Base model
Also known as: base LLM
A base model is the model as it comes out of pretraining — before any instruction tuning or alignment. It knows a lot and predicts text well, but it’s not an “assistant.”
Ask it a question and it may continue the question rather than answer it, because it was only trained to continue text.
That’s the gap that SFT and RLHF fill.
Why the distinction exists: researchers and fine-tuners work with base models (maximum flexibility), while users mostly meet the *aligned* version (the assistant). Same weights family, very different personalities.