What the Katacombs Pilot Taught Me About Luna High and XHigh

Two distinct technical studies were conducted. The first pilot explored using GPT-6-Luna and XHigh for Katacombs implementation, following a collaborative Zoom session with Steven Diamante to evaluate local model viability. The second project focused on fair content moderation by training two models on the Civil Comments dataset, consisting of 1.8 million public comments. This trust and safety analytics exercise utilized Kaggle notebooks and public data to measure toxicity, insults, and threats through a browser-based demo.