Texti #45: Agentic AI in the Omniverse

Last week Mr. Huang, the elephant 🐘 🐢 that holds the entire AI Industry in his hands, spoke recently at the AI summit in India, and he presented his vision of the evolution of the rest of industries onwards. It must be said, it’s an interesting approach.

TLDR:

  • Agentic AI - Micro AI’s trained to serve a specific purpose 🤖
  • Omniverse - Isolated virtual environment where AI can be trained to control a robot, so that it later can do that properly when installed in a real robot. 🌐
  • Insane AI Scaling - AI will scale so faster than you can learn how to learn to ride a bike. 🚲

Long story:

First of all, Mr. Huang came in like a wrecking ball into the AI industry and his stocks went through the roof. In the last 3 years he became one of the most influential people on the planet. He’s fighting problems in his company nobody has fought before, the stocks of his employees sky-rocketed so high, that the employees realised they don’t ever need to work again, because they’re rich enough. 😅

He is also the person who said that by 2030 developers won’t exist, and you don’t need to learn programming anymore. As a developer I found this message unreasonable, and I still believe he’s wrong about it, but time will show. ⌛

Anyways, Agentic AI - despite the fancy name, this means, AI models that are purpose built. Unlike GPT4o or any other model you’ve previously likely used, these type of models are going to be specialised in a certain industry/area of industry. For example, this could be an Agent for solving tax related problems 💶, or AI Lawyer 🧑‍⚖️, or even an AI Doctor 👩‍⚕️. Don’t freak-out though, these are not people replacements, these are more human helpers.

Most industries have a vast amount of information floating around, which is hard to be remembered and taken into account by a single human. This is why for example for complex surgeries there is a consilium of doctors to debate if this is the right thing to do or not.

Now an AI agent here would be an ideal use of data aggregation and suggest options suggestions.

Look, a vast amount of data in the hands of a non-professional doesn’t serve a great purpose, all it is just theory with lack of understanding if this is actually useful or not.

Mr. Huang, said though, a phrase with which I agree and this should be probably my focus as a developer in the coming years.

“Developers of the future, will mainly develop AI Agents, for niche businesses” - Jensen Huang

✅ I cannot agree more here. I believe the future will be reshaped, and this shape is likely going to look like a billion small models that will be used from case to case. There will be an AI to coordinate most of them, but if it's a business niche, that's a model tailored for the business's needs.

Next Omniverse - this is basically a virtual world 🌐 where you can model whatever you want, and let AI Agents learn how to do a specific physical job. The context that is presented here, is factories where you have to do a quite repetitive job, like for example manufacturing, but not limited to it, their website also advertises autonomous driving.

Basically what you’d do as a company would be, create a small realistic video game, where all that happens is let’s say vacuum assembly. You’d then show the model what it needs to do, and it’d learn how to do the action time and time again, you could run this over millions of iterations, and the model would learn the most efficient and fast way to do it, including deviations from the usual stuff. 🤖

Another use case you’d consider doing which would make even more sense would be to simulate a virtual world with all the chemical elements known to human, and let an AI model do an infinite combination of other things with them in different aggregation states to research new and improved materials, potentially even cheaper. ⚛️

I find this fascinating as this could in theory propel us forward really fast. This is pretty much Comic Book territory, where in an enclosed lab somebody builds up the next spider man 🕷️ 👨

This introduces us into Insane AI Scaling, given the fact that now we’re able to build AI Agents and dedicated Omniverses for them, we’re approaching a state where we could deploy computing power into producing new materials 🧪, new computing algorithms 🧙‍♂️, new machines 💻, new theories, new everything. This would potentially mean we could scale indefinitely ♾️, and remove every barrier we had in our heads so far.

  • Is Jensen Huang talking about a potentially incredible world? Well of course.
  • Is he mad to believe such a thing? Absolutely, but you have to be insane to try and build the thing of the future.

Let us see what the outcome of this will be. I believe we’re going to be living not in an apocalyptic future, but rather a beautiful harmony of humans and robots like The Jetsons.


That's it folks, see you next week ❤️️️️️️️

Happy Prompting!

Remember to invite your friends to subscribe at https://newsletter.texti.app

{{ address }}
Unsubscribe · Preferences

© 2024 Texti Newsletter