I built this little tool to help me understand what it's like to be an autoregressive language model. For any given passage of text, it augments the original text with highlights and annotations that tell me how "surprising" each token is to the model, and which other tokens the model thought were most likely to occur in its place. Right now, the LM I'm using is the smallest version of GPT-2, with 124M parameters.

Send me a message or webmention
Back to feed