AI Daily News May 13 2025: 🤳AI Analyzes Face Photos to Predict Biological Age and Cancer Outcomes👨‍💻Google's Jeff Dean: AI at Junior Engineer Level Within a Year 🔋Apple Developing AI-Powered Battery Management for iOS 19
13:36 Share

Welcome to a new deep dive here on the deep dive. You've shared some really fascinating material with us today. It's quite a collection, isn't it? Yeah, all focused on the incredible momentum we're seeing right now in artificial intelligence. Just a whirlwind of advancements. We've got a collection of reports detailing some of the most significant AI developments that have, well, just come out recently. That's right. Fresh off the press, so to speak. And our mission today is...

you know, to cut through the hype a bit. Get to the core of it. Exactly. Extract the core insights, those aha moments that really tell us where things are heading. And crucially, what it all means for you, the listener. Absolutely.

By the way, this is a new episode of the podcast AI Unraveled. Created and produced by Etienne Newman. That's right. Senior engineer and passionate soccer dad from Canada. And if you're enjoying these deep dives. Please remember to like and subscribe on Apple. It really helps us out. It does indeed. Okay, let's dive in. Right. Where should we start? It's striking how these, well...

seemingly disparate news items actually paint a cohesive picture. Yeah, it really feels like like an A.I. revolution unfolding across multiple fronts all at once. Not just small steps anymore, is it? No, it feels like foundational shifts, how we build software, how we interact with devices, health care. Even how nations are thinking strategically about A.I. Precisely. OK, so speaking of building software.

That seems like it's right on the cusp of a huge change. You mean the Google prediction? Yeah. Google's chief scientist, Jeff Dean, he made a pretty significant statement at the AI Ascent 2025 conference. The one about AI junior engineers. Exactly. He predicts that within roughly a year, AI systems could function at the level of a junior software engineer. Which is awesome.

Wow, a year isn't long at all. Right. And he wasn't just talking about, you know, basic code snippets. No, it was more comprehensive. Running tests, debugging. Understanding documentation, even learning from senior engineers, using standard tools. And potentially working 24-7. That part's key, too.

I mean, think about the acceleration that could unleash. Dean did acknowledge current limits, though. Sure. He mentioned the path forward likely involves reinforcement learning. Right. The AI learning through trial and error, essentially. And just accumulating vast amounts of agent experience. The implication really is a potential paradigm shift in software development. A massive force multiplier, as they say. Could be.

freeing up human engineers for maybe more complex creative work. But it does raise some questions, doesn't it? Like how do new people enter the field? That's a very important point. If AI handles those entry-level tasks. What does that mean for training?

For onboarding. The industry might really need to adapt. Maybe the focus shifts? Shifts how? Well, perhaps towards skills AI can't easily replicate. Creative problem solving, system design, overseeing the AI itself. So more emphasis on uniquely human skills and continuous learning. Absolutely. Which actually, speaking of skills and learning, have you heard about Jamgit Tech? Tell me more.

It's an AI powered app also created by Etienne Newman, who produces AI Unraveled. Oh, interesting. Yeah, it's designed to help people master over 50 different in-demand certifications. Like what kind of fields? Things like cloud computing, cybersecurity, finance, business.

It uses AI to personalize the learning, quizzes, performance questions, labs. Sounds incredibly relevant, especially given these potential shifts we're talking about. Definitely worth checking out if you're looking to upskill. Good tip. Okay, so moving from software to something that honestly feels like science fiction. The brain-computer interfaces. Yes.

Apple is reportedly looking into BCI technology. Allowing users to control devices with neural signals? Mind control, almost. Well, maybe not quite mind control, but the goal seems to be enhancing accessibility. Right. Specifically for people with severe motor impairments. That's the primary aim mentioned. And they're reportedly working with a neurotech startup, Synchron. Yes. And their Stentrode device, which is fascinating in itself. How does that work again?

It's implanted via a blood vessel, a vein, and it detects thought patterns from the motor cortex. Less invasive than some other approaches. Wow. And Apple might release a specific interface standard for BCI devices later this year. That's the report. And integrating it into their existing switch control accessibility framework. It suggests a much bigger long-term vision, doesn't it?

Absolutely. Revolutionizing human computer interaction. Potentially creating a whole ecosystem. Incredible potential for accessibility. Now, something maybe a bit more immediate, everyday AI. The battery management in iOS 19. Yeah, the Bloomberg report.

Apple developing an advanced AI system for that. The core idea is optimizing battery life by learning user habits. And proactively adjusting power consumption based on those habits. It ties into their whole Apple intelligence strategy. And could even enable, maybe, slimmer iPhone designs without sacrificing battery life. It's a great example of on-device AI becoming more sophisticated.

How does it actually work, though? Well, it observes your patterns when you charge, when you use certain apps, uses that data to train the AI. And then makes adaptive changes to power use behind the scenes. Exactly. Plus, there's that new lock screen indicator for estimated recharge time mentioned. So AI working quietly to make the phone more efficient. Precisely. Enhancing user experience without you needing to do anything. Okay, so from personal devices to the geopolitical stage.

Saudi Arabia. Right. They're making some very significant investments in AI infrastructure. Part of their national AI strategy, diversifying beyond oil. And they're engaging multiple chip suppliers, which is interesting. Not just relying on one source. No. They've selected Grok, known for their LPUs for AI inference tasks. Inference. So running the trained AI models. Correct. And they're planning a data center expansion in Damem with Grok. But they're still buying training chips from NVIDIA.

Seems so. They're covering both basis training and inference. And they launched a new AI company too, Humane. Yes. Backed by their sovereign wealth fund, it's set to leverage NVIDIA's platforms to deliver AI services, build data centers, develop models. So it's a multi-pronged strategy. Diversify hardware, build national capability. Position themselves as a global AI hub. It's a serious commitment. It really shows how critical AI infrastructure is becoming globally. Absolutely.

Now speaking of big players making moves,

Google search. Ah, yes, the potential changes to the homepage. They're reportedly experimenting with making AI mode more prominent. Even potentially replacing the I'm feeling lucky button. That's iconic. It is. Some tests show it replacing that button. Others show it sitting next to the main search button, maybe with a rainbow border to highlight it. Part of a bigger push to expand AI mode availability, right? Though it's still limited now. Yeah, currently in their experimental labs, small percentage of US users. But what does that signal?

A major shift in their search strategy. It certainly could be. Underscoring their commitment to deeper AI integration. Guiding users towards more conversational AI-driven ways to find information. That seems to be the direction. Moving beyond just keywords. Interesting times for search. Okay, shifting gears completely now, healthcare. The FaceAge tool. Yes, developed by researchers at Mass General Brigham.

an AI that estimates biological age from a face photo. - And the striking part is the correlation with health outcomes. - Right, the study in the Lancet Digital Health found this face age often differs from chronological age. - And it's a significant predictor of survival outcomes, particularly in cancer patients. - That's quite something. It could help clinicians predict short-term life expectancy, especially for palliative care. - That's the potential. It provides a non-invasive biomarker trained on

What, tens of thousands of photos? Yeah, analyzing subtle facial characteristics, they found cancer patients looked about five years older on average biologically. And higher face age correlated with worse survival. It even improved physician accuracy when they used the face age scores. And there was a correlation with a gene linked to cellular aging, too. So it seems to be tapping into something genuinely physiological. The implications are huge, aren't they? Non-invasive health checks from photos.

Insights into resilience. Potentially tailoring treatments based on this predicted biological age. It's fascinating. Truly remarkable. Okay, another research-led development.

Sakana AI in Tokyo. Continuous thought machines or CTMs. A completely different neural network architecture. Yes. The key idea is that they process information and think step by step over time. Unlike traditional AI that often does one big pass. Exactly. CTMs have an internal self-generated timeline.

It's inspired by how biological brains synchronize neural activity for reasoning. And they demonstrated this. Yeah, with things like complex mazes where the CTM visibly traces potential paths. You can sort of see it thinking. Kind of. And image recognition where it spends more time, more internal steps on difficult parts of the image. So the potential is more flexible AI, more adaptable, maybe easier to understand. That's the hope.

Better for tasks needing iterative reasoning, planning, that kind of nuanced understanding. Very cool concept. Now, something maybe more immediately practical for many listeners, AI tools for video. Oh, yeah. Turning video libraries into computers.

Content gold mines, as the source put it. Repurposing video content seems to be getting much easier. Definitely. These tools can automatically transcribe, generate summaries, pinpoint key moments. Convert video scripts into blog posts, articles, social media snippets. It just extends the lifespan and reach of your original video work so much. The example given used Notebook LM.

Right? Yeah. Uploading video, getting a transcript, then using prompts to refine it. Yeah, a pretty straightforward workflow. It democratizes content creation and marketing, really. Saves a ton of time and resources, maximizes the value you already have. And speaking of maximizing value and resources, especially when learning new skills. Let me guess, Jenga tech again. You got it.

Etienne Newman's app. If you're dealing with complex information like maybe prepping for certifications inspired by all this AI talk. It's designed for that, right? Cloud, cyber, finance, business, healthcare. Over 50 certs with AI helping you learn via performance-based questions, quizzes, labs, simulations. It definitely sounds like a powerful tool in today's fast-moving tech landscape. For sure. Okay.

OK, one more major development, OpenAI's release of HealthBench. Ah, yes. An open source benchmark for evaluating large language models in health care. Developed with input from over 260 physicians, which adds a lot of credibility. Absolutely. And it's composed of like 5,000 conversational scenarios, multi-turn, multilingual. Simulating real health care interactions.

And a detailed rubric with tens of thousands of criteria. Over 48,000, yeah. Looking at clinical accuracy, communication quality, context. And the results show newer models are improving significantly. Like OpenAI's latest scoring, much higher than GPT 3.5 Turbo. Big jump. O3 hit 60% accuracy compared to 16% for 3.5 Turbo on their scale. And even smaller, cheaper models are getting much better.

Right. GPT 4.1 nano outperformed older, larger models while being more efficient. And OpenAI is open source, the evaluations and the data set itself. Which is crucial for transparency and allowing others to build on it. So the significance here is ensuring safer, more effective AI in health care. Exactly. A standardized way to test, promoting transparency, guiding development towards more dependable tools.

Yeah, it definitely feels like a necessary step. Okay, just to quickly round things off, a few other headlines caught my eye. It's been a busy period. Google DeepMind launched an AI futures fund. SoftBank's big $100B OpenAI Stargate investment might be facing delays. Saw that, yeah. Funding massive projects is complex. Perplexity AI is reportedly set to raise another huge round, $500 million. They're growing fast in the AI search space.

Carnegie Mellon researchers publish work on Lego GPT. Sounds fun. Combining AI and Lego. Intriguing. Saudi Arabia.

Officially unveiled Humane, which we discussed. Right. And the US FDA plans to deploy AI across the agency by the end of June. AI adoption speeding up everywhere. Wow. Okay. So as you can clearly see, the landscape of AI is just incredibly dynamic. Rapid advancements across so many different sectors. Software, personal devices, healthcare, search. Hopefully this deep dive has given you, our listener, a clearer picture of some of the most important developments. And important.

and importantly, their potential impact down the road. It really makes you wonder, doesn't it? Considering this pace,

What aspects of our daily lives do you think will be most fundamentally changed by AI in, say, the next few years? That's the big question to ponder. And hey, if any of these advancements have sparked your interest in maybe diving deeper into a new field or getting certified. Remember Etienne's AI-powered Jamgatech app. Exactly. It's there to guide you. A fantastic resource from someone clearly passionate about AI and empowering people with knowledge.

Well put. Thanks for diving deep with us today. Thanks for joining us.