πŸ“– Step 9: AI/LLM#303 / 350

Reinforcement Learning from Human Feedback

Reinforcement Learning from Human Feedback (RLHF)

πŸ“–One-line summary

A reinforcement learning technique that aligns model behavior using human preference feedback.

πŸ’‘Easy explanation

A training method where humans pick 'this answer is better,' and the model learns from that feedback. It's why ChatGPT answers politely instead of chaotically.

✨Example

인간 ν”Όλ“œλ°±μœΌλ‘œ λͺ¨λΈ 쑰율

πŸ‘ A

예의 λ°”λ₯Έ λ‹΅λ³€

πŸ‘Ž B

λ¬΄λ‘€ν•œ λ‹΅λ³€

↓ μ‚¬λžŒμ΄ Aλ₯Ό 선택 ↓
🧠 λͺ¨λΈμ΄ A λ°©ν–₯으둜 ν•™μŠ΅