NeuroNomixer
  • Home
  • Blog
  • Visual Guides
  • Authors
  • Contact
Sign InSign Up
HomeBlogAuthorsContactPrivacy Policy

© 2026 NeuroNomixer — Built with Next.js & Tailwind CSS

Visual Guides/RLHF: How Human Feedback Shapes AI
Applied AI

RLHF: How Human Feedback Shapes AI

Play the role of a human rater and watch the reward model learn your preferences. See how RLHF turns human judgments into AI behaviour.

0% complete · 0/5 ratings · 0/3 phases explored

Sign in to save progress

Section 1: The Three Phases

Expand all three phases to complete this section.

Section 2: Be the Human Rater

For each question, pick the better AI response. Your choices train the reward model.

Round 1 of 5

Prompt

Explain photosynthesis

0/5 ratings collected

Section 3: Reward Model Training

Illustrative curve: reward model accuracy (agreement with held-out human preference labels) vs. number of human comparisons. Shape and endpoints anchored to published InstructGPT-era results.

52%
100
61%
1K
67%
10K
71%
100K
X: Human comparisons (training examples)Y: Agreement with held-out human labels (illustrative)

At 100 comparisons the reward model is barely better than chance (50% on pairwise picks). With large comparison datasets, InstructGPT-scale reward models reached roughly 69-73% agreement with held-out labelers (Ouyang et al. 2022), close to the ~73% rate at which labelers agree with each other. That ceiling is real: the reward model stays imperfect, which is why reward hacking remains a problem.

Section 4: RLHF vs DPO

RLHF
DPO
Complexity
High (3 phases)
Low (1 phase)
Stability
Lower
Higher
Performance
Comparable
Comparable
Compute
2–3× more
1×
Used by
InstructGPT, Llama 2-Chat
Zephyr, Tulu 2

DPO (Direct Preference Optimisation) skips the separate reward model entirely: it re-frames preference learning as a classification problem directly on the policy, eliminating the PPO loop. RLAIF (RL from AI Feedback) replaces human raters with a strong AI model like Claude or GPT-4, dramatically cutting annotation cost.

Known Problems

Reward Hacking

The policy discovers inputs that score highly on the reward model but are not actually preferred by humans, exploiting its blind spots.

Over-Optimisation

Without a KL penalty, the model drifts far from the original SFT policy, producing degenerate outputs that fool the reward model.

Cost

Collecting high-quality human preference data is slow and expensive. OpenAI spent millions on annotators for InstructGPT.

Key Insight

InstructGPT (2022) showed that a 1.3B RLHF model was preferred over a 175B GPT-3 by human evaluators. Scale doesn't matter as much as alignment. This was the breakthrough that led directly to ChatGPT.

← All GuidesNext Guide →