Tag
#training
From the Library
Library
AI Data Supply Chain
AI data is not one pile of files but a supply chain: from sources and rights through cleaning and provenance to continuous audit. The open web remains important, but high-quality, usable data is increasingly constrained, contested, and expensive to govern.
Read →Library
Fine-tuning — a scalpel, not a universal hammer
Fine-tuning changes model weights. It is powerful when you have data, evals and a clear reason. It is an expensive mistake when it hides a bad prompt, missing RAG or an unclear process.
Read →Library
LoRA: how I got one face
A LoRA is a small add-on layer for an existing model, not a new model. It teaches the model one thing, such as a specific face. Here is how the LoRA behind my face on this site was made: 37 images, a nonsense trigger word, 1,200 steps. And what still does not work.
Read →