Updating the model weights

SFT
Supervised fine-tuning

RL post-training
DPO · GRPO · RLVR

LoRA
Low-Rank Adaptation

imitate correct trajectories; needs labeled examples of the right behavior

sample, score against a reward or preference signal, reinforce what wins

limits the set of parameters that can change; cheaper, safer updates

They need: benchmark + evaluator.