Unconventional AI releases Un0 image model and claims 1,000x less power use

In short: Unconventional AI has released Un0, an image-making AI system, and says its new computer design could eventually use up to 1,000 times less electricity for AI results.

What happened

Unconventional AI, led by Naveen Rao, the former head of AI at Databricks, announced its first AI model, called Un0. Un0 is a tool that generates images, similar to systems like Stable Diffusion and OpenAI’s image tools.

The company says the main point is not the pictures themselves. It is how Un0 is built. Unconventional AI is working on a different kind of computer design based on “oscillators,” which are repeating electrical signals (like a metronome that keeps time).

Right now, Un0 does not run on a real Unconventional AI chip. The company said it built Un0 using a software simulation, which is like testing a new car design in a driving simulator before you build the actual car. In a research paper, the company says the simulated system can match today’s top “diffusion models” (a common method for generating images by refining visual noise step by step).

Rao told TechCrunch the company plans to release schematics for an actual chip soon. Longer term, Unconventional AI says it wants to build a full service where prompts go in and AI results come out, but with far less power use.

Why it matters

AI systems can be expensive to run because they use a lot of electricity, especially when they answer user requests, which is often called “inference” (the work of producing an output after the AI is trained). If Unconventional AI’s approach works on real hardware, it could lower the energy cost of using AI and reduce pressure on power grids as more people and companies use these tools.

Source: TechCrunch AI