Match each alignment method to its learning mode and model count

Match pairs·Medium·4.0 · 0·~2 min·Asked atCharacter Ai Fiddler Ai

Attempt it

Drag each answer to line up with its matching prompt

PPO RLHF

Group relative reward normalization; no separate critic/value network

DPO

Offline preference pair optimization; usually policy + reference

GRPO

Online rollouts; usually policy + reference + reward + value components

TL;DR

PPO RLHF is usually online with more model components, while DPO is offline with fewer components and GRPO removes the critic.

Memory aid

Easy to grasp

Imagine three training styles. One learns while driving in live traffic (online PPO), one studies recorded driving videos (offline DPO), and one compares groups of trial runs without a dedicated scoring teacher (GRPO). They all aim to improve behavior from preferences, but they differ in when data is collected and how many moving parts are required.

Key concepts

RLHF

Concept explanation~2 min read

Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.

Interviewers ask how PPO, DPO, and GRPO differ by learning mode because it exposes whether you understand RLHF as a system or only as a buzzword. At toy scale, teams can get away with a reward-up narrative. At production scale, the real limiter is online vs offline alignment modes and model components, and that limiter interacts with method-specific stability constraints in ways that decide whether improvements are durable. A strong answer therefore starts by separating objective, constraint, and measurement before discussing tactics.

This deep dive follows that exact structure. We begin with the optimization mechanics, then map where pipelines choke, then list early warning metrics, then cover method-level tradeoffs, and finally describe an operational loop that keeps alignment quality stable across releases. That progression is intentional: most regressions happen when one link in this chain is skipped. If you can explain the full chain, your answer sounds like someone who has actually shipped post-training rather than memorized terminology.

Mechanism first: objective, anchor, and control surface

Start with a clean mental model. RLHF-style training is not one metric chase; it is controlled optimization under uncertainty. The policy is pushed toward preferred behavior through a reward-like signal while a stability term prevents catastrophic drift from the pretrained baseline. When these elements are collapsed into one headline score, teams lose the ability to reason about failure causality.

A compact expression of this tradeoff is:

J(\pi)=\mathbb{E}_{x,y\sim\pi}[r_\phi(x,y)]-\beta D_{KL}(\pi(\cdot|x)\Vert\pi_{ref}(\cdot|x))

The reward term encodes preferred behavior, while the KL term acts as a trust region around language competence and style priors from pretraining and SFT. If beta is too weak, optimization can exploit reward shortcuts. If beta is too strong, policy updates stall near the reference and quality gains flatten. This is why mature teams operate with target bands for both reward and divergence, not single scalar goals.

For this question, tie the equation to lived operations: when reward rises but sample efficiency, compute overhead, and policy drift turn unstable, the correct response is to inspect signal quality and constraint strength before scaling run length or batch size. That framing demonstrates control-theory thinking, which interviewers look for in hard RLHF discussions.

J(\pi)=\mathbb{E}[r_\phi]-\beta D_{KL}(\pi\Vert\pi_{ref})

Why scaling exposes hidden bottlenecks

Failure signatures and diagnostics that matter early

Method-level choices and tradeoff policy

Deployment loop and durable alignment practice

Situations where this technique stops working.

2–4 min · Everything important, quickly.

Real products, models, and research that use this idea.

Many open LLM post-training stacks adopt DPO for lower operational complexity when offline preference datasets are strong.
Reasoning-model alignment work has popularized GRPO-style variants to reduce critic overhead.

What an interviewer would ask next. Try answering before peeking at the approach.

QWhat metric would you watch first if this started regressing after deployment?

Pick one stage-specific metric linked to the failure mode, then explain why that signal moves earlier than aggregate quality scores.

1 more follow-up an interviewer would ask next. Sign in to reveal them.

Red flags & common mistakes

The phrases that signal junior thinking. Click to expand.

Most common mistake

Candidates often mix up learning mode and model count, especially confusing DPO as online RL or forgetting critic removal in GRPO.

60 second bullets to scan on the way to the call.

Online vs offline definition
PPO component inventory

Primary sources. Browse if you want the original framing.