Research
Exploring the fundamental principles of intelligence through rigorous experimentation and theoretical analysis.
Topics
Join our Research Team
We are always looking for exceptional researchers to join our mission.
We present Omni-1, a transformer-based model capable of processing text, image, and audio inputs within a single shared embedding space, achieving state-of-the-art results on 14 benchmarks.
A novel quantization technique that reduces memory requirements by 60% while maintaining 99% of model performance, enabling local deployment of 70B parameter models.
Proposing a framework where AI models critique and refine their own outputs based on a set of constitutional principles, significantly reducing harmful outputs without human intervention.
Automating the design of vision transformer architectures using evolutionary algorithms, resulting in models that are 2x faster and 15% more accurate than ViT-L.