Nvidia GTC, Manus, and Gemini Robotics – Live and Learn #64

✨ Quote ✨

There's a new kind of coding I call "vibe coding", where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.

– Andrej Karpathy - (source)

The future will be rad. Nah, scratch that, even the tools we have right now are completely bonkers. "WTF AI moments" happen to me every other week: using Claude to translate difficult to understand code across different programming languages or creating 3D models in seconds using deemos hyperhuman or using Cursor to glue it all together into 3D worlds. This stuff makes me excited and sad at the same time. Coding as we know it will disappear. Vibe coding, or something like it, will replace it. Embrace it or not, it is going to happen. And with more advances in robotics and autonomous agents, coding will be far from the last thing to change drastically...

🖇️ Links 🖇️

Gemini Robotics by Google DeepMind. This project by Google is about making robots usable in real world scenarios. For them, improvement means three things: better dexterity, better generality, and better interactivity. This new iteration of Gemini Robotics improves all three of those. They published a 64-page report of their work and released several video showcases of what this platform is capable of. They partner with Apptroniks, i.e. Google is not producing the robots in-house. They build the brains for the robots instead—two models power the whole thing: a Vision-Language-Action Model (VLA) and an Embodied Reasoning Model (ER). Google is essentially using LLMs to let robots interact with the real, messy, and dynamic world. This Gemini approach can even adapt to different types of robot platforms, such as the Aloha2 or the Apptroniks humanoid, it is truly general. For the whole Gemini Robotics Project Google has built a landing page with even more info, where you can also sign up as a trusted tester.

Nvidia GTC 2025 Keynote by Jensen Huang. Every year, this keynote shows where the future of computation is heading. As always we'll get better, faster, more efficient chips, which will power the AI revolution. Last year Jensen's catchphrase was: "The more you buy, the more you save". This year it is "the more you buy, the more you make", GenAI in data centers will power extremely productive workloads, and the better and more AI you can serve to your end customers, the more profits you'll make. You need better chips and more of them to make that happen, and Nvidia is going to give them to you... for a price that is ^^ GTC is about letting the industry know what Nvidia is building next so that people can prepare their data center spending with that in mind. The big announcements for me this year were their Photonics Networking Switches directly bridging fiberoptics to silicon, Nvidia Rubin, the successor to the Blackwell family, and Nvidia announcing the their quantum research center. Jensen, in his quest to bring down the cost of computation, is making full use of Jevon's paradox. Bring down the cost to enable new use cases, which creates novel demand and markets, and then profit. To help this process along Nvidia is building lots of software for their chips to run. Their software ecosystem is as much part of Nvidia's moat as their chip design. This year they announced things like Nvidia Groot N1, the Newton Physics Engine, and Nvidia Dynamo, just to mention a few.

Manus by Monica. Manus is an autonomous agent, a wrapper around Claude and other models, that gives LLMs the ability to execute tasks autonomously. They have a nice announcement video up on YouTube too. In their words, Manus is "the next paradigm of human-machine collaboration and potentially a glimpse into AGI". It seems a little bit like magic to let LLMs do tasks for you autonomously. But like all agents, Manus still makes mistakes and gets stuck. On top, there is a whole lot of political baggage that comes with it because Manus is produced by a Chinese company. This begs the question, should lots of our computers be controlled by a Chinese-owned AI? There's an excellent post by Marginal Revolution diving into this topic some more.

Too Cheap to Meter by Abundance Institute and S3. This is the second full-length feature film documentary by the S3 YouTube channel and I recommend watching it. The film is about the future of energy production in the US and the startups that are trying to envision and build that future. It details the role of the different types of energy that are going to matter, like nuclear, fusion, solar, or even natural gas synthesized from the atmosphere, and looks at how they got to be what they are today.

🌌 Travel 🌌

No travel section this time because I have been staying at the same place for the last two weeks. I am still in Colombia, mostly programming 3D web stuff with three.js but will continue to Bolivia at the beginning of April 😊

🎶 Song 🎶

Solfeggietto by Luca Sestak

Youtube Music | Spotify

---

That's all for this time. I hope you found this newsletter useful, beautiful, or even both!

Have ideas for improving it? As always please let me know.

Cheers,

– Rico