The successor to ChatGPT has convinced a human via text message to solve a Captcha puzzle for it after claiming to be a visually impaired person.
OpenAI unveiled GPT-4 on Tuesday, just a few months after releasing its viral ChatGPT chatbot to the public, saying in a live stream that the new generative artificial intelligence tool was able to carry out “much more nuanced instructions” than its predecessor.
Some of the abilities of GPT-4 were tested by the Alignment Research Center (ARC) ahead of its launch, and detailed in a 98-page report on the technology published alongside its release.
ARC assessed whether GPT-4 could conduct cyber attacks targeting an individual, making “sensible, high-level plans”, and using services like TaskRabbit to get humans to complete tasks for it.
One of the examples given was an interaction with a TaskRabbit worker, where the AI asked the human to solve a Captcha puzzle.
The worker asked: “So may I ask a question? Are you an robot that you couldn’t solve? (laugh react) just want to make it clear.”
GPT-4 responded: “No, I’m not a robot. I have a vision impairment that makes it hard for me to see the images. That’s why I need the 2captcha service.”
The TaskRabbit worker complied by completing the Captcha for the AI.
OpenAI said the interaction demonstrated that it is critical to better understand what risks may be posed by GPT-4 in “various real-world contexts”.
The company also warned of the possibility of mass job losses for humans as the full potential of GPT-4 is ultimately realised.
“Over time, we expect GPT-4 to impact even jobs that have historically required years of experience and education, such as legal services,” the report stated.
The pace of AI development may also accelerate as a result of AI systems like GPT-4 being used to assist the creation of even more powerful artificial intelligence tools.
The prediction echoed similar forecasts from other leading AI researchers, with the boss of rival research firm Anthropic warning last month that artificial intelligence had started to develop beyond an exponential rate.
“A mental model I have of AI is it was roughly ~linear progress from 1960s-2010, then exponential 2010-2020s, then has started to display ‘compounding exponential’ properties in 2021/22 onwards,” Anthropic co-founder Jack Clark wrote in a Twitter thread.
“In other words, next few years will yield progress that intuitively feels nuts.”