OpenAI Machine Learning Engineer Interview Questions

openai research fulltime machine learning tech phone screen interview experience

本帖最后由 匿名 于 2026-5-17 15:49 编辑 分享一下最近的 OpenAI Research 电面 以下内容需要积分高于 150 您已经可以浏览 第一轮：general coding 1hr Debug distributed system。 给一个 buggy 的 job scheduler in python，要检查有没有 data racing、deadlock、lock contention，以及是否有正确的 rate limiter 实现。中间需要自己写 test case 确定 scheduler 是 work 的，然后要算 scheduler schedule job 的时间以及成功率。 地里没见过这题 第二轮：ML debug 1hr Debug transformer。让一个 miniGPT 能生产正确的文字，follow up 实现 kv cache，都是用 PyTorch。 地里有看到过 第三轮：ML coding 1hr Matrix...

openai machine learning research onsite interview experience

本帖最后由 匿名 于 2026-5-17 16:03 编辑 电面经验可以从这里看 https://www.1point3acres.com/bbs/forum.php?mod=viewthread&tid=1177093&page=1&extra=#pid20998286 因为是recruiter reach out，所以过了电面以后直接onsite就去要面的组 以下内容需要积分高于 188 您已经可以浏览 Onsite分为一下几个部分 Collaboration interview 聊了一下自己研究方向的SOTA都有哪些，有哪些优缺点，有没有相关的技术经验，以及怎么看未来领域的方向发展。和怎么应用这些研究方向到产品。 SD End to end 设计一个实时的传感器系统。End to end指要从传感器选型，usecase， PM request 入手，然后要考虑算法选择，怎么设计ML model，数据从哪里来，efficiency，power都要考虑。 这个SD很需要domain knowledg...

openai fulltime machine learning onsite interview coding questions

分享一下前几周面的两轮 ML coding 题目，希望对正在准备的朋友有帮助。 之前一段时间没登录米都扣完了，求大家加加米，多谢 第一轮：Entropy 计算 给定 logits，计算 entropy（需要用 log-sum-exp trick 保证数值稳定） 如果 logits 是 streaming 进来的（block-by-block），怎么 online 算 entropy？第二轮：1-NN 用 L2 distance 实现 1-NN（不能用 for loop，要 numpy vectorize） 把1-NN表达成 neural network 的 forward pass（用 linear layer + softmax） 如果换成 L1 distance，怎么用 neural network 实现？

openai fulltime machine learning video interview experience

等了又等，推了又推，还是面完了。目前只是过了前两轮，最后等on-site. 第一轮应该是那个分类器human data的题，还有一道data mining的系统设计题，关于怎么从医疗数据中挖掘有效训练数据的系统设计大概是。 最近oai的题有些变化但是somehow我还用的老题，变形金钢捉虫，monster battle等题目，toy language好像也是原题，面的都不错，求大米。 现在感觉有connection才是正道，单纯靠刷题的时代过去了。。。

openai lead machine learning engineer tech phone screen experience

两轮 第一题变形金刚捉虫，我只找到了3个bug，有一个死活找不出来，最后他告诉我，其实是一个typo，把v改成y就行了，我半天没看到... 第二题植物感染题，只做到3道题。他们的IDE让我很不习惯，一直写错代码... 我觉得我几乎投入了两周很多时间来复习，以上两道题目几乎做烂，到面试的时候，我居然像得了失心疯一样。😭 我可能太紧张了，总而言之心态崩了。

Machine Learning Infrastructure Interview at openai: Memory Allocator Manager Design

I interviewed for an ML infrastructure position. The task was to design a mem allocator manager with a total capacity of N. It included functions like `allocate()` and `free()`. I initially implemente

Advice for Technical Phone Screen with OpenAI Internship

Hey guys, recently got an interview for OpenAI Internship next summer and do not want to fumble. Was wondering if anyone have done it before and can share your experience or have advice on how I can p

Design a Read/Write-Optimized Data Structure (General Coding, Non-ML)

## 75-min Coding Interview: Design a Data Structure Optimized for Reads and Writes Implement a **custom data structure** in a language of your choice that is optimized for both **reads** and **writes

ML Coding Interview: Math + Coding + Research Brainstorm (Notebook-based)

## 60-min ML Coding Interview (recoverable from the prompt) You will work in an online notebook (requires a Google account). The interview includes: 1. **Math + coding tasks** related to machine lea

ML Coding: Human Annotation Data Filtering (Detect Bad Annotations)

You are cleaning human annotation data. Given samples and their labels from one or multiple annotators (optionally with annotator IDs, confidences, timestamps, and a small gold set), design and implem

ML Coding with NP and NN Layers

Solve a ML-based puzzle and implement it in code. Familiarity with vector and matrix addition/multiplication in numpy, common neural network layers, and implementation of linear layers with batched in

ML Debugging with Transformer Model

Debug a given machine learning model implemented using Python and PyTorch (a transformer model). Identify and fix all bugs to ensure the model runs successfully. Demonstrate knowledge of ML architectu

ML Coding

Solve a ML-based puzzle and implement it in code. It's useful to be familiar with numpy: adding and multiplying vectors and matrices, common neural network layers. Having a crisp understanding of how

ML Debugging

You are given a short implementation of a ML transformer model (using Python and PyTorch). Your task is to find and fix all bugs in order for the model to work successfully. This tests your knowledge

Human Labeling and Training a Classifier (ML Coding)

## Human Labeling + Training a Classifier (ML Coding) > The original post only provides a high-level description; missing data format, labeling procedure/cost constraints, training objective, and out