Claude in BOSL2

[raster]

How is Claude (AI, LLMs) being used in the development of BOSL2?

[revarbat]

Currently, mostly in the creation of GitHub workflows, and some ongoing experimental development of rather complex NURBS code.

Probably we will be using it to create test suites in the near future.

Using Claude to develop code does carry a risk of it writing slop. If you keep a close eye on it, though, test its output thoroughly, and review what it wrote, it can be a very helpful timesaver for experienced developers. Verifying code can take a lot less effort than the initial development.

[thoastbrot]

just saw this thread - I recently tried to use claude to rewrite a gridfinity library (written in native openscad). The issue: it's producing a whole lot of slop since it has no way to verify the outcome - which is by nature visual. A CSG tree or similar representations can look correct to the machine, but make no sense at all.

Claude (and other LLM agents) need a ground truth, a test function which tells them how well they did. Without, they produce slop in the most cases. Maybe deduction is possible after feeding enough openscad code/good models written with BOSL2.