I’m a software engineer and applied researcher living in San Francisco. I’m currently a Member of Technical Staff and engineering manager at OpenAI.

I build tools that extend the reach of human comprehension and mastery. The world is growing increasingly complex for many reasons, including AI. I build things that make it possible for people to understand what’s going on and take more effective action.

I’ve worked in a variety of research and production settings:

• From 2014-2016, I was a product manager at Microsoft working on Bing search relevance. My job was to make Bing better at understanding queries and the web so it could return more relevant results. I thought of myself as a scout and portfolio manager. There were many different ways we could improve relevance with different levels of risk and reward. My job was to spot the most promising opportunities and structure the portfolio to maximize long-term gains in relevance while also delivering a consistent stream of wins along the way. I was among the first at Bing to realize that modern deep learning was going to revolutionize natural language understanding and helped start many of the first deep learning projects in Bing.
• From 2016-2024, I worked in Google Research (a part of the org that later merged into Google DeepMind). I started as a product manager, working with a team developing an end-to-end reinforcement-learned conversational AI six years before ChatGPT. A year or so in, I realized that I wanted to shift into a more hands-on role as a software engineer. It was clear by that point that deep learning was one of the most important technologies in the history of computing and I wanted to be one of the people building with it directly. I went on to work on a ton of different things:
  • I built systems that summarized opinion-shaped free text data across Google products like Shopping and Maps.
  • I helped build a system for training text classifiers and embedding models that served hundreds of billions of inference requests every day in Search, YouTube, Play, and many other products.
  • I led a small team to build a tool that used embedding visualizations to make sense of large text datasets, helping trust & safety teams identify novel patterns of abuse.
  • On the People & AI Research (PAIR) team, I helped build a tool for analyzing and improving Gemini post-training and eval datasets.
• In 2024, I joined OpenAI, where I now lead a team of software engineers building tools that help analysts and investigators detect and disrupt high-severity misuse of OpenAI’s products.

I have a blog because it’s useful to think out loud. It helps me think better and helps me use my past thoughts as a scaffold for future ones.

It also makes me and my work more legible to others, which increases the chance that I’ll meet interesting people. If you read the things I write here and have any inkling to reach out, please do.

Contact

I love hearing from people with shared interests.

The most reliable way to reach me is to email me at [email protected]

If you prefer, you can also DM me on Twitter: @gradysimon