179 questions
No questions match those filters.
← All questions
Training & Fine-tuningHard· RLHF· SFT· Hallucination
What is the relationship between SFT data quality and h...
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansHigh factual depth in SFT data is only safe when the model already possesses the underlying knowledge. If there is a capability gap, SFT teaches structural completion rather than knowledge retrieval. RLHF motivates a shift toward correctness-conditional signals. In PPO, rewards are only propagated for completions the model generates that are judged correct, which prevents the model from taking gradient steps on knowledge it does not actually possess.