Open Superintelligence Lab

Open Superintelligence Lab is a globally distributed and open AI research lab that aims to accelerate AI research 2x to 5x.

Currently the progress is happening in top tier universities and companies, however, there is a massive amount of people all over the world who just need a structured environment to start contributing to AI research, which will in turn accelerate technology and improve living standards for everyone.

We allow absolute research freedom. You may get inspired by our suggestions, projects, tasks, ideas, or bring your own ideas to the table.

We are replicating the exact process top universities and labs use to produce top tier researchers.

Open Tasks

Does RoPE mess with semantics of the vectors, what would you do differently? ➝

Make LLM pretraining accessible for everyone ➝

Articles

Space State Model (SSM): Encode n-gram

Claim: For any n-gram language model, there exists a state space language model that can simulate it with arbitrarily small error.

DeepSeek Sparse Attention

Advanced research on DeepSeek's innovative sparse attention mechanisms for efficient long-context processing.

Tiny Recursive Model

How a 7M parameter model beats 100x bigger models at Sudoku, Mazes, and ARC-AGI using recursive reasoning.

Pretrain LLM with NVFP4

NVIDIA's breakthrough 4-bit training methodology achieving 2-3x speedup and 50% memory reduction.

47x Faster Image Generation

Diffusion Transformers with Representation Autoencoders achieve state-of-the-art FID 1.13 on ImageNet.

QeRL: Beyond Efficiency

Quantization-enhanced Reinforcement Learning for LLMs enables RL training of 32B models on a single GPU.