The mii-llm Manifesto

We believe the future of artificial intelligence will be built through diversity, specialization and cooperation. Frontier language models are becoming extraordinarily capable, particularly in programming, mathematics, reasoning and interacting with digital environments. Their progress is remarkable, but intelligence is much broader than the ability to solve verifiable tasks or automate digital processes.

Human knowledge is the result of thousands of years of cooperation, different cultures, languages, disciplines and shared objectives. Everything we use to train language models comes from this extraordinary accumulation of collective intelligence. We believe AI should reflect this diversity.

A model specialized in understanding a language and its culture may require different knowledge and training objectives from one specialized in scientific research, legal reasoning, creative writing or interacting with digital environments. We do not believe that optimizing every model toward the same capabilities will necessarily produce the best intelligence for every purpose.

Our vision is an open ecosystem of specialized language models, capable of becoming exceptionally good at specific tasks and cooperating with one another. These models should be designed for their purpose, with appropriate architectures, datasets, training methods and evaluations. They should be efficient enough to run where they are needed, from local and edge devices to large computing infrastructures.

We believe openness is fundamental to this vision. Open weights, datasets, training recipes and reproducible evaluations allow researchers, organizations and individuals to understand, improve and adapt models to their own needs.

They also make it possible for smaller communities to contribute to the frontier of knowledge. At mii-llm, we want to contribute to this future by researching, training and releasing specialized models, together with the datasets, tools and benchmarks needed to develop them.

We also believe that cooperation between models deserves to become an important research direction. Specialized models may eventually work together, sharing knowledge, distributing tasks and combining different capabilities into more complex systems. Such cooperation may open new paths toward broader forms of intelligence.

We are already living through an extraordinary technological revolution. Our goal is to make this revolution more open, diverse and accessible, while exploring new ways for intelligent machines to extend human capabilities.

We believe the future of AI lies in specialized models, each developing distinct capabilities and contributing to a broader intelligence through cooperation.