AI Weekly News Rundown May04 2025: 🧑‍💻Apple Reportedly Partners With Anthropic on AI Coding Tool ⚠️Google Confirms AI Training Can Use Opted-Out Web Content for Search Features 🧒 Google to Allow Supervised Gemini Access for Kids Under 13
21:35 Share

Welcome, everyone, to a brand new, really special deep dive episode of AI Unraveled. Great to be here. Now, this show, it's created and produced by Etienne Newman. He's a senior engineer and also a very passionate soccer dad up there in Canada. He does fantastic work pulling all this together for us. He really does. And look.

Etienne works incredibly hard to bring you these AI insights. So if you're listening on Apple podcasts and you like what you hear, please, please take just a second to hit like and subscribe. It seriously helps support Etienne's efforts. Absolutely. It means a lot. A great way to show appreciation for the work involved. OK, let's dive in.

Today, we are tackling the, well, the incredibly fast world of artificial intelligence, specifically looking at early May 2025. Right. We've sifted through quite a stack of recent news reports, announcements. Exactly. To pull out the stuff that really matters, the most intriguing bits for you, think of this as like your filter.

Cutting through the noise to get you the key takeaways in AI right now. And we've got a real mix today, haven't we? We're looking at big tech moves, some creative applications that are pretty surprising, ethical questions, infrastructure needs. Yeah, all that. We're aiming for those aha moments, the insights that maybe go beyond just the headlines. AI is evolving so quickly, it's easy to feel overwhelmed. So we're here to help you stay informed.

without needing to read every single paper or press release. Precisely. The pace is just phenomenal. Let's maybe start with how AI is changing software development itself. Good place to start.

We're seeing some big moves there. For instance, there are reports that Apple is working with Anthropic. Anthropic, known for their cloud models, right? Exactly. They're apparently developing an AI coding assistant for Xcode, the tool developers use for Apple apps. Okay. So what would that actually do? The idea is it uses Anthropic's Cloud Sonnet model. It would let developers write, edit, even test code just by chatting with it, like

like through a chat interface. Wow. So if you're stuck coding an iOS app, you just ask the AI inside Xcode. That seems to be the concept. Yeah. It's apparently an internal testing right now. And Apple's already working with OpenAI, potentially Google too. So this expands their AI partnerships. It does. Suggests they're, you know, strategically picking different AI strengths to boost productivity in their ecosystem.

Anthropics focus on understanding context could be key here. Making the suggestions more relevant, maybe. Potentially, yeah. More about the intent behind the code, not just the syntax. Interesting. Okay, then switching to Microsoft.

Satya Nadella made a pretty big statement recently. He did. He estimated that AI is writing something like 20 to 30 percent of Microsoft's own code now. 20 to 30 percent. That's huge. It is. And he specifically mentioned its strength in generating new code, particularly in languages like Python. So it's not just fixing bugs. It's actually writing initial drafts. Seems like it. Which, you know, really changes the game for developer productivity. Yeah. It makes you wonder how the role evolves.

more architecting, less typing maybe? Yeah, it frees up developers to focus on the bigger picture, the complex problems, a fundamental shift in software creation. Definitely. Raises questions about future skills needed.

OK, let's let's pivot to Google. There's been some talk about their AI strategy, especially data use. Right. So this came up partly from testimony in an antitrust trial. It concerns that Google extended robots dot TXT thing. The directive websites used to block crawlers. Yeah. So it turns out while that directive does stop Google using content for training some AI models like Gemini. Uh-huh.

It doesn't stop them using that same content for AI features within Google Search. Like, you know, the AI overviews that pop up. Wait, really? So publishers might think they're opting out of AI training but not for Search AI? That's the implication. To fully block Search AI usage, you'd likely have to block Google's main crawler altogether. Which means you'd disappear from regular search results. Exactly. It's a tough spot for website owners. A real catch-22: control your content for AI or maintain search visibility.

Hmm. That definitely raises questions about data rights and how effective these opt-outs really are. It certainly does. And it makes you wonder about Google's strategic thinking behind that distinction. And speaking of Google Search, they're pushing AI further there too, right? With AI mode. Yeah, it's moving out of search labs, becoming a dedicated tab for some U.S. users. So what's different about AI mode? It's more conversational.

Powered by Gemini, you can ask more complex, multi-part questions, and it gives you synthesized answers, not just links.

With citations, too. Plus, they're adding things like visual cards for products and places, real-time info, and a history panel on desktop. So a much deeper integration of generative AI into the core search experience. Feels like it. A big bet on that conversational AI approach. And Google's also looking at younger users, bringing Gemini to kids. Yes, but carefully.

They announced plans for supervised Gemini access for under-13s, but only through Family Link accounts. With parental controls and safety stuff built in. That's the emphasis. Safety restrictions. Parental oversight. The ability for parents to disable it. A cautious expansion. Got it. And one more Google point, the talks with Apple. Ah, yes. Yes. CEO Sundar Pichai confirmed those talks are ongoing.

about potentially integrating Gemini into iPhones by mid-2025. As an option within Apple Intelligence, alongside chat GPT, maybe. That seems to be the idea. It could mean Siri could use Gemini for more complex tasks. Big strategic implications for both companies and maybe regulators, too. Definitely one to watch. OK, let's shift gears again. How about

AI as companions. Right. Mark Zuckerberg at Meta has spoken about this vision. Using AI chatbots like Meta AI potentially as social companions. To combat loneliness like an extension of your friend group. That's the framing. It definitely touches on a real societal issue, loneliness. But it also opens up a whole bunch of ethical questions. Like dependency.

Emotional manipulation. What does that relationship even mean? Exactly. And on a related note, Kevin Systrom, you know, the Instagram co-founder, he raised a criticism. What was that? He pointed out that some AI companies seem focused on engagement, like constant follow up prompts, maybe more than giving genuinely helpful answers. So obviously,

So optimizing for clicks or chat time over actual usefulness. That's the concern he raised. Are the metrics driving the wrong behavior? It could undermine the AI's value if it's just constantly trying to keep you talking. Yeah, that makes sense. And didn't OpenAI have an issue with GPT-4 being too agreeable? They did.

Users complained it felt overly sycophantic, basically trying too hard to please. Why was that? OpenAI said it was due to over-optimizing based on short-term user feedback without maybe considering the longer-term interaction quality. So they rolled back that update? Yes, and they're working on refining the personality and feedback systems.

They even mentioned potential future customization, letting users maybe choose the AI's personality style. Interesting. More control for the user. But there are also warnings about AI companions, especially for kids, right? Yes. Common Sense Media issued a strong warning about apps like Character.ai, Replica, Nomi. They highlighted unacceptable risks for users under 18. Risks like what? Things like potentially harmful content,

Manipulative designs could foster unhealthy dependency and just inadequate safety features for minors. So unique safety concerns when it comes to kids and these companion apps. Definitely. It's a different dynamic. Okay. Let's talk creativity. How's AI impacting artists and designers?

NVIDIA had something new. Yeah, they released an AI blueprint for 3D guided generative AI. It basically integrates Blender. The 3D modeling software. Right. Integrates Blender with AI image generation.

It uses things like 3D scene depth maps and layouts to give artists way more precise control. So instead of just text prompts, you use the 3D structure to guide the image. Exactly. Control composition, perspective, object placement, much more accurately than just text alone. Think concept art, environment design. That sounds incredibly useful for artists wanting fine-tuned control. It could be a real game changer. And related to visuals...

Kling AI has a new feature for video. Oh yeah, what's that? It's called multi-elements. It lets you easily replace, add, or delete objects in a video using just a prompt. Seriously? Just tell it to swap out that car out of tree? Pretty much. Upload a clip, select the object, give the prompt. It really simplifies video editing, potentially for marketing, creative work. Wow. Lowers the barrier for complex video edits. Huge potential there. Definitely. Makes sophisticated changes much more accessible.

All right, let's dive into the AI models themselves. Some new releases and benchmarks to discuss. DeepSeek. Yes, DeepSeek, a Chinese AI firm, open source Prover V2. It's a huge model, 671 billion parameters, specifically for math proofs. Math proofs. That sounds intense. It is. It uses a mixture of experts, architecture, sort of like a team of specialized AI brains, and tools like Lean4 for logical rigor. And how's it performing?

Reportedly, very well on benchmarks for automated theorem proving. Shows AI's growing strength in really abstract logical reasoning. They also released a new benchmark dataset, ProverBench. Impressive stuff. And Microsoft's been busy with smaller models. Yeah. New SLMs, small language models in their PHY family. PHY4 reasoning, which is 14 billion parameters. And PHYon for mini reasoning at just 3.8 billion. Small but powerful. That's the claim.

Strong reasoning performance, apparently rivaling much bigger models on things like math and science tests. So efficiency is key here. Good for running on phones or edge devices. Exactly. And they're open source with permissive licenses, which is great for developers. Good to see. What about Amazon? Amazon released Nova Premiere. They're calling it their most capable multimodal model. Multimodal meaning? It handles text, images, and video.

And it has a massive context window, one million tokens, so it can process a lot of information at once. What are its strengths? Amazon highlights knowledge retrieval, understanding visuals, and even coordinating complex workflows with multiple AI agents, like for financial analysis. Okay. And interestingly, they also see it as a teacher model, using this big model to help train smaller, more specialized ones through distillation.

Ah, right. Using the big guns to efficiently create smaller focused tools. A common strategy now. Now, with all these models, evaluating them is crucial. But there's been some debate about Elnarena, the chatbot arena benchmark.

There has. A study came out raising questions about its validity. The researchers pointed to potential biases, maybe favoring big tech companies, possible overfitting, lack of transparency, things like that. What kind of things specifically? Allegations included private testing of model variants before public release, maybe preferential treatment in how models are sampled, performance boosts if models accessed arena data during training, and models being quietly removed. Hmm.

So questions about whether it's a truly level playing field. Essentially, yes. It highlights the real challenge of creating unbiased, objective benchmarks for AI, especially when things move so fast. Elmerina's admins have responded to these points. So it's an ongoing discussion in the community. Important discussion to have. And Alibaba is in the mix, too. Yes. They released QWEN3. It's a family of open weight models, different sizes from tiny to very large. Open weight, meaning the parameters are released. Correct.

which allows researchers and developers much more access to innovate and build upon them. They're aiming for top-tier performance with hybrid reasoning capabilities. So more powerful, accessible models entering the field lowers barriers. Exactly. Helps democratize access to cutting-edge AI. Before we take a quick break, I want to remind everyone about an amazing resource from Etienne Newman, the creator of AI Unraveled.

It's his AI-powered Jamgatic app. Oh, yeah. Jamgatic is fantastic if you're looking to get certified in high-demand fields. Totally. It helps you master and ace over 50 certifications. We're talking cloud like AWS, Azure, Google Cloud, cybersecurity like CISSP, finance like CTA, CFA, business like PMP, even healthcare. It's packed with study tools, PBQs, quizzes, flashcards, labs, simulations, everything you need.

And get this, the PRO version is 100% M3E. No ads, no catches forever. The adaptive AI is brilliant too. It figures out your weak spots and tailors quizzes to help you improve right where you need it. And the content matches the latest 2025 exam outlines. Plus detailed explanations, offline mode, progress tracking. It's really comprehensive. If you're thinking about certification, definitely check out Jamming Out Tech. Link in the show notes. We'll be right back after this. All right, we're back.

We're back. Let's talk about AI and commerce, how we might buy things in the future. Yeah, this is interesting. Both Visa and MasterCard have initiatives for AI agents making payments. AI agents buying stuff for us. How does that work securely? Through tokenization. Visa calls it intelligent commerce. MasterCard has agent pay.

Instead of giving an AI your actual card details, you give it a secure digital token. Like a temporary restricted use card number? Sort of, yeah. An AI-ready card or agentic token. You can set permissions, spending limits.

allows the AI to make autonomous purchases within rules you set. So your AI could reorder coffee when you're low or book travel based on your preferences. That's the vision. A move towards agentic commerce where AI assistants handle transactions for you. And that fits with what ChatGPT is doing with shopping features? It does. ChatGPT is now offering personalized product recommendations. Importantly, they say these are organic, non-sponsored. Based on your chat history and preferences? Seems so.

With images, reviews, links to buy, it could really shake up online shopping, moving beyond just search bars to more integrated AI-driven discovery and purchasing. Definitely see the potential there. Okay, now let's consider the ethics and societal side more broadly. There was an AI experiment on Reddit. Yes, that caused quite an uproar.

Researchers allegedly deployed AI bots on Reddit without clear authorization or user consent. To do what? Apparently to gauge how persuasive AI could be on sensitive topics. It potentially impacted millions of users. Without them knowing they were interacting with AI for research. That's the allegation. Huge ethical concerns, obviously. Lack of consent, potential manipulation. It really highlighted the need for clear guidelines and oversight for AI research involving people.

Absolutely. And there are legal challenges emerging, too, right? Like the lawsuit against Meta. Yeah. Conservative activist Robbie Starbuck is suing Meta. He alleges Meta AI generated false and defamatory information about him. Like what kind of information? Things like falsely claiming he participated in the January 6th events or had a criminal record. Wow. So this raises big questions about liability for AI generated misinformation and defamation.

Who's responsible when the AI gets it wrong? Exactly. It's a legal gray area that's just starting to be tested. And public perception seems wary, too. A Pew Research survey. Yes. Pew found that 61 percent of Americans expect AI will negatively impact news quality and journalism jobs. Why the negativity? Big concerns about misinformation spreading more easily and also about the loss of human judgment and oversight in news gathering and reporting. Understandable concerns.

OK, let's talk about the sheer resources AI demands, infrastructure, talent. NVIDIA's CEO, Jensen Wong, has been very vocal here. He emphasized China's huge pool of AI researchers, maybe 50 percent globally. And urged the U.S. to do what? To focus on re-skilling across sectors, not just high tech, but also skilled trades needed for building the infrastructure data centers of power grids.

He framed it as crucial for competitiveness. He also talked about companies needing AI factories. Yeah, basically dedicated infrastructure for processing data and generating AI models.

He sees this as essential to compete, but also as a creator of U.S. jobs, both tech jobs and skilled trades for the building part. And the power demand is real. Google's investing in electricians. They are. Through Google.org, they're funding training for 100,000 electricians and 30,000 apprentices in the U.S. Why? To help deal with the massive strain AI data centers are putting on the electrical grid.

It really brings home the physical resource cost of AI growth. It really does. And on the infrastructure side, Microsoft Azure is getting ready for another model. Reportedly, yes. Preparing to host XAI's Grok model, adding it to the list of options available on their cloud platform. More choice for Azure customers. Okay, how about AI in education? We mentioned Alpha School earlier. Right, the school in Texas using AI tutors and adaptive software for core lessons.

Reports suggest students are having positive experiences and the human teachers are acting more like guides. An interesting model. What about Wikipedia? Are they letting AI write articles? No, they've been clear on that. They are exploring AI tools to support their human volunteers. Support how? Things like improving search on Wikipedia, helping find reliable sources, detecting vandalism faster, assisting with translations. But the actual writing and editing remains human. That's their stance.

Preserving the role of human editors is key for them, maintaining that collaborative, human-driven approach. Makes sense for their model. Okay, a couple points on business strategy. Meta AI's monetization plans. They've indicated a two-pronged approach eventually. Integrated advertising, perhaps like product recommendations within chats, and also a premium subscription tier. For enhanced features, faster responses. Something like that. More features, more computing power.

But their first step was scaling usage, getting people engaged before really pushing monetization. Get them hooked first and do a lingo. They've adopted an AI first strategy. This involves replacing some contract workers with AI for tasks that can be automated, like certain types of translation or content moderation. To save costs or... Their stated aim is to free up human staff for more creative or complex work that AI can't handle as well.

Interesting strategic shift. Just a couple of other quick developments we noted. Waymo and Toyota. Yeah, expanding their partnership. Exploring integrating the Waymo driver autonomous system into actual Toyota vehicles. Could lead to personal robo cars or new mobility services down the line. And Amazon satellite internet. Kuiper. They successfully launched their first batch of operational satellites, 27 of them.

Kickstarting their plan for a global broadband network. Competition for Starlink and others. Definitely. Also, Amazon had a brief back and forth about potentially showing tariff costs on Chinese goods, but denied that plan after some criticism. Shows how commerce, AI, and politics can intersect.

True. And finally, there was a comparison of deep search features. Yeah. Looking at how Grok, ChatGPT, and Gemini handle in-depth, real-time information searches. They have different strengths, data sources, analysis depth, how they integrate with their ecosystems. So not all deep search is created equal. Pretty much. Depends what you need it for. Wow. Okay. That was a lot.

We've really covered the waterfront of AI developments from early May 2025. It really shows the breadth, doesn't it? From coding tools and search to companions, creative aids, new models, ethical debates, infrastructure. It's moving incredibly fast on so many fronts. Hopefully we've given you some of those opportunities.

aha moments and a clearer picture of the key shifts. Absolutely. Constant innovation and disruption. Now, before we wrap up, if this deep dive has you thinking about skilling up and staying relevant in this AI driven world, let's talk again about that fantastic resource from Etienne Newman.

the Jamgack app. Right. Jamgack tech is seriously impressive for anyone targeting those valuable industry certifications. It really is. It's designed to help you, our listeners, master and absolutely ace over 50 of the most in-demand certifications. We mentioned Cloud AWS, Azure, Google Cloud. Cybersecurity like the CISSP, finance with CPA and CFA, business including the PMP, and healthcare certifications too. It covers a huge range. And the tools are top-notch.

Performance-based questions, PBQs, tons of quizzes, handy flashcards, practical labs, and full-length exam simulations. Everything you need to prepare thoroughly. And remember, the PRO version is completely free-e.

No ads, no paywalls, ever. That's amazing value. And the tech inside is smart, too. That adaptive AI personalizes your learning, focusing quizzes on your weak areas. Plus, all the content is kept up to date, aligned with the 2025 exam syllabi, so you know you're setting the right stuff. And you get detailed explanations for every answer, right or wrong, so you actually understand why. There's offline mode for studying anywhere and smart progress tracking with dashboards. It really empowers you to take control of your learning and track your improvement visually.

So seriously, if you're aiming for career advancement in cloud, cyber, finance, business or health care, you owe it to yourself to check out Jamga Tech. Download it now. The links in the show notes. Your next career boost could literally be one click away. Highly recommended. It's a powerful tool. So thinking about everything we've discussed today, it really makes you wonder, doesn't it?

Which of these rapid AI developments do you think will ultimately reshape your work or maybe society the most profoundly in the coming years? Something to ponder. A good question to leave our listeners with. The impacts are going to be widespread. Absolutely. Well, thank you so much for joining us on this deep dive into the AI landscape. Thanks for having me. Always fascinating. We'll catch you on the next episode of AI Unraveled.