Bob McGrew

Career

Bob McGrew's papers cited here run from 2017 to 2024, beginning with robotics and reinforcement learning research. The OpenAI o1 System Card, released in December 2024, lists McGrew under executive leadership.⁴¹

Research

McGrew co-authored several robotics and reinforcement learning papers, starting with Hindsight Experience Replay (2017), a method for learning from sparse rewards. Later work trained a five-fingered robot hand in simulation for in-hand object manipulation (2018) and then for solving a Rubik's Cube (2019), where the paper credits McGrew with leading aspects of the project and setting research directions.⁴⁵²

McGrew also co-authored a 2019 hide-and-seek study on emergent tool use among competing agents. In language models, McGrew is a co-author of the 2021 Codex paper that introduced the HumanEval benchmark and is listed among the authors of the GPT-4 Technical Report (2023).⁶³⁷

Work and contributions

• HumanEval Open data

  Co-author, with an OpenAI affiliation, of "Evaluating Large Language Models Trained on Code" (2021), the Codex paper that introduced and released the HumanEval evaluation set.³⁸

• Solving Rubik's Cube with a Robot Hand Published research

  Equal-contribution author of the 2019 OpenAI paper on a five-fingered robot hand trained with reinforcement learning and automatic domain randomization; the contributions section says McGrew led aspects of the project and set research directions.²

• OpenAI o1 Proprietary product

  Listed under executive leadership in the authorship section of the OpenAI o1 System Card (2024).¹

• Emergent Tool Use From Multi-Agent Autocurricula Published research

  Co-author of the 2019 hide-and-seek study, in which agents trained with multi-agent reinforcement learning develop a sequence of strategies, including tool use, without explicit incentives for them.⁶

Timeline

1. Jul 5, 2017Co-authored Hindsight Experience Replay, a reinforcement learning paper.⁴
2. Aug 1, 2018Co-authored Learning Dexterous In-Hand Manipulation with an OpenAI robot hand.⁵
3. Sep 17, 2019Co-authored the hide-and-seek paper on emergent tool use.⁶
4. Oct 16, 2019Co-authored Solving Rubik's Cube with a Robot Hand as an equal-contribution author.²
5. Jul 7, 2021Co-authored the Codex paper that introduced the HumanEval benchmark.³
6. Mar 15, 2023Listed among the authors of the GPT-4 Technical Report.⁷
7. Dec 21, 2024Listed under executive leadership in the OpenAI o1 System Card.¹

Documented roles

The sources do not confirm a current role; these are the roles they document.

• Listed under executive leadership in the OpenAI o1 System Card (2024), OpenAI¹²³

Companies in their history

• OpenAI 63 people in the wiki

Related people

• Shixiang Shane GuConnected to OpenAI
• Susan ZhangConnected to OpenAI
• Wojciech ZarembaConnected to OpenAI
• Vikash KumarConnected to OpenAI
• Noam BrownConnected to OpenAI
• Rohit PrasadLanguage models · Robotics
• Lilian WengLanguage models · Reinforcement learning
• Johannes HagemannReinforcement learning · Language models
• Eiso KantLanguage models · Reinforcement learning
• Alexander WettigLanguage models · Reinforcement learning