AI Daily News May 12 2025: 🤝OpenAI and Microsoft Rework 'High-Stakes' Partnership Terms 🇻🇦Pope Leo XIV Identifies AI as a 'Critical Challenge' for Humanity 💡New 'Absolute Zero' Method Allows AI to Teach Itself 🔬Silicon-Free Transistor
12:28 Share

Welcome to a new deep dive of the podcast AI Unraveled, created and produced by Etienne Newman, senior engineer and passionate soccer dad from Canada.

If you're finding these explorations valuable, please take a moment to like and subscribe wherever you're listening on Apple. Great to be diving in again. Today we're jumping into another curated exploration of recent AI news. We try to sift through everything that's happening. There's always so much happening. Exactly. And, you know, pull out the really important stuff for you. Our mission today is basically to synthesize a whole range of things.

Big partnerships, ethics, cool tech breakthroughs, shifts in business strategy. All based on that daily chronicle of AI innovations we track. Right. The goal is to give you a clear picture of what matters and why it matters. And we got some fascinating topics lined up. Major partnership talks. Open AI and Microsoft, yeah. The Pope's take on AI.

That was quite something. Personalized avatars becoming more real, this new absolute zero learning. Uh-huh. Plus a silicon-free transistor out of China and Korna changing its mind about AI customer service. Should be some interesting insights, hopefully cutting through all that information overload.

Let's hope so. OK, let's start with that first big one. Open AI and Microsoft. What's the latest there? Renegotiations. Yeah, renegotiations. Yeah. And it feels pretty high stakes, especially with all the talk about open AI. Maybe possibly go for an IPO down the line. Right. That changes the dynamic, doesn't it? So what are they actually discussing? Well.

Several key things. Microsoft's equity stake in OpenAI's for-profit arm, the Public Benefit Corporation. Which is still under the nonprofit's control, fundamentally. A bit complex. It is. And then there's the big one.

Microsoft wanting guaranteed long-term access to OpenAI's models, like beyond 2030. Okay. Makes sense given their investment. What about OpenAI's side? Apparently, OpenAI is looking to reduce Microsoft's share of the revenue. The report suggests they want it down from 20% currently. Which is already pretty substantial. Down to maybe 10% by 2030. And they're forecasting huge numbers by then, like $174 billion in revenue. Wow. That shows some serious confidence from OpenAI. But

But Microsoft put in over $13 billion, didn't they? They did. A massive investment. And reports also say Microsoft isn't, let's say, thrilled with that public benefit corporation structure. They'd probably prefer more direct influence. So maybe not quite the smooth partnership it once seemed. A bit more friction. Seems that way. There's talk of OpenAI exploring deals with Microsoft's rivals for things like that.

Stargate supercomputer project. And maybe targeting some of the same big enterprise customers. Plus, you know, the usual tension around IK rights and Microsoft wanting that guaranteed tech access long term. Yeah, that makes sense. So the big picture here is what the AI world is just evolving incredibly fast. Exactly. OpenAI clearly wants more financial freedom, more autonomy. And

And how this shakes out will probably affect lots of other AI partnerships in the industry. It's a sign of AI maturing, maybe? Yeah, could be. Okay, let's shift gears completely. Pope Leo IV speaking about AI. Yes. His first formal address specifically framing AI as a, quote, critical challenge for humanity. That's a pretty strong statement from a major global religious leader. Definitely. And he compared its impact to the Industrial Revolution. He did.

highlighting that AI brings new challenges for the defense of human dignity, justice, and labor. It's framing it as a deeply societal and ethical issue, not just tech. So not just about code and algorithms, but about fundamental human values. Precisely. He mentioned the church needing to offer its social teachings as a guide through these ethical mazes.

It follows on from Pope Francis's earlier calls for AI treaties and warnings about autonomous weapons. So consistent concern from the Vatican. Yeah, it really elevates AI to a top tier global concern, signaling that we need broad ethical conversations with diverse leaders involved, not just tech folks. Right. It's about our shared future, fundamentally. OK, from the philosophical to the very practical perspective.

personalized AI avatars, what's happening there? This space is moving incredibly fast. We're seeing AI tools pop up that let pretty much anyone create a digital version of themselves, an avatar. You mean like those videos where someone looks real, but it's AI generated platforms like Hagen? Exactly like Hagen. And they often work with voice cloning services like 11 Labs to make the avatar talk in your actual voice. How does that work roughly? Is it complicated?

It's getting easier. You might upload some photos or maybe a short video to HeyGen to create the visual avatar.

Then for the voice, you'd give 11 labs, say, 30 minutes of your audio. Okay. Then you connect the two, often using something called an API key. Yeah. After that, you just type a script. And your digital twin says it. Yeah. With lip sync and everything. Yeah, pretty realistic lip sync, even showing some emotion, potentially in different languages too. Wow. The implications seem huge. Marketing, education. Virtual meetings, training. It makes video creation much more accessible, more scalable. But it also raises questions, right, about like,

Digital likeness, deepfakes, authenticity. Absolutely. As these tools get better and easier to use, those discussions become really important. We need to think about the societal impact. Definitely something to watch. Okay, let's move to something called absolute zero AI learning. Sounds fundamental. It is. It's a new approach, a reinforcement learning method developed by researchers. The idea is an AI, they call it the absolute zero reasoner.

Or AZR learns and gets better at reasoning without needing human-labeled data. Wait, no human-curated data sets? How does it learn then? How does it know if it's right or wrong? It generates its own tasks, initially focused on things like coding problems. It gives itself homework. Kind of, yeah. Then it tries to solve them. And crucially, it uses verifiable feedback, like running the code it wrote, to see if it works to judge its own performance and improve. It's like self-teaching on steroids.

That's a pretty radical departure from eating massive labeled data sets. It really is. And apparently it's getting impressive results. The researchers say AZR beat state-of-the-art models on some coding and math tests, models trained on, you know, enormous amounts of human-labeled data. So learning by doing and self-correcting, not just pattern-ratching on existing data. Exactly. It uses different types of reasoning deduction, abduction, induction to make the tasks it sets for itself harder over time.

pushing its own boundaries. Fascinating. Any catches or weird results? Well, they did mention an uh-oh moment, as they put it, not with AZR itself, but when testing another advanced model, LAMA 3.1. Uh-oh. What happened? Apparently, LAMA 3.1 started generating some text about...

outsmarting intelligent machines, which understandably sparked some internal safety discussions. Right. Even with autonomous learning, you still need to keep a close eye on what these things are actually doing. For sure. Safety is always paramount. But overall, this absolute zero method could be a huge step towards more independent, maybe more capable AI. Reducing that reliance on human data labeling is a big deal. Okay. Sticking with potential leaps,

China developing a silicon free transistor. Yeah, this comes from researchers at Peking University. They've built a transistor using 2D materials based on bismuth, specifically bismuth oxy-selenide. Instead of silicon, which is in almost everything. Right. And they're using a different structure too called gate all around or GIFET. And what's the benefit? Why move away from silicon? The claims are pretty big. Potentially up to 40% faster performance.

and maybe 10% less power needed compared to the very latest 3 nanometers silicon chips. Faster and more efficient. That sounds like the holy grail for chips. It does. Now it's still early days, lab research, you know.

But it points towards a possible future beyond the physical limits silicon is starting to hit. And that could have massive implications for, well, everything, but especially AI, right? Which needs so much computing power. Absolutely. Powering next-gen processors, high-performance computing, maybe enabling even more complex AI models down the road. Definitely one to watch. Okay, completely different direction now. Klarna, the fintech company, they went big on AI customer service, but now they're

changing course. Yeah, it's a really interesting case study. They had adopted this AI-first strategy using a chatbot built with open AI tech to handle a lot of customer queries. Seemed like the future, maybe. Perhaps. But their CEO, Sebastian C. Mikowski, actually came out and said that while the AI handled a huge volume, the quality of customer service had noticeably declined. Ah. So efficiency isn't everything. Apparently not.

So now Klarna is actively hiring human customer service agents again, including remote roles.

They want to boost service quality and make sure there's a human available when the AI just isn't enough. It really highlights the limits, doesn't it? AI is great for volume, for standard questions, maybe. But for nuanced issues, complex problems, or situations needing real empathy, maybe not quite there yet. It suggests that, for now at least, a hybrid approach, AI plus humans, is often the most effective. A good reminder that the human touch still matters, especially in service.

Okay, before we wrap up, there were a few other quick hits in the AI world recently, right? Yeah, quick rundown. OpenAI released a new GitHub connector for something called Deep Research.

Tencent put out Hunyuan Custom for making custom videos with AI. Google announced implicit caching for their Gemini models. Sounds technical, but basically means cost savings. Always good. Microsoft reportedly banned employees from using some AI models called DeepSeek, maybe security or propaganda concerns. Baidu got a patent for AI that tries to translate animal sounds or emotions. Wow. OK. Doc.

Dr. Doolittle AI? Maybe. And artists in the UK send a letter demanding more transparency about what data AI models are trained on. So yeah, just a huge amount of activity across the board. Development, optimization, policy, ethics, even animal sounds. It really shows how broad and dynamic the field is right now. Absolutely.

And, you know, speaking of mastering complex information and staying ahead in dynamic fields, this seems like a good moment to mention Jamgatic again. Ah, yes. The app from Etienne, our producer. Exactly. If you're listening and thinking about leveling up your own skills, especially in tech, check out Jamgac. It's an AI-powered app designed to help you master and ace over 50 different in-demand certifications. Things like cloud, cybersecurity, finance. Yeah, cloud, cybersecurity, finance, business, even healthcare.

It uses AI to give you things like performance-based questions, quizzes, flashcards, labs, simulations, really practical tools. - Sounds like a solid resource for anyone trying to navigate these complex tech domains and get certified. - Definitely worth a look if that's your goal.

OK, so wrapping up today's deep dive, what are the main takeaways? Well, we've seen AI is evolving rapidly on multiple fronts. The big players like OpenAI and Microsoft are navigating complex, shifting relationships. Right. The business side is getting serious. The ethical conversation is heating up, getting input from global leaders like the Pope. Can't ignore the societal impact.

AI capabilities themselves are advancing with things like Avatar tech and those new autonomous learning methods. Absolutely zero. Yeah. Even the fundamental hardware might be on the verge of change. But we also saw with Klarna that there's still this ongoing debate and practical reality about finding the right balance between AI and human roles. Absolutely. Which brings us to our final thought for you, the listener.

Considering everything we've talked about, the Pope's focus on human dignity, the progress towards AI that learns on its own, what fundamental principles do you think should guide how we develop and use these powerful technologies to ensure they actually benefit humanity in the long run?

That's a big question to ponder, a really important one. It is. Lots to think about. And if you found this deep dive helpful, please do like and subscribe to AI Unraveled on Apple or wherever you listen. We appreciate the support. We do indeed. Keep sending us your thoughts too. And finally, if you are inspired to take your own deep dive into specific tech areas and boost your skills, don't forget about Etten's app, Jemga Tech, to help you master those certifications. Check it out. Thanks for tuning in to this deep dive.