Brain dump - LLM use

[TuThoThai]

Please drop here all your current use of LLM and thoughts about it, @thbar, @atibaut, @hfjelstad

[hfjelstad]

Breadcrums = Use comment in root README.md to direct LLM to machine specific context

This would be a compact context package for the LLM: Indexes formed either as ontologies , Table of contents, describing shortly how to navigate the information. Description of templates, tools available etc.

Have scripts checking consistency and validate example files on commit/pr

Always have verified and valid source material, never "fire and forget", it usually takes several iterations to get the details right, keep the it short and to balance the output. The best output is when you have detailed knowledge or interest of the topic.

Ask open questions, most of the LLM's seem to agree to much if you prompt them statements, becoming echochambers

What i have done so far:

Experimented with the foundations of a profile
Creating guides
Used the profile foundations + XSD to discuss different modelling approaches -> Identified possible profile improvements
Used profile foundations + XSD to create a working NeTEx 2 EDIFACT converter
Experimented with building ontology based on XSD and real life examples
Building validation and consistency scripts and other useful micro tools
Creating rudimentary pages to try making it a bit more "sexy" (Not my strongest side =D )