Advanced Reinforcement Learning Interview Questions, issue 11, Feb 6, 2026

The Context Leakage Trap

Principal ML Engineer interview at Anthropic, and the interviewer asks:

We need to scale our RLHF dataset 10x. To maximize variety, we present human labelers with two model outputs generated from different user prompts (e.g., Response A to Prompt X vs. Response B to Prompt Y). We ask them to pick the better response.

Breaking prompt consistency in RLHF collapses the Bradley-Terry signal, causing the reward model to learn prompt difficulty instead of response quality.

The full answer, with the mechanism and the arithmetic, is for paid subscribers on Substack.

Read it on Substack

Get the next one

Free on Substack. Unsubscribe in one click.