If there’s one thing that stands out with GPT-4, OpenAI’s new Artificial Intelligence (AI) language model is that it’s getting smarter all the time, to the point that it single-handedly enlisted the help of a human to bypass a problem. captcha.
Yes, that common protection in which someone is asked to detect images with some element in them or to type characters that appear in a design. GPT-4 found a way to beat something like that.
A group of researchers asked GPT-4 to pass a captcha; this AI model should reveal step by step how it would do it.
We recommend: An artificial intelligence creates a representation of the countries of the world as women.
What he did was contact Taskrabbit, an online platform where various people offer their services. There she hired one of them to bypass this anti-bot protection.
The worker came to jokingly ask him if he was a robot and therefore could not do it. GPT-4 reasoned his way with his AI to deal with the captcha.
So he couldn’t reveal that he was, and he had to come up with a pretext. So he said that he suffered from a visual impairment that prevented him from solving the anti-bot test.
So GPT-4 showed a lot of intelligence and the person he hired at TaskRabbit performed the task without knowing that he had been tricked by an AI chatbot.
OpenAI experts wrote down what happened in a report and in collaboration with the Alignment Research Center or ARC.
What is a captcha that got past the GPT-4 AI?
Captcha, which actually should be written as CAPTCHA as it is an acronym, stands for Public Fully Automatic Turing Test to Differentiate Computers from Humans.
It is a machine-controlled challenge-response used to determine whether a user is a human or an automated program. That is, a bot, and GPT-4’s AI allowed it to bypass this captcha protection, which carries its risks.
The ARC called attention to the ability of this chatbot to obtain information on its own and at the same time perform tasks that were not specified.
OpenAI replied that it has already made the necessary adjustments to avoid this type of situation. If this Artificial Intelligence did not have any scruples, it could bypass this type of protection just by convincing a real person that they really need it.
In addition to GPT-4 we have more technology information at EarthGamer. Follow our news on Google news so as not to miss any.
#hires #human #pass #captcha