we keep improving the underlying models, but most interactions still boil down to typing a box of text and hoping the model understands what we want
that seems strange when you think about how much information a person normally uses to understand a task
files, previous work, deadlines, preferences, software state, examples, feedback, and the actual goal behind the request.
what would an AI interface look like if it were designed around the task rather than around a chat box?
[link] [comments]