##### ###### ##### ### # # ### # # ###### ## ## ## ## ## ## ## # # # # # ## ##### #### ##### # # # # # # # #### ## # ## ## ## ## # # # # # ## ## # ###### ## ### # ### # ######
##### ###### ##### ### # # ### # # ###### ## ## ## ## ## ## ## # # # # # ## ##### #### ##### # # # # # # # #### ## # ## ## ## ## # # # # # ## ## # ###### ## ### # ### # ######
SFT teaches models what to say. Alignment teaches them how to say it well. I'll walk you through preference-based training methods that make models more helpful, harmless, and honest.
You'll learn DPO, GRPO, and other techniques that have replaced traditional RLHF for most practitioners.
