Alex Beutel
OpenAI
About Me
I am currently the tech lead for our Safety Research teams at OpenAI.
I was previously a senior staff research scientist, tech lead, and manager at Google Research, co-leading a Responsible ML team and driving research spanning
, , , , and . I led work from basic research to product impact, with >50 launches across multiple products.I got my
in computer science at , advised by and . My research focused on large-scale user behavior modeling, covering , , and . Over the course of graduate school, I interned with Facebook's Site Integrity team and News Feed Ranking teams, Microsoft's Cloud and Information Service Lab, and Google Research.Before graduate school, I majored in computer science and physics at
. While there, I worked with and on computational geometry for terrain modeling.AI Safety
-
The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions
... , , , , , Alex BeutelPreprint, 2024
-
Practices for Governing Agentic AI Systems
... , , , Steven Adler, Cullen O’Keefe, Rosie Campbell, Teddy Lee, , , Alan Hickey, Katarina Slama, , Paul McMillan, Alex Beutel, ,White paper, 2023
-
Controlled Decoding from Language Models
... , Jong Lee, , YaGuang Li, Tao Wang, Yanping Huang, Zhifeng Chen, , , , Alex Beutel,ICML, 2024
Recommendation Systems
-
Building Healthy Recommendation Sequences for Everyone: A Safe Reinforcement Learning Approach
... , , , , , , Alex BeutelFAccTRec, 2020
-
Top-K Off-Policy Correction for a REINFORCE Recommender System
... , Alex Beutel*, , , Francois Belletti,WSDM, 2019
-
Latent Cross: Making Use of Context in Recurrent Recommender Systems
...Alex Beutel, , , Can Xu, Jia Li, Vince Gatto,WSDM, 2018
-
Recurrent Recommender Networks
... , , Alex Beutel, , How JingWSDM, 2017
Robustness
-
Diverse and Effective Red Teaming with Auto-generated Rewards and Multi-step Reinforcement Learning
...Alex Beutel, , ,Preprint, 2024
- Blog post
- Press: MIT Technology Review
-
Understanding and Improving Robustness of Vision Transformers through Patch-based Negative Augmentation
... , , , , Alex Beutel,NeurIPS, 2022
-
CAT-Gen: Improving Robustness in NLP Models via Controlled Adversarial Text Generation
...Tianlu Wang, , , , Kang Li, , Alex Beutel,EMNLP, 2020
Fairness
-
First-Person Fairness in Chatbots
... , Alex Beutel, , , Anna-Luisa Brakman, , Meghan Shah, , ,Preprint, 2024
- Blog post
- Press: MIT Technology Review
-
Measuring and Reducing Gendered Correlations in Pre-trained Models
... , , , Alex Beutel, , , , ,Preprint, 2020
-
Fairness without Demographics through Adversarially Reweighted Learning
... , Alex Beutel, , Kang Lee, , , ,NeurIPS, 2020
-
Fairness in Recommendation Ranking through Pairwise Comparisons
...Alex Beutel, , , Hai Qian, , Yi Wu, , , , ,KDD (Applied Data Science Track), 2019
- Incorporated in a large-scale production recommender (Google Research's 2019 Year in Review).
-
Putting Fairness Principles into Practice: Challenges, Metrics, and Improvements
...Alex Beutel, , , Hai Qian, , Christine Luu, Pierre Kreitmann, Jonathan Bischof,AAAI/ACM Conference on Artificial Intelligence, Ethics, and Society (AIES), 2019
Machine Learning Systems
-
The Case for Learned Index Structures
... , Alex Beutel, , ,SIGMOD, 2018
-
Learned Indexes for a Google-scale Disk-based Database
... , , Alex Beutel, , , , Xiaozhou, Li, Andy Ly,ML for Systems workshop at NeurIPS, 2020
Fraud Detection
-
CopyCatch: Stopping Group Attacks by Spotting Lockstep Behavior in Social Networks
...Alex Beutel, Wanhong Xu, , Christopher Palow,WWW, 2013
- ACM Computing Review Editor's Highlight on CopyCatch
- Patent by Facebook (Patent Number 9077744)
- Discussion by Facebook
- Included in courses at Carnegie Mellon and University of Florida
-
(Best Paper Award)
FRAUDAR: Bounding Graph Fraud in the Face of Camouflage
... , , Alex Beutel, , ,KDD, 2016