What I Learned From Going Through Meta's AI Researcher Interview Loop

I made it to Meta's final round for an AI Researcher role. Two years of industry research experience, relevant publications, solid ML fundamentals. I thought I was ready. I wasn't — not because the technical bar was higher than expected, but because I prepared for the wrong version of the interview. Here's what the loop actually looks like from the inside, and where the real difficulty sits. The loop structure (briefly) For context, the Meta AI Researcher loop typically covers: Coding rounds (ML-adjacent implementation, not pure LeetCode) Research presentation (deep dive into your own work) Research discussion (your opinions on the field) ML system design Behavioural Most write-ups focus on the coding. That's not where candidates lose this interview. Research taste is the hardest thing to prepare for The research discussion rounds aren't testing whether you've read the papers. They're testing whether you have opinions about the field — independent, defensible views on what matters, wha