You’re in an AI Engineer interview at Anthropic and the interviewer asks:
“Our model is lagging on the Chatbot Arena leaderboard. We need to boost our ELO score by 50 points next release to match GPT-5. How do you adjust the post-training pipeline to make this happen?”
Don’t say: “We need to improve our reasoning capabilities on hard math benchmarks like …
