The Best-of-N Paradox

Senior AI Engineer interview at Anthropic, and the interviewer asks:

“You’re reranking 16 agent trajectories with a critic model to boost your SWE-bench score. Accuracy went up. Now walk me through the exact shape of that accuracy-vs-rollouts curve, and tell me when this 16x spend is actually worth it in production.”

Don’t say: “More rollouts means more chances to get it right, so accuracy goes up.”