AI Daily News March 20 20205: ⚡AI Capabilities Surpassing Moore’s Law 💥Nvidia to Invest Hundreds of Billions in US Manufacturing 👀Apple Reshuffles AI Leadership to Improve Siri 🧠Nvidia Open-Sources Reasoning Models to Advance AI Collaboration
17:41 Share

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

If you're finding these deep dives valuable, please take a moment to like and subscribe to the podcast on Apple. So you're tuning in because you want to get a handle on all this AI news, right? You want to go beyond just the headlines. Yeah, exactly. You want the substance. That's the mission. That's what we do here. Today, we're going to zero in on a single day, March 20th.

2025. Oh, wow. And we're going to pull out the really crucial developments in the AI world. We've got a bunch of news and some insightful summaries. Just think of this as your fast track to being informed about AI. I like it. Okay. Let's dive right in. One of the big stories was NVIDIA's plan for manufacturing. And we're talking about

hundreds of billions of dollars for U.S. semiconductor and electronics manufacturing over the next four years. That's a lot of money. And it really points to a very fundamental shift in strategy.

For a long time, the production of these components has been concentrated in Asia. And NVIDIA's move here is really about building up the infrastructure domestically in the U.S., which aligns with the government's efforts to boost domestic production, make our supply chains more resilient. And I think what's also really interesting about this is

This could trigger a kind of silicon race. Oh, interesting. Where other major tech players follow suit. Yeah. And that could lead to a diversification of manufacturing capabilities beyond just NVIDIA's efforts. Right. So this could be the start of something really big.

Yeah. Okay. So it's not just about building their own factories, right? It sounds like they're also looking at partnerships. Right. The reports mentioned collaborations with TSMC and Foxconn. Yeah. Big names. To manufacture their cutting up systems here in the States. Yeah, absolutely. It's about leveraging that existing knowledge and those capabilities. Yeah. Right. While simultaneously establishing a strong US base,

And, you know, when you consider the scale of what they're trying to do. Yeah, it's massive. And how complex semiconductor manufacturing is this kind of multi-pronged approach makes a lot of sense. And then in addition to that, they also had that big acquisition, right? They bought a synthetic data company. Right. Called Gretel. Gretel, yeah. For over $320 million. Yeah, and that highlights a really critical bottleneck in AI development. And it's data. Right. You know, to train AI models, you need massive amounts of data, high quality data.

And it also has to respect people's privacy. So Gretel specializes in creating what's called synthetic data. This is artificially generated data that behaves like the real world data. Interesting. But doesn't contain any of that sensitive personal information. So think about creating these realistic training examples.

without using anyone's actual private details. So it's all about basically bolstering their cloud-based AI offerings, right? Yes. By directly addressing these data challenges. Exactly. And privacy concerns. Precisely. By bringing Gretel's technology in-house, NVIDIA can then offer its customers access to more data for training their AI.

And that could really speed up innovation while also helping them to navigate this increasingly complex world of data privacy regulations. So it's fascinating how these two moves, the manufacturing investment and the Gretel acquisition,

they really support each other in fueling the expansion of AI. It's really interesting. - Yeah, so a stronger semiconductor industry in the US, more stable supply chain, and ultimately just the ability to meet these increasing demands of AI. Okay, now let's switch gears a bit. Another big player, SoftBank, also made a significant move on March 20th, agreeing to acquire Ampere Computing for $6.5 billion cash. - Big money. - Big money. - So tell me about Ampere. - So Ampere is a US-based company, designs chips.

And they've made a name for themselves by focusing on these high-performance processors that are also very energy efficient. Their target market is cloud computing and AI workloads. So very competitive. Yeah, super competitive. Rapidly expanding sector.

Right. And so this fits very neatly into SoftBank's strategy to get into the AI infrastructure game. Yes. They've been making investments in this space for a while now. And this really strengthens their position. I mean, they've clearly identified AI infrastructure as a key area for future growth.

And acquiring Ampere gives them a substantial foothold in the processor market. Right. Which is really important. What's interesting to note here is that Oracle and Carlyle, the previous investors in Ampere, they're selling their stakes. So this is going to make Ampere an independent entity under SoftBank. Gotcha. So Ampere will focus on ARM-based processors for data centers. Right.

which is a really interesting technology. And they'll continue to operate in Santa Clara, California with their team of around 1,000 engineers. So SoftBank is making a big bet on the future of AI and semiconductors. Absolutely. With this acquisition. And this could lead to some interesting developments in AI computing down the road. Yeah, some serious advancement. Okay, now Apple, they also made some interesting changes in their AI leadership. Oh, Apple, yeah. It sounds like they're trying...

trying to really give Siri a much needed upgrade. Yeah, it looks like they're really trying to revamp Siri and their entire approach to AI.

You know, they're definitely feeling the pressure in that voice assistant market. For sure. And they're trying to bring new ideas and new expertise to that area. So it's a really interesting move. And the big news was Mike Rockwell, who was previously in charge of the Vision Products Group, is now going to lead the Siri team. Big change. Big move. Yeah. Well, I mean, Rockwell's experience in leading the development of a product, as groundbreaking as the Vision Pro suggests that Apple is doing,

really serious about taking Siri to the next level. Right. So they're bringing a fresh perspective, a more ambitious vision perhaps for Siri. Meanwhile, John J. Andre is going to continue to focus on broader AI research and other AI technologies within Apple. So they're not

abandoning their core research efforts. And Paul Mead is taking over for Rockwell. Yeah. Leading the Apple Vision Pro team. It's a bit of a ripple effect at the executive level. It is. But all of these changes really signal a very focused effort to prioritize and enhance Apple's AI capabilities. Yeah.

Particularly when it comes to Siri. The hope is that this is going to lead to a much more intuitive and natural and intelligent interaction. Right. For Apple users down the line. So it's a wait and see, but it's a very promising development. Yeah. So these leadership changes. Okay. Now let's zoom out for a second and look at the bigger picture. One of the really interesting things that emerged on March 20th was that the pace of AI innovation may be exceeding Moore's Law. Oh.

Oh, wow. That's a pretty profound statement. That is a big statement because Moore's Law has been the gold standard for decades, predicting that roughly every two years, the number of transistors on an integrated circuit would double. That's Moore's Law.

But now in the field of AI, what we're seeing is an even faster acceleration. Interesting. Some are calling it Hyper Moore's Law. Hyper Moore's Law. And the performance of these leading AI models, in some cases, it's been observed to double every six months. Wow.

That's a big difference. That's a big difference. So what's driving this acceleration? It's more than just packing transistors onto a chip now, right? Yeah, you're right. It's not just about the hardware anymore. Right. Of course, advancements in hardware are still contributing, but this rapid acceleration is also being fueled by improvements in the algorithms. Okay. And the vast amount of data we have available now. Makes sense. So better algorithms are able to achieve more with the same computing power.

And the availability of these huge data sets is allowing AI to really learn and generalize in ways that just weren't possible before. Okay, so if AI is truly on this HyperMores law trajectory, what are the implications? Well, we could see faster progress towards AGI. AGI, okay. Artificial General Intelligence. This is the idea of AI that can understand or learn any intellectual task that a human being can't.

we might see breakthroughs in complex reasoning, real-time cognitive abilities happening much sooner than experts had predicted. So very exciting, but also presents challenges, right? Yeah, for sure. We have to think about our technological infrastructure, our education systems, and particularly ethical frameworks, which will need to adapt to these evolving capabilities. Yeah, that's almost hard to fathom how fast it's moving. It's really moving fast. Okay, now shifting gears,

We saw Hollywood push back pretty hard against some proposed AI copyright rules. Yeah, interesting development. It seems like a pivotal moment for the creative industries. It really is. There's a growing concern among actors, screenwriters, musicians about AI companies, you know, the big ones like OpenAI and Google.

who are lobbying for certain exemptions from copyright law. Okay. Exemptions that could allow AI models to be trained on copyrighted work. Gotcha. Without permission or any compensation to the creators. And over 400 very prominent creatives. Yeah. Including Aubrey Paza, Paul McCartney. Yeah, big names.

Calling for government intervention to prevent what they call AI theft of creative labor. Yeah, that's a really powerful statement coming from such influential figures. Yeah. So, you know, they see this as a fundamental issue of protecting their work, their ability to earn a living from their craft. Right. And on one side, you have the principle of strong copyright protection for creators. On the other side, you have the concept of fair use. Right.

And the argument that access to large amounts of data, including copyrighted material, is essential for AI development. Yeah. Particularly in those creative fields like music, writing, art. Right. And also for AI's ability to understand and model our culture. So it's a really interesting debate. So what are the potential consequences? Well, you know, if AI companies get these broad exemptions, it could really destabilize the creative industry to value the work of human artists. Yeah. But on the other hand,

If very strict restrictions are imposed, it could slow down the progress of AI in those creative areas. Right. Limit its ability to learn from and build upon existing works. Right. It's a complex issue with both economic and cultural implications. Okay. Yeah.

- It is. Now back to NVIDIA. - Back to NVIDIA. - They were in the news a lot. They also made some significant announcements at their GTC conference. - Right. - Including the decision to open source some of their advanced reasoning models. - Yeah, that's a big move. - That's a major contribution to the AI research community. - It is. These aren't your typical AI models. They're making public these advanced reasoning models

designed to understand information from multiple sources. So text, images, videos, multimodal understanding. They're also designed for what we call symbolic reasoning, which is this more logical rule-based AI.

and the ability to break down complex tasks into smaller, more manageable steps. So it's really crucial for building more sophisticated AI agents. - Okay, and this is all about empowering other developers and researchers to build more advanced human-aligned AI agents. - Yeah, that's the whole point. They're democratizing access to some really cutting edge AI capabilities. This could accelerate progress towards AGI, encourage more collaboration in the academic research community,

And it could open up new opportunities for startups. Yeah, really lower the barrier to entry. Absolutely. Lower the barrier to entry for everyone. And in addition to all of that, they also unveiled a new line of personal AI supercomputers. Wow. The DGX Spark,

At the DGX station. Massive amount of computing power for individuals, researchers, developers. So they're basically giving data center capabilities in a workstation. Exactly. So these are high performance computing systems designed for individuals who need that power in their office or their lab. They use the latest Blackwell architecture from NVIDIA. So what this really does is it enables a more personalized approach to AI development.

So you can build and fine tune these models locally. Right. Potentially reducing the need for those expensive cloud computing resources. Yeah. Really fostering innovation at a personal scale. Exactly. Bringing the data center to your desktop. That's wild. Okay. Now Hugging Face launched a really interesting iOS app on March 20th. Tell us about that. Yeah. So Hugging Face is really committed to making open source AI accessible. Yeah. And

Their new iOS app uses AI powered image recognition. Oh, cool. To give you real time descriptions of what the user sees through their phone camera. Oh, wow. Pretty amazing stuff. It uses open source vision language models to generate these descriptions. And they're very detailed. That's cool. And contextually relevant. And super useful for people who have visual impairments. Exactly. It has the potential to be a game changer. Right. For people with visual impairments, helping them understand their surroundings.

But it's also useful for anyone who wants to learn more about the world around them. You can use it for education. You can use it while you're traveling. It's a really great example of how powerful these open source AI models can be when they're made available through these user-friendly apps. Okay. And there was some interesting research published in Nature about...

optimizing generative AI. Okay. Using feedback from language models, not just from humans. A little bit meta, isn't it? It is. So traditionally, when you want to fine tune these generative models. Yeah. Particularly for those subjective qualities like creativity and coherence. Yeah. You rely on human feedback, which is called RLHF, reinforcement learning from human feedback. But this new research is exploring a different approach.

using AI-generated evaluations of the output quality to fine-tune the performance of the generative model. So instead of having human experts review and rate, they're using other AI models to do the judging. Oh, that's cool. Provide the feedback. Interesting. So the system uses these AI-generated evaluations to adjust the parameters, essentially teaching it to produce better outputs. So what are the advantages?

Well, it could significantly reduce our reliance on expensive human reviewers. Yeah. And it also opens up the possibility of having these self-improving AI systems. Right. That continuously refine their outputs based on their own internal understanding of what quality is. So that could be a really big step towards those more autonomous, sophisticated AI agents. And then something a little closer to home maybe. Closer to home, yeah. Is the increasing role of AI in fast food. Right.

Sounds like our next trip to Taco Bell or Pizza Hut might be a little different. That's right. Yum! Brands which owns Taco Bell and Pizza Hut is rolling out AI-driven voice assistants at their drive-thrus across the United States. These are systems that are powered by conversational AI and large language models. Wow. They're designed to take orders, offer upselling suggestions, even handle the payment processing, all without any human interaction. Interesting.

So the next time you're at Taco Bell late at night, you might be talking to an NI. Wow, that's wild. The goal is to improve the accuracy of the orders. Okay. Reduce labor costs. Speed up service for the customer. But there are some downsides too, right? Well, yeah, there are always concerns about job displacement. Yeah, absolutely. Especially in the fast food industry. And also, how reliable will these systems be in those noisy, chaotic, real-world situations? Mm-hmm, right.

It's going to be interesting to see how it plays out. Now, there were a few other notable AI developments. Oh, yeah. A few more things. March 20th was a big day. It was a very busy day that maybe didn't make the headlines. So researchers at Google AI and UC Berkeley proposed this new idea called inference time search. OK. It's a new method for scaling AI models where the model generates several potential answers to a query and then uses reasoning to select the best one. Interesting.

LG unveiled Exon Deep, a reasoning AI that achieves the comparable performance to larger models but with just 32 billion parameters. Wow, very efficient. Muse launched the Muse S Athena headband, a wearable for AI-powered cognitive fitness training. NVIDIA and XAI joined the AI infrastructure partnership,

with other major players like Microsoft and BlackRock. XAI also debuted its image generation API. Oh, wow. Featuring their Grok 2 Image 12 model. Keeping busy. Microsoft partnered with Innate to develop brain-inspired AI systems. So just a ton of activity on this single day.

A lot going on. Highlights just how fast the field is moving. Yeah. I mean, March 20th, 2025 gives us a snapshot of this rapidly evolving AI landscape. It really does. So what stands out most to you from all these developments? Which areas do you anticipate will have the biggest impact? Yeah, for me, the thing that really stands out is that the acceleration of AI capabilities beyond Moore's Law. Right. I mean, if that trend continues, it could lead to some really profound changes across industries in our lives.

The advances in reasoning models. Yeah. The emergence of personal AI supercomputing. Yeah.

It's very exciting to see where this all leads. Absolutely. And on that note, we want to remind you, our listeners, if you find these deep dives valuable and want to support the show, please consider making a donation. You can find the links in the show notes. Yes, every little bit helps. We want to keep this show free and accessible to everyone. And for those of you looking to reach a really engaged audience of professionals interested in technology and AI, consider advertising your business here on our deep dives. Yeah, it's a great way to reach a large audience. So that's it. Another fascinating look at a

single day in AI. With AI evolving so rapidly, it really makes you wonder, are we prepared for a world where Hypermores law is the norm? It's a big question. And how will our understanding of intelligence itself have to adapt? That's something to think about. Thanks for diving deep with us. Thanks for listening. We'll see you next time.