Kieth–Have you considered using AI to help those of us who struggle with Compling? Here’s link about the latest from Claude that assists programmers to write scripts: https://www.youtube.com/watch?v=d8lKfXLKe9Q
The cool solution would be a ive AI assistant that knew where you were in the compiler and could respond appropriately. Scrivener’s compiler is all GUI manipulation, and I very much doubt Claude or any other AI would be able to do much of use at the moment. Multimodal models (vision + text) are fairly new, and while they seem to reason from a single image, a GUI is actually an abstract visual “maze” where buttons lead to new states and features, toggles can cause contextual effects elsewhere etc. Many visual states with distracting information. This is a hard problem, you’d need to label what document outputs you want then train a sort of chain of thought algorithm across all possible permutations of visual information.
A simpler solution: only use text and train it using the Scrivener manual, the KB and forum posts, this is called a fine-tuned model and could be hosted by L&L and accessed via an web API from Scrivener. How useful it would be, who knows?
Now if Scrivener’s compiler could be automated (aka Scrivener had its own API), then things would really sizzle! Writing scripts is orders of magnitude simpler for an AI, they get trained on billions of lines of code, then their statistical guesstimatore can work well, even for relatively novel problems.
You can already come close to this with the Markdown + post-processing script workflows. These can be set up with very simple options in Scrivener, passing most of the heavy lifting through to, for instance, LaTeX.
4 Likes