Pioneering Research

We are dedicated to deep research in AI, transformers, multimodality, and novel architectures.

Core Research Areas

Novel Architectures

Designing next-gen neural networks beyond current paradigms, focusing on efficiency and new capabilities.

Multimodality

Creating models that seamlessly understand and process text, images, audio, and other data modalities.

Scaling Laws

Investigating the principles that govern model performance with scale to build more predictive models.

AI Safety & Alignment

Developing robust techniques to ensure our models are safe, interpretable, and aligned with human values.

Our Philosophy

We believe in a foundational, first-principles approach to research. Our goal is not just to advance the state-of-the-art, but to develop a deeper understanding of the core principles of intelligence. We encourage open collaboration, rigorous experimentation, and a long-term vision for creating safe and beneficial AGI. This means tackling fundamental challenges and sharing our findings with the broader community to accelerate collective progress.

Publications

A New Era of Attention: Gated Mechanisms in Transformers

NeurIPS 2025

Generative Sonics: High-Fidelity Music Synthesis

ICML 2025

Principled Scaling Laws for Transfer Learning

Preprint

Towards Robust and Verifiable AI Alignment

Preprint

Join Our Research Team

We are looking for passionate researchers and engineers to join us on our mission. If you are excited by the prospect of solving fundamental challenges in AI, we would love to hear from you. We value curiosity, rigor, and a drive to make a positive impact.