How do you make good explanations for what a neural network does? We provide a framework for analysing explanations of the behaviour of neural networks by looking at the hypothesis of how they would act on a set of given inputs. By trying t...

Exploring the OthelloGPT model.

We trained a simple Convolutional Neural Network on a poisoned version of the MNIST dataset. Some elements of the dataset include a watermark, for which the label has been modified. We describe the process for uncovering the path through th...