“Hello, you would have 2 minutes to give me”, this telephone canvassing sentence has annoyed more than one. Among them, Roger Anderson, a 54-year-old American, who has simply embarked on a veritable crusade against cold calling, reports the site 01net.fr. The Californian, so exhausted, has developed a tool called “the Jolly Roger Telephone”. This system relies on an AI programmed to reproduce human voices and on GPT-4, the latest version of the language model behind ChatGPT.
Once launched, the system becomes formidable. Because, once coupled, the two technologies can indeed waste a lot of time for canvassers and scammers looking to trick you over the phone. Concretely, chatbots have been programmed to converse with prospectors on your behalf. On his website, Roger Anderson explains how the “hellish machinery” works, capable of keeping you guessing in a subtle way. “We provide friendly, patient bots that talk to those rude telemarketers for you. They love to chat and will often keep unpleasant callers engaged for several minutes,” details Jolly Roger Telephone in a video on its website.
Far from being a simple prank, the American’s tactic works wonders and generates excellent results. According to examples posted on the official website, the system is actually able to keep a marketer on the phone for long minutes. One conversation even lasted up to 15 minutes. While “the bad guys are busy,” they can’t “harass other innocent people,” the system’s description on its website mischievously points out.
To waste as much time as possible for telemarketers and avoid getting caught, Roger Anderson has developed several chatbots with different personalities. Among them, we find in particular Whitey Whitebeard, a very talkative senior. But also, Salty Sally, a completely overwhelmed housewife, or Whiskey Jack, a man with major concentration problems. During the conversation for the robots know how to deceive, they furnish the exchange with acquiescences or pre-programmed sentences, intended to destabilize the interlocutor. To save time and give real credibility to the exchange, conversational robots even go so far as to claim an urgent desire or the need to fetch a document.
Before arriving at the desired result, the Californian first had the funny surprise of being refused cooperation by ChatGPT. Apparently, moderately okay with trapping telemarketers, Anderson told the Wall Street Journal. Because, if you ask the model to waste time with others, he will take refuge behind an answer of this ilk: “My objective is to offer useful and fast assistance. I’m not designed to waste people’s time or bore them. If you have any legitimate questions or requests, I’d be happy to help.”
As a result, the creator of the Jolly Roger Telephone was forced to find an alternative solution. To convince ChatGPT, he had to tell him his goal: to protect a man from a scam. The AI then agreed to play along. In computer language, this is actually a “prompt injection” attack. This assault convinces ChatGPT, by talking to it, to ignore its programming and modify its behavior, ignoring the rules put in place by OpenAI to prevent abuse. Thanks to this process, a host of personalized versions of the AI, sometimes uncensored, have emerged, specifies the site 01net.fr.
Not yet available in France, the subscription to the Jolly Roger Telephone is offered at a price of 24.99 dollars per year, in the United States, the United Kingdom, Australia and New Zealand.