GenAI Basics · Question 29 · Hard
In Reinforcement Learning from Human Feedback (RLHF), what is the reward model trained on?
- Human preference judgements, such as rankings of which of several model responses is better
- The raw web text used for pre-training
- The model's own token probabilities
- Image labels collected from the internet
Answer and explanation
Answer: A. Human preference judgements, such as rankings of which of several model responses is better
Explanation: Humans compare model outputs; a reward model learns to predict these preferences, and the language model is then optimised with reinforcement learning to score highly on that reward.
Previous: A model's two candidate next tokens have logits 2 and 0. With softmax at… · Next: Why do transformer models add positional encodings (or positional embeddings)…
Practise genai basics
Try a fresh set of genai basics questions with an explanation after every answer, free and without signing up: GenAI Basics practice. Or sit a full placement mock test.
More genai basics questions
- In a RAG system, what is the main purpose of a vector database?
- A query embedding is q = (1, 2, 2). Four document embeddings are: D1 = (2, 1…
- A model's two candidate next tokens have logits 2 and 0. With softmax at…
- Why do transformer models add positional encodings (or positional embeddings)…
More technical & programming practice
Pseudo-code · C Programming · Java · Python · Data Structures · Algorithms · Object-Oriented Programming · DBMS
LogicWiz Exams home · All practice topics · Placement mock tests · Daily 10 challenge