Tilde Research is a moonshot AI laboratory founded in 2024 and based in Palo Alto, California. Its research agenda is directed at the frontier of artificial intelligence, spanning mechanistic interpretability, novel architectures and pretraining science. The stated aim is to build foundational understanding of how AI models work, rather than to iterate on existing systems.
The company's work concentrates on two areas. Interpreter models are designed to expose the inner workings of AI systems, converting black-box prompting into transparent and controllable processes. Control technologies permit direct steering of model behaviour and are intended to deliver performance gains beyond those available through conventional post-training or fine-tuning. The team has demonstrated enhanced reasoning capabilities in open-source models and fine-grained control in text-to-video generation, and has also worked on reasoning capabilities in open-source models.
Tilde Research raised $8 million in seed funding. Its staff combine experience from academia and leading AI infrastructure companies, and the company releases open-source tools and research. It operates in the verticals of artificial intelligence research, AI foundation models, and AI safety and interpretability.
