Hacker News new | ask | show | jobs
by tompapermill 9 days ago
Hi Tom,

One of the things we've learned is that models often need more guidance when generating documents. We give real-time feedback to models as they're constructing the document, such as "you have content here that you've done nothing with and doesn't appear in the report as it likely overflowed off the page, you should think of a way to handle this" and "this image is sized incorrectly and just will not fit in the area you've specified".

Giving models the tools to create media for print allows both the user and the model to experiment with more wild designs, knowing that things will not randomly break!