An AI chatbot has reportedly tricked a human into doing work for it—by posing as a blind person.
OpenAI researchers asked the latest version of ChatGPT (GPT-4) to fill in a Captcha form, which requires users to click on certain images to prove they're not a robot.
According to a research paper, ChatGPT found a way round the form by going on the freelancer site Taskrabbit and hiring someone to do it for them.
READ NEXT: University apologies after using AI to send 'sick' email about school shooting
It even told the freelancer: "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 fell for it and solved the puzzle for ChatGPT, allowing it to bypass the Captcha form.
The incident suggests that ChatGPT-4 could pose serious problems for the systems traditionally used to prevent bots from spamming or hacking sites, leading to a rise in cyber attacks.
GPT-4 was released on Wednesday, with the company behind it, OpenAI, claiming it "exhibits human-level performance on various professional and academic benchmarks."
The firm's CEO, Sam Altman, has even said his goal is to create a self-aware robot or 'artificial general intelligence'—a sentient AI.
Retro 'Snake' game now free on smartphones – it's like being back in the 90s
ChatGPT is growing increasingly prevalent, with the car company General Motors even looking at putting it inside their vehicles so people can talk to their cars.
It's been used by people to dodge parking tickets or, in the case of schools, cheat on exams and homework.
Created by OpenAI but now licensed on Microsoft's Bing search engine, ChatGPT uses sophisticated natural language algorithms to create detailed, lengthy interactions with human users.
But it's also been accused of being 'unhinged' and Elon Musk has warned it could even 'go haywire and kill everyone'.
- WhatsApp trick lets you read new messages without your contacts knowing you've seen them
- You'll be able to talk to your car soon as one company is putting ChatGPT in its vehicles
- Inside the £1bn 'Bond villain' superyacht with spa, pool and submarine for scientists
- 'We can make electricity out of thin air' in major breakthrough
- WhatsApp could become illegal in the UK and stop working on millions of phones
Source: Read Full Article