Tilde Research @ Stanford

​Join us for a discussion on how mechanistic interpretability informs model design. Pizza and boba will be served.

​We’ll share examples from our Telescope project and other internal research showing how circuit-level analyses can reveal structure and behavior within large models, and how those insights have guided architectural decisions and training interventions. By tracing how specific features emerge and interact, we illustrate ways that mechanistic interpretability can inform practical design choices.

​Tilde Research is an SF-based AI research lab working on mechanistic interpretability and model architectures. We’re backed by Khosla Ventures, Yufei Zhao, and researchers at OpenAI and Anthropic. For more information, please visit tilderesearch.com.