About
Guide labs is a product-focused AI research company where we are building a new class of interpretable AI systems that humans and domain experts can understand and reliably debug. We make these models available via our API.
The prevailing paradigm in machine learning and AI is to train models as monoliths optimized solely for performance. However, this results in models that are difficult to work with, and leaves critical model properties at the mercy of the optimization process. At Guide Labs we are taking a different path. We believe you cannot reliably debug, steer, and align a model you don’t understand. We are rethinking the model architecture, loss function, and entire pipeline to constrain the model training process such that the models we get are more easily understandable, their errors easier to identify and fix, and easy to align.
San Francisco