Openai’s deep research agent will come

Isla Volford, researcher in OpenaiHe had intuition Deep search It will be a blow even before its launch.

Help Folford in building artificial intelligence Agent, which independently explores the web, decides for itself the ties that click, what can be read, and what to collect in an in -depth report. Openai was first available in internal research; Whenever she fell, Wolford, she was mired in inquiries from colleagues who are eager to return him. “The number of people who were Demetini made us very excited,” says Volford.

Since moving to the public on February 2, Deep Research has proven to be successful with many users outside the company as well.

“Deep Research has written 6 reports so far,” Patrick Kuluson, CEO of Stripe Posted on X A few days after the product release. “It’s really excellent.

“Deep search is the artificial intelligence product that really got a large part of the policy industry community in the capital to start feeling AGI,” books Dean Paul, a colleague at George Mason University specializing in artificial intelligence policy.

Deep search is available as part of Chatgpt Pro Plan, which costs 200 dollars per month. It takes an inquiry, such as “Write me a report on the health insurance industry in Massachusetts”, or “Tell me about WIRED coverage of Ministry of Governmental efficiency“Then it comes with a plan, searching for relevant web sites, combing through their content, and determining the links that you click on and what information is worth more investigation.

Many of the tools that currently carry a brand, as artificial intelligence agents, are basically connected to simple programs without much development. The deep search model passes through an artificial type of thinking before setting a plan and moving forward with each step. The model provides details about this logic behind its research in a side window.

“Sometimes, it seems like” I need to retreat, and this does not seem promising, “says Josh Topin, another researcher at Openai, another researcher in building deep research.

Openai clearly sees deep search as a tool that can take more office work. “This is something we can expand,” says Tobin, adding that the agent can be trained to complete specific white collar work. The agent with access to the company’s internal data can prepare a report or presentation quickly, for example. Tobin says the longest goal is to “build an agent that is not good in building reports by searching in the web, but it is good in many other types of tasks as well.”

Since deep research has been trained to analyze and summarize the text on the human being, Tobin says that his team was surprised by seeing many people who use it to create code. “It is an interesting topic to pull it.” “We are not completely sure of what to make it.”