Tilde Research is an AI research lab founded in 2024 and based in Palo Alto, California. It works on the frontier of artificial intelligence across three areas: mechanistic interpretability, novel architectures and pretraining science. The company describes itself as a moonshot lab, with a stated focus on building a foundational understanding of how AI models work.
Two lines of work form its main output. Interpreter models are designed to reveal the inner workings of AI models, turning black-box prompting into transparent, controllable systems; these have demonstrated enhanced reasoning capabilities in open-source models. Its control technologies aim to steer model behaviour directly, with fine-grained control demonstrated in text-to-video generation. The company positions both as alternatives to conventional post-training or fine-tuning approaches. Additional technical domains include interpreter models, reasoning enhancement and text-to-video generation.
Tilde Research raised $8 million in seed funding. The team combines expertise from academia and leading AI infrastructure companies, and the company actively publishes open-source tools and research.
