OpenAI·Machine Learning Engineer·Technical Phone Screen
- How would you handle numerical overflow when training with FP8 and BF16 in a mixed-precision setup?
“This tripped me up more than I expected.”
Mid-sized
Most of these OpenAI write-ups come from Software Engineers (460), then Machine Learning Engineers (137) and Backend Engineers (23). The most reported rounds are the technical phone screen (335), the system design round (229), the onsite coding round (32) and the behavioral / leadership round (22).
667 interview notes · updated Jul 2026
“This tripped me up more than I expected.”
“This felt manageable at first and then quietly became a lot.”
“Prepped this one from a video walkthrough and it paid off.”
“This is basically a full system design in one question.”
“This one is deceptively massive.”
“Classic question but I always fumble the 'how you reacted' part because I focus too much on explaining the situation.”
“They wanted more than just 'it was assigned to me.' Had to actually explain the reasoning behind picking it, which I wasn't fully prepared for.”
“I started with the data model and probably spent too long there.”
“I fumbled around for a bit trying to remember how L1 differs from L2 in terms of gradients and what that means for a network you'd actually train.”
“I knew the event loop stuff but fumbled a bit explaining why JS is single-threaded by design versus just a limitation.”