Our Mission

We are a non-profit, decentralized lab operating across continents with highly skilled programmers and researchers. We tackle the most challenging problems in computing and artificial intelligence, prioritizing long-term impact over scale. We believe that high-quality datasets and innovative architectures deliver far greater value than arbitrarily large datasets or inflated ambitions.

Key Research Focuses

Dataset Compilation & Curation

Collecting and cleaning terabytes of data to create domain specific datasets for training advanced models. Instances include audio, textual and video datasets. Includes: writing custom processing and compilation pipeline.

Training models and architecture

Developing novel architecture and training large-scale models for scaling laws and domain specific reasons. It includes fine-tuning and post-training paradigms.

Scaling laws and limits of Compute

We study compute returns and scaling law of architectures for Deep learning and advanced artificial intelligence models. Heavily inspired by LAION and Open-SCI.

Benchmark and Evals

Evals are core of all training models, we invest time and resources understanding the nuances and asking fundamental questions. It was quite inspired by Xeophone and Yoav Goldberg.

Publications

High-Impact Research

Currently under preparation. Our initial findings will be published soon.

Stay tuned for peer-reviewed papers on music, evals and scaling laws.

Collaborate & Support

We actively seek collaborations with academic institutions and industry partners on cutting-edge AI projects.

Worked with