The language model GPT-2 from OpenAI is one of the most coherent generative models for text out there. While its generation capabilities are impressive, it’s ability to zero-shot perform some...
03 Nov 2019 - 7 minute readIt’s always great fun to visualize neural networks. We previously saw how visualizing a neuron’s activation revealed fascinating insights. For a supervised learning setting, the training process of a neural...
21 Jun 2018 - 14 minute readI recently came across a research paper (Radford et al.,) that quite fascinated me. The authors discovered that a single neuron captured the sentiment of an entire piece of text....
12 May 2017 - 12 minute read