tissues. Exactly. That's really promising. It is. But moving on from healthcare for a second. Sure. It seems like AI isn't always so seamless in other areas. Yeah. Like with this Apple lawsuit. Yeah, this class action lawsuit against Apple. Right. Is pretty revealing. About AI's integration into our tech.

So what happened? Well, they've been accused of false advertising. OK. Because they delayed certain AI features in iOS and Siri. Oh, so consumers feel like they were misled about the timing of these upgrades. That's the central claim, yeah. The lawsuit suggests that Apple's marketing efforts

you know, created this expectation among consumers that these advanced AI functionalities would be available. Right. You know, within a specific timeframe. And they weren't. And they weren't delivered as advertised. So what are the implications of this lawsuit for Apple and for the tech industry as a whole? Well, I think beyond the immediate legal ramifications for Apple. Okay.

This case really underscores a growing tension. Between? The hype surrounding AI and the much slower reality of deploying it. Right. So I think this might mean that you as a consumer might want to be a bit more discerning.

Marketing claims related to AI. Okay, so like taking those promises with a grain of salt. Exactly. And perhaps expecting greater transparency. Tech companies. Regarding. AI roadmaps and timelines. For when these features will actually be available. Yeah. Okay, interesting. It could set a precedent. How companies are held accountable. For the AI capabilities that they advertise. Precisely. It's a good reminder that while AI is advancing rapidly. Yes. It's still under development.

Very much so. Okay, well, let's move on from this legal battle. Sure. And talk about a company that's actively pushing forward with AI. Okay. That's Google. Ah, yes, Google. And they've rolled out some pretty sophisticated video AI features for Gemini. Yeah, it's really interesting to see AI move beyond...

analyzing like static text and images to actively processing live video feeds. That's pretty amazing. It is. So what kind of features are we talking about here? We're looking at real time video interpretation, live transcription, summarization of video content, even

context aware assistance during video calls or when you're sharing your screen. That's incredible. Yeah, it's pretty amazing. And all of this is powered by Gemini Astra. Yes. Which is an AI assistant that can essentially see what's happening in a video in real time and respond intelligently. Exactly. That's incredible. Yeah. So Gemini Astra is designed to analyze visual information. Okay. And provide these relevant and timely responses.

So is this available to everyone? It's available to subscribers of Gemini Advanced. Okay. At the moment. So Google is really trying to stay ahead of the game here. That seems to be the strategic motivation, yeah. It's really interesting to see how these different companies are competing. Yeah.

Yeah. While other major tech companies are reportedly experiencing delays in launching similar real-time video AI features. Google is just moving full speed ahead. It seems so, yeah. So what kind of impact could this have on our lives? Well, you can imagine in like remote collaboration. Okay. You have an AI that can follow a discussion. In a virtual meeting. Yeah. Provide a concise summary afterward. Exactly. That would be amazing. Or even in customer support. Okay.

Okay. And AI could visually guide someone through a technical issue. I see. Troubleshooting something. Okay. It could really help with accessibility too. How so? Real-time language interpretation during a video call. Oh, wow. Imagine breaking down all those communication barriers. That would be incredible. It would be a game changer. But are there any concerns with these capabilities? Well, of course not.

As with any new technology, there are always concerns about privacy and potential for surveillance. Especially in professional or educational settings. Exactly. So these are definitely ethical implications that we need to consider. As this technology becomes more widespread...

It's a lot to think about. It is. Now let's move on from the digital world to the physical world. Let's talk cities. Cities. Cape Town is exploring an AI-powered traffic management system with a really ambitious goal. So city officials in Cape Town are evaluating a system that aims to reduce the number of vehicle stops.

At intersections. At traffic intersections by a significant 30%. 30%. That's the goal. That's huge. Yeah. So the idea is that the AI would dynamically adjust the timing of traffic signals. Okay. Based on a constant flow of real-time data. Taking into account things like traffic congestion. Exactly. Congestion accidents. Okay. Pedestrians. All kinds of factors. So instead of traditional fixed timing for traffic lights. Right.

This system would constantly be learning and adapting. To the actual traffic conditions. That's fascinating. Yeah. So this dynamic adaptation is really the core innovation. And how would it improve things?

By optimizing these green light intervals and signal patterns in response to real time conditions, the system has the potential to significantly improve traffic flow, reduce air pollution. From idling vehicles. Exactly. And ultimately shorten commute times for everyone. A 30% reduction in stops would make a huge difference. Absolutely. Yeah. In daily commutes. Yeah. That's the hope. So if this is successful, Cape Town could be a model for everything.

other cities. That's certainly the hope. Looking to leverage AI to address their own urban mobility challenges. Right. It's exciting to see how AI is being used. To tackle these large scale infrastructure issues. It is. Yeah, absolutely. Now, speaking of large scale applications. Okay. We also saw the Trump administration launching a new AI platform. Yes. To be used across the federal government.

That's right. And I think it's really notable. What is? This increasing recognition within government. Of AI's potential. Exactly. To enhance efficiency and streamline operations. Yeah. So what are the objectives of this new platform? Well, they're stating things like accelerating document analysis. Okay.

Providing more informed decision support for policymakers. Okay. And improving service delivery to citizens. So they want to modernize some of those slower, more bureaucratic processes. That seems to be the main drive. Are there any concerns about this? Well, of course, as with any new technology, you know, there are always potential downsides. When you increase integration of AI into federal infrastructure,

It invites scrutiny. How that data is being used. The level of transparency in algorithms and the potential for unintended biases. So this could really impact the future of AI regulation. I think so. And standardization within different agencies. Yeah. Okay. Let's shift gears back to consumer technology. Sounds good. Back to applications.

Apple. It seems they are looking for ways to embed AI into their wearables. Yes, specifically the Apple Watch. Oh, interesting. So there are reports suggesting that they're looking into integrating cameras. Okay. The future versions of the Apple Watch. So both the standard series and the ultra model. That's right. Cameras on an Apple Watch. Yeah. What's the idea behind that? Well, it seems to be linked to this concept they're calling visual intelligence. Okay. So the aim is to equip the watch with

with the ability to perceive its surroundings. Perceive. Essentially. And then? Provide the wearer with context-aware information based on what it sees. So it could identify landmarks. Yeah. Translate text. Recognize objects. Precisely. Those are the types of functionalities they're hinting at. That's really interesting. It is. So Apple is really focusing on

on weaving AI into their existing products. That seems to be their strategy. Rather than venturing into new categories. Of dedicated AI devices. Like some other companies have done. Right. But when can we expect these camera equipped Apple watches? Well, the report suggests they are still some years away. Okay. Potentially a few generations away. Okay.

But of course, this raises significant privacy concerns. That Apple will have to address. Absolutely. Okay, let's talk about the technology driving all these advancements. Okay. AI models themselves. The heart of it all. DeepSeek has released a major upgrade to their AI model. Yes. DeepSeek's newly updated model. Called? DeepSeek V30324. Okay.

seems to offer substantial performance enhancements. In? Particularly coding. Oh, so what's new? Developers are reporting it can generate much larger amounts of error-free code. Okay, that's great for coders. It is. But is there anything negative about this new model? Well, some have observed that the model's output has a more robotic tone.

Interesting. So it kind of highlights the ongoing balancing act in AI development. Between... Achieving high performance and maintaining a natural and engaging interaction. I see. So it suggests that DeepSeq may have prioritized functional improvements. At the expense of a more natural language generation. Right. But they did change their licensing to MIT license. Yes.

which is generally considered quite permissive. And open source friendly. Exactly. So that's a positive. Yeah, that could encourage wider adoption. By researchers and developers. Exactly. Leading to more innovation. Precisely. And how big is this model? It's quite large. Okay. It's 641 gigabytes. Wow. Or 352 gigabytes. Wow. With quantization. Quantization. It's a technique that reduces its memory footprint. Oh, okay. But what's remarkable is...

A developer was able to demonstrate it running at a pretty fast speed on regular consumer grade computer hardware. Over 20 tokens per second on regular hardware. That's really impressive. It is quite impressive. And it's available for testing on the OpenRooter platform. Yes. So developers can easily try it out. That's right. And it can handle creative prompts. Yes. On OpenRooter. Yeah. So it shows that it's not just for coding.

So this release really highlights the... Certainly ambitious. So what does Perplexity plan to do if they acquire TikTok? Well, they've laid out some pretty major plans. They want to rebuild TikTok's algorithm with a focus on transparency. Okay. Open sourcing it. Right. Ensuring American oversight. Okay. And integrating their own AI search technology. So they're really committed to transparency? Yeah, open sourcing the algorithm is a big part of that. Okay. They also want to integrate AI-powered technology.

citation capabilities. Directly into TikTok videos. Exactly. So users can verify information in real time. That's the idea. That's a huge step towards combating misinformation. Yeah, absolutely. So what about the technical side of things? Well, they're proposing to use NVIDIA's Dynamo technology

to improve TikTok's recommendation engine. How would that help? Well, the goal is to achieve significant improvements in speed and scalability. Okay. And their vision also includes this cross-platform integration. Where? TikTok videos would appear in perplexity search results. Okay. And perplexity's engine would power search within TikTok. So they're really going all in. They are.

It's pretty amazing to see a startup like Perplexity making such a bold move. It is. But will this bid actually be successful? Well, the deadline is approaching quickly, so only time will tell. Okay. Well, before we wrap up. Yes. We want to take a moment to ask for your support. Of course. If you enjoy these deep dives. Yeah. And you want to help us keep this podcast free. Free for everyone. Please consider making a donation. Every little bit helps. You can find donation links in the show notes. Thank you.

Now let's quickly go over some other noteworthy AI developments. From March 23rd, 2025. Anthropic introduced a new think tool. For their Clawed AI model. Designed to enhance structured reasoning. For complex tasks. There were also reports of a potential partnership. Between OpenAI and Meta. With India's Reliance Industries. Which could lead to a price reduction for ChatGPT in India. That would be huge. It would.

We also saw comments from Kai-Fu Lee, a prominent figure in AI, about Sam Altman and his startup Zero One AI shifting towards deep-seek models. And there were glimpses into Apple's internal research and development for camera-equipped Apple Watches with AI visual intelligence aiming for a 2027 release. Zapier also launched its MCP protocol.

which aims to simplify how AI assistants interact with thousands of apps, over 8,000 to be exact, and there were studies exploring a potential link between increased chat GPT usage and feelings of loneliness and emotional dependence.

Which raises some interesting questions. About the social and psychological considerations of AI adoption. It does. And lastly, browser use secured significant seed funding. For its AI-friendly website navigation solutions. That's right. A very busy day for AI innovation. It was. Yeah. So what

does all of this mean for you? That's the big question. We've seen AI becoming increasingly integrated into our lives. From healthcare to our devices to our cities and even government. The speed of innovation is remarkable. Presenting us with exciting possibilities and some big questions. About responsible development and deployment of this powerful technology.

We've only scratched the surface here. Each of these advancements could be explored in much greater detail. So what stood out to you? What was the most impactful trend or maybe the most concerning one? What questions are these rapid advancements raising in your own mind?

We'd love to hear your thoughts. And if you're looking to reach a highly engaged and professional audience that is keenly interested in technology and innovation, we invite you to spread the word about your business or service by advertising right here on AI Unraveled. Reach thousands of professionals seeking to stay informed about the latest advancements. Reach out to us to learn more about our affordable and effective advertising opportunities. And thank you for joining us on this deep dive into the

ever-evolving world of AI. We hope you found it valuable. And we encourage you to continue exploring these fascinating topics. And forming your own informed perspectives. On the future shaped by AI. See you in the next episode. Goodbye. Bye.