Hacker News Reader

> predict the next word in 40GB of Internet text

This could do wonders for lip reading correction.

OpenAI trained the initial 1.5B model on ~160G of text.. so I’m sure it’s already going to give amazing results.