Christopher Ong | Research Acceleration: The View Inside OpenAI

**Christopher Ong | Research Acceleration: The View Inside OpenAI**

In the last year, AI models have rapidly improved on public benchmarks measuring their ability to carry out research tasks autonomously. Such performance suggests that models may be capable of automating parts of the AI R&D process itself, thereby accelerating the pace of AI progress. However, there is still little empirical data on how these new capabilities have changed the research process within the AI labs at the research frontier. This talk will share findings from OpenAI's recent report on AI-powered research acceleration, which aims to help fill this gap. We will also discuss future research directions for the economics of recursive self-improvement (RSI).

The report documents several stylized facts. First, models are increasingly capable of autonomously completing research tasks that would take a human researcher several hours or even days to finish. However, completing longer tasks remains challenging and more frequently requires human intervention. Second, agentic use is rapidly increasing, such that total hours worked by the agent workforce is several times that of the human workforce. Finally, researchers intensely use agents for execution-oriented tasks such as writing code, reviewing technical work, and managing training runs, but high-level planning remains primarily bottlenecked on humans.

**Details:**

Time: 12:00 pm - 1:00 pm PT

Location: Gates Computer Science Building, Room 119, 353 Jane Stanford Way, CA 94503

You can find additional event details, speaker bio, and information about upcoming seminars on our website.