1 min lesson

Fine-tuning vs reinforcement learning: what each is for

Use the lesson to respond to this. "A model produces well-formatted, syntactically valid tool calls but the actual edits it makes are often wrong. Which stage is most likely the lever and why?" Keep the answer plain.

SFT vs RL: what each is actually forthe distinction interviewers test

Both stages update the same weights toward different targets: one imitates, one scores.

SFT VS RL, SIDE BY SIDE

Objective: SFT: Maximize likelihood of curated demonstrations (behavior cloning). · RL: Maximize expected reward, not imitation likelihood.

diagram: compare

The distinction interviewers test: shape versus score.

Fine-tuning vs reinforcement learning: what each is for

Step 1 of 2

SFT vs RL: what each is actually forthe distinction interviewers test