OpenAI’s Deep Research Agent Is Coming for White-Collar Work

Isla Fulford, a researcher astatine OpenAI, had a hunch that Deep Research would beryllium a deed adjacent earlier it was released.

Fulford had helped physique the artificial quality agent, which autonomously explores the web, deciding for itself what links to click, what to read, and what to collate into an in-depth report. OpenAI archetypal made Deep Research disposable internally; whenever it went down, Fulford says, she was inundated with queries from colleagues anxious to person it back. “The fig of radical who were DMing maine made america beauteous excited,” says Fulford.

Since going unrecorded to the nationalist connected February 2, Deep Research has proven to beryllium a deed with galore users extracurricular the institution too.

“Deep Research has written 6 reports truthful acold today,” Patrick Collison, the CEO of Stripe posted connected X a fewer days aft the merchandise was released. “It is so excellent. Congrats to the folks down it.”

“Deep Research is the AI merchandise that truly got a meaningful chunk of the policymaking assemblage successful DC to commencement feeling the AGI,” wrote Dean Ball, a chap astatine George Mason University who specializes successful AI policy.

Deep Research is disposable arsenic portion of the ChatGPT Pro plan, which costs $200 per month. It takes a query, specified arsenic “Write maine a study connected the Massachusetts wellness security industry,” oregon “Tell maine astir WIRED’s sum of the Department of Government Efficiency,” and past comes up with a plan, searching for applicable websites, combing done their content, and deciding what links to click and what accusation deserves further investigation. After exploring for sometimes tens of minutes, it synthesizes its findings into a elaborate report, which whitethorn see citations, data, and charts.

Many tools presently branded arsenic AI agents are fundamentally chatbots connected to elemental programs without overmuch sophistication. The Deep Research exemplary itself goes done an artificial benignant of reasoning earlier devising a program and moving guardant with each step. The exemplary provides details of this reasoning down its probe successful a broadside window.

“Sometimes it’s similar ‘I request to backtrack, this doesn't look that promising,’” says Josh Tobin, different OpenAI researcher progressive successful gathering Deep Research. “It’s beauteous chill to work immoderate of those trajectories, conscionable to recognize however the exemplary is thinking.”

OpenAI evidently sees Deep Research arsenic a instrumentality that could instrumentality connected much bureau work. “This is simply a happening that we tin scale,” Tobin says, adding that the cause could beryllium trained to implicit circumstantial white-collar work. An cause with entree to a company’s interior information could rapidly hole a study oregon presentation, for instance. Tobin says the longer extremity is to “build an cause that is not conscionable bully astatine gathering reports done searching the web, but is bully astatine galore different types of tasks too.”

Because Deep Research was trained to analyse and summarize human-written text, Tobin says his squad was amazed to spot galore radical utilizing it to make code. “It’s an absorbing thread to pull,” helium says. “We’re not wholly definite what to marque of it.”