What is the difference between base models and instruct models?

• A. Base models are smaller in size than instruct models.
• B. Base models predict the next token, while instruct models are fine-tuned to follow instructions.
• C. Base models handle images; instruct models handle text.
• D. Base models include memory; instruct models do not.

Answer: (B)

The main distinction is how they’re trained and what they’re tuned to do. Base models learn by predicting the next word in a sequence, using vast amounts of text. That gives them broad language skills and general reasoning, but they aren’t specifically trained to follow user instructions. Instruct models start from that same base and are then fine-tuned with data that pairs prompts with preferred, instruction-following responses. This alignment process (often using supervised fine-tuning and sometimes reinforcement learning from human feedback) teaches them to understand and comply with explicit instructions, produce useful step-by-step answers, and avoid straying off task. The other options aren’t the defining difference—size, modality, or memory aren’t what separates base from instruct models.