Unconventional AI: Databricks' Ex Revolutionizes Energy

Unconventional AI: A Radical Overhaul of Computer Architecture

In the ever-evolving world of artificial intelligence, the quest for the next big technological breakthrough has given rise to ambitious projects. Among them, Unconventional AI stands out for its commitment to rebuilding computer architecture from the ground up. Under the leadership of Naveen Rao, former head of AI at Databricks, the company aims to make inference processing significantly more energy-efficient.

The key element of this transformation lies in a new innovative computer architecture based on the use of oscillators. This approach promises to revolutionize how AI systems consume energy, potentially reducing their energy footprint significantly.

Launch of the Un-0 Model: A Step Towards the Future

Last Thursday, Unconventional AI unveiled its first AI model, named Un-0. This image generation model marks an important milestone by demonstrating how the company's technology can mimic traditional AI systems. Accompanied by a detailed research paper, this launch explains how the team managed to create a fully functional image generation model using a software simulation of the new architecture.

The Un-0 model, comparable to tools such as Stable Diffusion or OpenAI's GPT Image 1, stands out for its performance achieved through an oscillator-based architecture. This approach differs radically from the conventional chips used in current computing and large language models (LLMs). Naveen Rao estimates that this technology could reduce the energy consumption of AI systems by up to 1,000 times.

Towards a Revolutionary Infrastructure

Although the infrastructure necessary to realize this vision is still in development, Unconventional AI is making significant progress. Currently, the Un-0 model operates on a software simulation of the oscillator chips developed by the company. However, plans for the creation of actual chips are underway, with the goal of building a complete inference stack from scratch.

The company's ambition is to provide computing power comparable to that of current providers, but with drastically reduced energy consumption. "We are going to build a new type of system composed of our chips," says Rao. "We will run AI models on this system, with queries coming in and inferences going out, all using only 1/1000 of the usual power."

A Major Challenge for a Small Team

Unconventional AI's goal is extremely ambitious, especially considering that the company currently has fewer than 50 employees. However, in the face of the rapid expansion of AI and rising energy costs to meet inference demand, this project could be one of the few initiatives capable of tackling this challenge.

Naveen Rao emphasizes that energy supply will be a major constraint for AI development in the coming years. "The scalability of AI is difficult because of energy. This is going to be the fundamental limit in the coming years. You simply cannot go beyond that. It will ultimately be a problem limited by energy," he states.

Thus, Unconventional AI positions itself as a key player in the search for sustainable solutions for the future of artificial intelligence, directly addressing one of the industry's biggest challenges: energy consumption.