“To put it bluntly, it felt at times like watching a 12-year-old use my computer.”
That’s Dr Junade Ali’s assessment of the new OpenAI browser, which was released on Tuesday evening.
The browser, called Atlas, appears designed to challenge Google Chrome and potentially supplant the most profitable segment of Google’s parent company, Alphabet.
But after testing out Atlas, AI experts have told Sky News one of its key elements feels “very clunky”.
As well as fully incorporating ChatGPT into the browser so it follows you around the internet, the “most exciting part” of the browser is its AI agent, according to Dr Andrea Barbon from the University of St Gallen.
The feature is currently only available in “preview mode” for certain users.
“When you activate the agent, you can ask it to do something, and then it will take control of your computer or your mouse,” he explained after trying it out.
“It will start clicking around on the website to perform the task that you requested,” he said – but the feature left him disappointed, and he stopped using the browser within minutes.
“I tried a couple of websites where the workflow is a bit complex and ChatGPT was just not able to handle it, not at all,” he said.
“After a few minutes, I closed the browser and I didn’t uninstall it, but I could – I’m not going to use it, right?
“Maybe I will use it in the future, if they release versions that are actually working,” he said.
Dr Ali, a fellow at the Institute of Engineering and Technology, said the AI agent would “struggle around a little bit to try and do the task in the most effective way and it would sometimes get stuck”.
“It definitely seemed very primitive – but it is a really neat concept,” he said.
OpenAI said its agent mode is “an early experience and may make mistakes on complex workflows”.
“We’re rapidly improving reliability, latency and complex task success.”
Will it…
