Base vs instruction-tuned (-it) models

Placeholder — full lesson coming.

What the "-it" suffix means: a base model only continues text, so asked a question it may just produce more questions. Instruction tuning (supervised fine-tuning on instruction/response pairs, then alignment such as RLHF/DPO) is the extra training stage that turns it into an assistant that follows instructions and holds a conversation. To cover: base vs instruct behaviour, the SFT→alignment pipeline, why our runtime is base-only (pretraining), and chat templates.