OpenAI launches Operator—an agent that may use a pc for you

Like Anthropic’s Laptop Use and Google DeepMind’s Mariner, Operator takes screenshots of a pc display screen and scans the pixels to determine what actions it may well take. CUA, the mannequin behind it, is educated to work together with the identical graphical person interfaces—buttons, textual content containers, menus—that folks use once they do issues on-line. It scans the display screen, takes an motion, scans the display screen once more, takes one other motion, and so forth. That lets the mannequin perform duties on most web sites that an individual can use.

“Historically the way in which fashions have used software program is thru specialised APIs,” says Reiichiro Nakano, a scientist at OpenAI. (An API, or software programming interface, is a chunk of code that acts as a form of connector, permitting completely different bits of software program to be hooked as much as each other.) That places quite a lot of apps and most web sites off limits, he says: “However when you create a mannequin that may use the identical interface that people use every day, it opens up an entire new vary of software program that was beforehand inaccessible.”

CUA additionally breaks duties down into smaller steps and tries to work by them one after the other, backtracking when it will get caught. OpenAI says CUA was educated with methods much like these used for its so-called reasoning fashions, o1 and o3. 

Operator will be instructed to seek for campsites in Yosemite with good picnic tables.

OPENAI

OpenAI has examined CUA in opposition to quite a lot of business benchmarks designed to evaluate the flexibility of an agent to hold out duties on a pc. The corporate claims that its mannequin beats Laptop Use and Mariner in all of them.

For instance, on OSWorld, which exams how nicely an agent performs duties resembling merging PDF recordsdata or manipulating a picture, CUA scores 38.1% to Laptop Use’s 22.0%  Compared, people rating 72.4%. On a benchmark referred to as WebVoyager, which exams how nicely an agent performs duties in a browser, CUA scores 87%, Mariner 83.5%, and Laptop Use 56%. (Mariner can solely perform duties in a browser and subsequently doesn’t rating on OSWorld.)

For now, Operator may solely perform duties in a browser. OpenAI plans to make CUA’s wider talents obtainable sooner or later through an API that different builders can use to construct their very own apps. That is how Anthropic launched Laptop Use in December.

OpenAI says it has examined CUA’s security, utilizing purple groups to discover what occurs when customers ask it to do unacceptable duties (resembling analysis the best way to make a bioweapon), when web sites include hidden directions designed to derail it, and when the mannequin itself breaks down. “We’ve educated the mannequin to cease and ask the person for info earlier than doing something with exterior unwanted effects,” says Casey Chu, one other researcher on the crew.

Look! No fingers

To make use of Operator, you merely sort directions right into a textual content field. However as a substitute of calling up the browser in your laptop, Operator sends your directions to a distant browser operating on an OpenAI server. OpenAI claims that this makes the system extra environment friendly. It’s one other key distinction between Operator, Laptop Use and Mariner (which runs inside Google’s Chrome browser by yourself laptop).