Language Model Sketchbook, or Why I Hate Chatbots

Explores new ways we might interact with language models beyond chatbots. It proposes “daemons” – background characters offering writing suggestions. “Branches” maps cause/effect chains to better understand issues. “Epi” provides a context menu of analysis tools for research and non-fiction. While chatbots suffice, these prototypes aim to leverage language models’ full abilities as “epistemic rubber ducks” for strengthening critical thinking. In showcasing these ideas, the author illuminates new interactive possibilities as we determine how best to utilize these neural network-grown language agents.