LLM System Design Interview, issue 70, Sep 14, 2026
The SFT Quality Trap
Staff AI Engineer interview at OpenAI, and the interviewer asks:
“Your team paid expert annotators to write the highest-quality SFT responses possible, detailed, sourced, with citations. After fine-tuning, factual hallucination rate went up. What happened?”
Don’t say: “The annotators must have written wrong references.”
Why paying expert annotators for flawless citations quietly spikes your hallucination rate and how elite teams filter training data against the model's internal knowledge boundary.
The full answer, with the mechanism and the arithmetic, is for paid subscribers on Substack.