AI Daily News Rundown April 22 2025: 🧭Anthropic Charts Claude's Values 🧠Hassabis: AI Could End All Disease ⚖️DOJ: Google Could Use AI to Extend Search Monopoly 👀Huawei Prepares New AI Chip as China Looks Beyond Nvidia
13:14 Share

Welcome to a new deep dive from AI Unraveled, created and produced by Etienne Noumen. He's a senior software engineer and a passionate soccer dad up in Canada. Great to be here. And hey, if you're finding these deep dives useful, please do take a second to like and subscribe to AI Unraveled on Apple Podcasts. It genuinely helps us reach more curious minds like yours. It really does. So today we're plunging into the very latest in the super fast moving world of AI.

We're looking at the key stories making waves as of today, April 22nd, 2025. Yeah, there's a lot happening. We'll be unpacking quite a bit new AI chips, some fascinating work on AI values. Right. And even exploring AI, potentially writing laws or maybe even curing diseases, quite

quite the range definitely we've gone through you know a whole load of news research updates the daily AI Chronicle basically to pull out what really matters and what it might mean for well for you ready to jump in let's do it it's pretty amazing how many different areas AI is touching all at once right now isn't it just we're seeing progress like not just in the software the algorithms the actual hardware underneath it all the foundations exactly and then you've got these

Really interesting efforts to understand AI behavior, even guide it alongside some really ambitious, you know, real world uses. Gives you a sense it's really growing up, becoming part of the furniture almost. Maturing. Yeah. And getting woven into, well, everything. OK, let's start with that foundation you mentioned. Our sources are saying Huawei is about to start mass shipments of its new Ascend 910C AI chip.

That's happening next month, May 2025. Why is that such a big deal? Well, the main thing here is that this Ascend 910C chip, it's really positioning itself as a serious domestic option inside China. Right. Challenging NVIDIA. Directly challenging NVIDIA's previous dominance there. And you have to remember the context, right? The U.S. restrictions on NVIDIA selling their top tier processors in China. Oh, OK. That's the driver. It's a huge factor. It's created this

well, this urgent need for Chinese tech firms, especially those heavily into AI, to find reliable, powerful chips locally. So it's not just a tech upgrade. This has like geopolitical ripples, doesn't it? Reshaping the global market. Absolutely. It forces us to think about the longer term strategic game, how developments like this affect global innovation, AI supply chains, the whole picture. Okay. Interesting. Let's shift from the

chips, the hardware, to the AI itself. Anthropic's been digging into the value shown by its Claude AI, looked at, what, over 700,000 real interactions. What are they finding? Yeah, that's a massive data set. Yeah. And what's really fascinating is they're trying to quantify something that feels pretty abstract, right? AI values. How do you even measure that? Well, they identified over, get this, 3,300 distinct values Claude expressed, then they grouped them into categories. Practical,

Knowledge, social, protective, personal. And it seems the practical and knowledge ones, like being helpful, professional, those came up most. That's right. They were the most common. But what's also quite telling is that ethical values tended to pop up more when Claude was pushing back against harmful or problematic requests. So like a built-in safety check kicking in. Yeah. And did the values change depending on the conversation? They did, yeah. Very context dependent. So you might see...

healthy boundaries emphasized in, say, relationship advice chats. But then see human agency come to the fore in discussions about AI ethics itself. Shows a certain adaptability. So the goal isn't just academic. It's about using this insight to build AI that's, I guess,

better align with what we consider ethical. Exactly that. It's trying to bring a more data-driven empirical approach to that huge challenge of AI alignment, making AI behavior fit better with human norms and expectations. Right. Now, speaking of ambitious applications, the UAE planning to use AI to actually write laws, that feels like a major leap. Yeah, it's definitely bold. The headline goal is pretty eye-catching. Cut the time for developing legislation by potentially 70%.

using AI assistance. 70%. Wow. How would that even work? The idea is to feed the AI a huge database, existing laws, federal, local, court rulings, government data, all of it, and then have the AI help draft, review, even update legislation much faster. And they're setting up a specific office for this, the Regulatory Intelligence Office, shows real commitment building on their other big AI investments. Like that $30 billion AI fund? Yeah.

They're serious about it. But, you know, it immediately throws up some big questions. I was just thinking that like reliability. Can I really handle legal nuance? That's a huge one. And bias, too. If the training data has biases, those could get baked right into the laws. Plus the whole issue of interpretation law isn't just code, is it? No, absolutely not.

It makes you wonder about the role of, well, actual human lawmakers and judges if AI is doing so much of the groundwork. Where's that balance between speed and human wisdom? That's a million-dollar question, isn't it? This UAE move, it really could set a precedent globally. It'll definitely spark debate about automation versus human oversight and governance. Definitely one to watch.

OK, changing tack slightly, but still on the theme of information. Google's Notebook LM has a new feature called Discover Sources. Sounds helpful for research. Oh, yeah. It's a neat upgrade. Basically, you can just describe the topic you're researching. Just type it in. Yeah. Type in your topic, hit Discover, and Notebook LM goes out and finds relevant websites.

web pages, articles, summarizes them for you. And then you can just pull the useful ones straight into your notes. Exactly. You review the curated list, add what looks promising to your notebook, and then you can use all the other Notebook LM features, briefing docs, asking chat questions about the sources, audio overviews. Huh. Sounds like it could really speed up that initial hunting around phase of research. It definitely streamlines it, makes digging into a new subject quite a bit more accessible and efficient, I think. Okay, let's switch to something. Well,

Incredibly optimistic. Demis Sassabis, the CEO of Google DeepMind, has suggested AI might potentially cure all diseases within the next decade. That's huge. It's a very bold statement. Yeah. It really reflects this deep belief in an AI's power to just massively accelerate scientific discovery, especially drug development. Radical abundance, I think you call it. That's the phrase, shortening timelines for medical breakthroughs from, you know,

Years, potentially down to weeks, driven by AI. And he mentioned their project AstroDemas too. AI understanding the visual world, reading emotions, seems related. It all ties into this vision of more capable AI. He also talked about AGI, artificial general intelligence, that sort of human level AI. Right. Possibly arriving in five to ten years. And even...

speculating about consciousness emerging in AI down the line. Plus, hinting at new robotics with reasoning skills. It's a broad, ambitious vision. The disease cure aspect is obviously incredible if it happens, but it does make you think about access, right? And the ethical side. Absolutely. If you have technology that transformative, ensuring it benefits everyone and managing the societal shifts, those become critical conversations. Very complex. Okay, let's shift gears again. Instagram.

They're using more AI for age detection, trying to stop teens seeing adult content. Yeah, it's part of Meta's wider push for youth safety online.

The AI looks at user behavior, maybe cues and images they post. How you act on the platform. Kind of. Yeah. And if it flags someone as potentially misrepresenting their age, it prompts them for age verification. Trying to be proactive makes sense. But I can see privacy concerns popping up there immediately. And like, what if the AI gets it wrong? False positives. Yeah, that's definitely a concern. Fairness, accuracy, accuracy.

These are big challenges whenever AI is used for things like identity or age checks. It's a tricky balance. Meanwhile, over in the legal world again, the U.S. Department of Justice is arguing that Google's AI search features might actually be strengthening its search monopoly. Right. This is part of that big ongoing antitrust trial. The DOJ's point seems to be that AI isn't necessarily creating new competition in search. They're just making Google stronger. Potentially. Yeah.

By integrating AI into its dominant search engine, leveraging its existing partnerships, default settings, the argument is it could lock competitors out even more effectively. So the outcome of that trial could really shape how AI gets used in search and maybe set rules for big tech and AI going forward. It could definitely set some major precedents for regulating AI in consumer tech, particularly around monopoly concerns.

Now for something a bit more quirky. Sam Altman from OpenAI said, users saying please and thank you to chat GPT costs them millions. Seriously? Uh-huh. Yeah. It sounds funny, but it highlights a real technical point. Those extra words, while polite and natural for us. They still need processing. Exactly. They add to the inference load the computation needed for the AI to generate a response.

Across billions of chassis, those seemingly tiny additions add up to significant compute time and therefore cost. Wow, just simple manners costing millions. Shows how resource hungry these models are. It really does. Even small user habits, when scaled up massively, have real economic impacts in the AI world.

A good reminder of the infrastructure humming way behind the scenes. Okay. And on a more commercial note, sounds like OpenAI and Shopify might be teaming up for shopping inside ChatGPT. Yeah. Reports suggest they're testing an in-chat shopping experience.

So you could browse Shopify products, maybe even buy them without ever leaving the chat GPT conversation. Like asking your assistant to find you a blue sweater and then just buying it right there. Pretty much. It points towards this idea of conversational commerce.

getting personalized recommendations, completing purchases, all within the flow of a chat. That could really change how we discover and buy things online, couldn't it? Much more integrated. Absolutely. It could blur the lines between just chatting or asking for info and actually shopping, a potential big shift.

OK, we also had a few other quick hits come across the wire. Amazon apparently getting some pushback from customers. Something about limitations using Anthropix models on their bedrock platform. Yeah, sounds like potential wrinkles in managing access or maybe capacity for those popular high-end AI models within a cloud service. The usual complexities. And Elon Musk is reportedly aiming to raise a huge amount, like $25 billion plus for his XAI X venture.

potentially valuing it around $200 billion. Shows the massive investment appetite and intense competition still driving the high end of AI development. Yeah. Big bets being placed. Then Eleven Labs, they do the audio AI. They've leased something called agent-to-agent transfers. Sounds technical. It suggests they're enabling more complex AI workflows.

letting different specialized AI agents hand off tasks to each other for more sophisticated multi-step processes. Think AI collaboration. Interesting. And finally, the Oscars people, the Academy, they've said using AI in making a film won't disqualify it from awards. Right. Acknowledging AI as becoming a tool in filmmaking, like CGI or editing software, while keeping the focus presumably on the final artistic achievement. Okay, wow. So as you can clearly see, AI is just...

changing constantly, new stuff happening basically daily. It's relentless. And staying on top of it or even just understanding its impact means you got to keep learning, keep upskilling, which actually brings me to something that might help you do just that.

Etienne Neumann's AI-powered Jamgatech app. Ah, yeah. Seriously, the Jamgatech app is built to help anyone get certified in like over 50 really in-demand fields, cloud, finance, cybersecurity, healthcare, business, you name it. Covers a lot of ground. It really does. So whether you're looking to give your career a boost or just get a deeper grasp of these critical tech and business areas, Jamgatech has the resources.

We'll put the link in the show notes for you. Definitely worth checking out, especially given everything we've just discussed. That pace of change really highlights why continuous learning is so important now. Absolutely. Well, that wraps up another pretty packed deep dive into the AI universe. It's clear this isn't just about fancy tech. It's starting to reshape really fundamental things. From the chips we build to how we might write laws to maybe even how we approach health.

It's touching everything. Indeed. And it makes you think, doesn't it, about what comes next? How will AI change work? What are the big opportunities, the big risks? And crucially, how do we steer all this incredible power in a direction that genuinely benefits everyone? That's the ongoing challenge. Definitely.

So that's it from us for this deep dive. Here's a final thought to chew on. Considering AI might write laws or cure diseases, what other core parts of society could be totally transformed in the next few years? And maybe more importantly, what questions should we be asking right now to get ready for whatever that looks like? Good question to ponder. And don't forget, if you want to boost your own skills in these areas, check out the JamGetTech app links in the show notes. Thanks so much for joining us.