From Autocomplete to Assistant

What this page will cover

• Base model behavior: it continues text, it doesn’t answer questions
• Instruction tuning on example conversations
• Reinforcement learning from human (and AI) feedback
• Side-by-side: same prompt to a base model vs. a chat model

Security lens

Safety behavior is trained in, not enforced by code. It’s a strong tendency, not a hard boundary, which is why jailbreaks exist.