The Nvidia show & Neuralink update - MetaVisions #22

Hi all, another week, another dolla! Yes, Nvidia is mentioned again, don’t blame me, their team keeps doing amazing work!

The Nvidia show!

This week we saw Nvidia’s annual event , the CEO’s (Jensen Huang) keynote had an euphoric atmosphere, which is a reflection of the company’s stock performance in the last two years. I don’t want to go in detail about every announcement, but I want to cover my two biggest takeaways:

Nvidia Omniverse: ‘everything that moves, will be robotics’, Nvidia’s CEO made this statement during his keynote. We can already see this becoming true, warehouses are using robots, cars are learning how to self-drive and I have even seen fully robotized coffee shops. For robotics to be developed and deployed efficiently, there needs to be a way to test and train prototypes. But, does it need to be physical prototypes?
No! We have reached a point in technology where platforms like Nvidia Omniverse offer organisations all the tools necessary to create life-like digital twins of pretty much anything.

Let’s imagine a scenario, Volvo wants to design and build a brand new manufacturing plant, it will leverage robots to distribute parts across the floor.
Creating a digital twin of that plant will be essential, not only will this allow the team to prepare ‘simple’ things such as what will the electrical wiring look like, but most importantly it will provide a 1:1 realistic training and testing ground for all of the robotics and machinery that will exist within that environment.
The digital twin can be filled with humans, machines and workflows that mimics what we expect will happen in the real world, we can even insert random accidents and errors. This will give us a realistic playground where our moving robots can be deployed to be programmed and tested. Once this is done, we can pretty much lift and shift it to our real manufacturing plant.

It is exciting to note that Jensen mentioned Omniverse’s integration with the Apple Vision Pro, highlighting the importance of being present inside the digital twin for some scenarios… For me, watching incredible high-fidelity 3D models in a 2D screen does not allow users to extract the most of the digital twin!

A new evolution of AI-ready infrastructure: Nvidia revealed the B200 GPU, and the GB200, a combination of two B200 GPUs and a single Grace CPU. This new development is claimed to increase LLM inference workload performance by 30 times and significantly improve efficiency, reducing costs and energy consumption by up to 25 times compared to the H100 model.

In the past, training a model with 1.8 trillion parameters required 8,000 Hopper GPUs and 15 megawatts of power. However, Nvidia now claims that the same task can be accomplished with just 2,000 Blackwell GPUs, cutting power usage to a mere four megawatts.

The debut of the GB200 “superchip” in conjunction with the Blackwell B200 GPU is a great achievement. According to Nvidia, in tests with the GPT-3 LLM containing 175 billion parameters, the GB200 outperformed the H100 by seven times and quadrupled the training speed.

Nvidia wants to continue leading the Generative AI revolution by providing all the amazing infrastructure.

On X, a Neuralink employee went on a livestream with Noland Arbaugh, a young man that was paralyzed from the shoulders down because of a freak diving accident! Noland received his chip implant in January, this is following permission to test the Neuralink chips on humans by the FDA in May of 2023.

The livestream showcases Mr Arbaugh moving his laptop cursor ‘magically’ to play chess, watching it feels like a supernatural scene, as if he had some sort of telepathy power. He also mentioned staying up until early hours in the morning playing Civilization VI, for eight hours straight!

Noland mentioned that at first it’s awkward and does’t always work, but overtime he started to learn the actions from the chip, how it reacted to his body. After some time, it because natural. I believe it might feel like using an eye tracking system, but with the added ability to command actions such as ‘right and left click’.

Musk mentioned that one of the ultimate goals for Neuralink is to give back movement to those that have been paralysed, a Swiss university has achieved this. The École Polytechnique Fédérale enabled Gert-Jan Oskam, who is paralysed, to walk just by thinking about the movements involved.

That was achieved by putting electronic implants brain and spine, which wirelessly communicate thoughts to his legs and feet.

Of course some people will be scared and skeptic around the potential dangers of this technology, and I get it. It is crazy to think that this is the starting point of our Cyberpunk journey. Even scarier to think that this is the WORST that this technology will EVER be, its capabilities will only improve.
In times like this, I believe it is important to embrace the positive aspects of innovation. How amazing would it be if we had a solid solution to those suffering from paralysis and other spinal problems?

See you next week,
Davi, MetaVisions

Reply

or to participate.