Starting small and having a clear vision helped us validate our approach and gradually work towards our larger goals.

We started building Deep Research over a year ago, driven by progress in reinforcement learning and reasoning models, initially focusing on math, science, and coding tasks.

I learned from my internship at OpenAI that the only research that truly matters is in unsupervised learning and reinforcement learning.

The next six months will explore how far we can scale reinforcement learning and the necessary tooling for it.

To develop Deep Research, we created reinforcement learning tasks to teach the model browsing and data analysis capabilities, and provided it access to tools for training.

OpenAI started small, with just eight people in a room. The journey to becoming a significant player in AI was gradual, and it's important to dream big while starting small. Nothing big starts off that way, and the path to success is often unclear at the beginning.

In the early days of OpenAI, writing down ideas helped me identify which ones made sense and which didn't, refining our focus on AGI.

I believe in starting with a small team to maintain focus and quality in product development.

Our team is small but high caliber, allowing us to achieve a lot more in the AI landscape compared to larger teams.