OpenAI Software Engineer Interview Experience — A Down-to-the-Wire Coding Round and “Design ChatGPT”

These were all questions I'd already seen before, and I still feel like I did so poorly — why can't I even execute cleanly on questions I already know...

The coding round was GPU Credits II. I asked the interviewer a few clarifying questions, but he kept telling me to just go look for the answer myself in the test cases. I explained a brute-force approach, and the interviewer said he didn't fully follow it but I could go ahead and try it. After that I started writing code, and I handled a few of the edge cases poorly. I ended up debugging for a long time and only got it working with hints from the interviewer. I finished right as time ran out, so there was no time left for follow-ups. I think I probably failed this round.

The system design round was to design ChatGPT. At first there was no need to persist any content at all — I started with the most basic, weakest possible approach. Then the interviewer kept adding more requirements. We did a deep dive into scalability and availability, and then the interviewer had me talk through fault tolerance. Finally we went over the system's metrics and that was it.