Isla Fulford, a researcher at Openai, had a hunch that deep research could take a hit even before it was released.
Fulford helped build an AI agent that explores the network independently, deciding on the links to click, what to read, and what to link into in-depth reports. Openai first provides in-depth research internally; Fulford says whenever it falls, she is overwhelmed by the questions her colleagues are eager to come back. “My number makes us very excited,” Fulford said.
Since residency to the public on February 2, in-depth research has also proven to be a hit with many users outside the company.
“In-depth research has written 6 reports so far,” Patrick Collison is Stripe CEO published on X a few days after the product was released. “It’s really great.
“In-depth research is AI products that really gain access to DC’s meaningful policy-making community to start feeling AGI,” wrote Dean Ball, a researcher at George Mason University.
Part of the Chatgpt Pro program offers in-depth research and costs $200 per month. It requires inquiries such as “write me a report on the Massachusetts health insurance industry,” or “tell me an introduction to the wired coverage of government efficiency departments,” and then propose plans, search for relevant websites, colour through content and determine which links to click on and other information should be investigated. Sometimes after exploring for dozens of minutes, it synthesizes its discoveries into detailed reports, which may include references, data, and charts.
Many of the tools currently using it as an AI proxy are essentially chatbots without the need for too high chatbots to connect to simple programs. In-depth research on the model itself goes through a kind of artificial reasoning before making plans and moving forward. The model provides detailed information on this reasoning after the side window study.
“Sometimes, it’s like, ‘I need to go backtrack, which doesn’t seem so hopeful’, said Josh Tobin, another OpenAi researcher who was involved in building an in-depth study. “It’s really cool to read some of these trajectories just to understand the idea of the model.” ”
Openai clearly sees in-depth research as a tool that can work more office work. “This is something we can expand on,” Tobin said, adding that agents can be trained to do specific white-collar work. For example, an agent with access to internal company data can quickly prepare a report or presentation. The longer goal, Tobin said, is to “build an agent that is not only good at building reports through search networks, but is good at many other types of tasks.”
Since the in-depth study has been trained to analyze and summarize personal texts, Tobin said his team was surprised to see many people using it to generate code. “It’s an interesting thread,” he said. “We’re not completely sure what to do.”
