Enterprise Neurosystem Announces High Performance Atomic Language Model Developed in Collaboration with Uganda’s Crane AI Labs

The Enterprise Neurosystem is delighted to announce the completion of the Atomic Language Model (ALM), an open source AI model developed in collaboration with Uganda’s Crane AI Labs.

The practical and economic future of AI has been questioned given the massive compute and power consumption needs of hosting LLMs at scale. For example, in the U.S. the likes of OpenAI and Meta are investing billions of dollars to build vast datacenters equipped with tens of thousands of GPU processors, and which require gigawatts of power to operate.

The Atomic Language Model is an answer to this pressing issue. It’s a recursive, high-performance language model that is less than 50KB in size, and which can run efficiently on a standard (low power) non-GPU processor.

Most modern language models are empirical black boxes — their internal workings are unknown or opaque. ALM stands out by anchoring its core in formal mathematical verification. Leveraging the Coq proof assistant, ALM’s recursive mechanisms are rigorously tested and proven to be correct — thus shifting the validation of AI into mathematical certainty.

The ALM was built in equal measure by Enterprise Neurosystem architects and a talented team of Ugandan AI engineers and researchers at Crane AI: Kato (Mubiro) Steven, Bronson Bakunga, Glorry Sibomana, and Gimei Alex. Their objective was to use this groundbreaking architecture as the foundation of the first locally developed large scale language technology dedicated to Ugandan languages.

The Enterprise Neurosystem was introduced to the Crane team by the Science, Technology and Innovation Secretariat of Uganda. Following a conversation held in Kampala with Kato Steven and Enterprise Neurosystem founder Bill Wright, a plan was developed to build a unique small scale AI model — the ALM — with powerful analysis capabilities and low energy consumption.

The ALM’s small size breaks new ground in fields where most large scale LLMs are unwieldy and impractical. The initial development of this approach was inspired by Alexander Flom’s, a researcher at the UOR Foundation, where his new theory of computation has led to substantial performance breakthroughs.

The use cases for ALM are many, and include:

Embedded and Edge IoT Devices: Providing self-contained AI on the smallest embedded platforms, with no network required.

Space Exploration: Enabling AI on missions where every byte, watt and gram is precious.

Climate and Environmental Sensing: AI logic can be run on remote, energy-efficient sensors without the need for network connectivity.

2G Network Compatibility: Delivering robust language tools in regions with limited or basic mobile connectivity.

For more information on the ALM, check out Kato Steven’s GitHub repository.