Keep pulling the thread on Will Brown & Johannes Hagemann.
Prime Intellect and the AI research lab RCI have collaborated to train and open-source some of the largest mixture-of-experts (MoE) models.
Johannes Hagemann considers the research on recursive language models (RLM), originally authored by Alex Sang, to be one of the most promising directions for solving long-horizon agent tasks.
Prime Intellect's mission is to make frontier AI lab training accessible to everyone.
Prime Intellect aims to provide startups, enterprises, and other labs with access to frontier AI infrastructure that is currently exclusive to large labs.
Prime Intellect's platform is a full-stack solution covering compute orchestration up to a complete post-training stack.
Prime Intellect's platform includes sandboxes for secure code execution and evaluations as part of its environment hub.
Will Brown believes that to optimize AI systems end-to-end, developers need direct access to model weights for deeper customization than is possible at the prompt level.
Will Brown predicts that every company will eventually become an AI company, and most of those will want to have their own AI research lab.
Prime Intellect's full-stack research platform is named "Lab" and is designed as a community platform similar to GitHub for sharing and collaborating on AI environments.
Johannes Hagemann asserts that large labs like OpenAI and Anthropic have a competitive advantage because they can optimize models for their specific products, such as ChatGPT and Claude Code.
The developer tool company Cursor built its own "composer one" model through large-scale post-training, using the Cursor application itself as the training environment.
Post-training encompasses multiple techniques, including prompt tuning, supervised fine-tuning (SFT), and online distillation, with reinforcement learning (RL) being the most significant for large-scale models.