Get the latest tech news

Prompting by Activation Maximization

Presenting prompt synthesis by activation maximization.

With even a handful of coefficients, the search space is tremendous, but backpropagation and gradient descent help find useful minima representing useful and productive functions. We load the model, initialize a subject, select a target (seven) and go.We do the forward pass, compute loss, back propagation and run the optimizer. Small, generatively pretrained models have deep knowlege about the world but aren't so intelligent they cant follow even simple instructions.

Get the Android app