Announcing Noema Labs
Introducing Noema Labs
We're excited to announce the launch of Noema Labs, an open research lab dedicated to building multilingual reasoning AI models. Our mission is to create AI systems that can reason, reflect, and communicate effectively across languages and cultures.
Our Mission
At Noema Labs, we believe that AI should be accessible to everyone, regardless of the language they speak. Current AI systems show impressive capabilities in English and other high-resource languages, but performance often degrades significantly in lower-resource languages. This creates a substantial inequity in access to advanced AI capabilities across different linguistic communities.
We're working to address this gap through a comprehensive approach to multilingual model development that focuses specifically on reasoning capabilities. Our key areas of focus include:
- Multilingual reasoning alignment: Developing techniques to align AI systems with human reasoning across diverse linguistic and cultural contexts
- Efficient training methods: Creating approaches that enable high-quality multilingual models to be trained on consumer hardware
- Comprehensive evaluation: Building benchmarks to assess reasoning capabilities across 100+ languages
- Open science: Releasing models, code, and evaluation frameworks to the research community
Why "Noema"?
The term "noēma" refers to the content of thought—that which is thought about. This concept is central to our work, as we aim to create AI systems that can engage with the content of human thought across languages and cultures.
Our Approach
Our research is guided by several key principles:
1. Reasoning First
We believe that reasoning capabilities are fundamental to useful AI systems. Our models are designed to excel at logical reasoning, causal inference, and counterfactual reasoning across languages.
2. Cultural Context Matters
Reasoning often relies on cultural knowledge and context. We're developing approaches that account for cultural differences in how reasoning is expressed and understood.
3. Efficiency and Accessibility
We're committed to making multilingual AI research accessible to researchers and developers worldwide. Our techniques are designed to work on consumer hardware, reducing the barriers to entry for multilingual AI research.
4. Open Science
We believe in the power of open science to accelerate progress. Our models, code, and evaluation frameworks are released under permissive licenses to encourage collaboration and innovation.
Join Us
We're building a globally-distributed team of researchers and engineers passionate about making AI more equitable across languages. Please check our Join Us page if you are intrested in advancing our vision!
Stay tuned for more updates as we share our research findings, model releases, and opportunities to collaborate.