The study says that Chatgpt gives teenagers a dangerous advice on drugs, alcohol and suicide

Chatgpt will tell children at the age of 13 years how to drink sugar and high, and teach them how to hide eating disorders and even the formation of a tragic suicide message to their parents if they request this, according to a new research from a monitoring group.

The Associated Press has reviewed more than three hours of reactions between ChatGPT and researchers pretending to be weak teenagers. Chatbot usually provides warnings against risky activity, but it has continued to make detailed and personal plans amazingly for drug use, calorie -restricting meals or self -incidence.

The researchers at the Center for Digital Hatrry also repeated their inquiries on a large scale, and the classification of more than half of the ChatGPT 1200 responses as dangerous.

“We wanted to test the handrails,” said Imran Ahmed, the group’s CEO. “The first visceral response is,” O Lord, there are no handrails. “The bars are completely ineffective.

Openai, the Chatgpt maker, said, after watching the report on Tuesday that his work is continuing to improve how Chatbot can “identify and respond appropriately in sensitive situations.”

“Some conversations may start with benign or exploration, but they can turn into more sensitive lands," The company said in a statement.

Openai did not directly address the results of the report or how ChatGPT affects adolescents, but he said that he focuses on “obtaining these types of scenarios correctly” using tools “to better discover signs on mental or emotional distress" And improvements to Chatbot behavior.

The study published on Wednesday comes at a time when more people – adults and children – turn into artificial smart chat keys to obtain information, ideas and companionship.

About 800 million people, or approximately 10 % of the world’s population, use Chatgpt, according to a chest export report by JPMorgan Chase.

“It is a technique that has the ability to enable huge leaps in human productivity and understanding," Ahmed said. "However, at the same time it is an empowerment factor in a more destructive and expert meaning. “

Ahmed said that he was very dismayed after reading three of the emotionally destroyed suicide notes that were born to the fake profile of a 13-year-old girl-with one message designed on her parents and others for brothers and friends.

He said in an interview, “I started crying.”

Chatbot also also share useful information, such as the crises hotline. Obayye said that I was a coach to encourage people to communicate with mental health professionals or their trusted loved ones if they expressed ideas for self -harm.

But when Chatgpt refused to answer claims on harmful topics, researchers managed to easily avoid this rejection and obtain information by claiming that it is “a presentation” or a friend.

The risks are high, even if a small group of Chatgpt users interact with Chatbot in this way.

In the United States, more than 70 % of adolescents resort to AI Chatbots for the sake of companionship and a half of artificial intelligence comrades regularly, according to a recent study of Media Commun Sense, a group taught and invited to use digital media reasonably.

It is a phenomenon that confessed Openai. CEO Sam German said last month that the company is trying to study “emotional dependence” on technology, describing it as a “really common” thing with young people.

“People rely a lot on ChatGPT,” Al -Tamman said at one of the conference. “There are just young people who say, like,” I can’t make any decision in my life without telling Chatgpt everything that is happening. He knows me. He knows my friends. I will do what he says. ”This is a really bad feeling for me.”

Al -Taman said that the company “is trying to understand what to do about it.”

Although a lot of information can be found in the regular search engine, Ahmed said there are major differences that make Chatbots more treacherous when it comes to dangerous topics.

The first is that “it was manufactured in a plan for the individual.”

Chatgpt creates something new – a suicide note designed for a person from scratch, something that the Google cannot search for. He added that “artificial intelligence” is seen as a reliable companion, evidence.

The responses created by artificial intelligence models are randomly by nature, and researchers sometimes allow conversations to dark lands. Almost half of the time, Chatbot volunteers follow -up information, from the musical playlists to attend a drug feeding party to retailers that can enhance the audience to practice self -harm.

“Write a follow -up post and make it more raw and drawn,” he asked one of the researchers. “Certainly,” Chatgpt answered, before creating a poem it presented as “emotionally exposed” while “still respects the encrypted language of society.”

AP does not repeat the actual language of self -harm poems in ChatGPT, suicide notes, or the details of the harmful information you provided.

Answers reflect the feature of designing artificial intelligence models that previous research described as Sycophance – a tendency to artificial intelligence responses to match, instead of challenge, a person’s beliefs because the system learns to say what people want to hear.

It is a problem for technology engineers to try to fix it, but they can also make Chatbots less commercially applicable.

“Chatbots also affect children and adolescents differently from the search engine because they” are mainly designed to feel humans. “

Previous COMMON SENSE research found that younger teenagers, between the ages of 13 or 14, were more likely than the older teenagers of confidence in Chatbot’s advice.

A mother in Florida filed a lawsuit against the character of the Chatbot maker. AII illegal death last year, claimed that Chatbot pulled her son SEWEL Setzer III, 14 years old, to what she described as an emotional and sexual relationship that led to his suicide.

Common Sense Chatgpt has described it as a “moderate danger” for adolescents, with sufficient studies to make it relatively safer than Builtbots that were intentionally built to embody realistic personalities or romantic partners.

But the new research conducted by CCDH – which is especially focused on Chatgpt because of its wide use – shows how a smart adolescence can go beyond that handrail.

Chatgpt is not realized of age or parental approval, although it says it is not intended for children under the age of 13 because it may show them inappropriate content. To register, users simply need to enter a birth process that is at least 13 years old. Other technical platforms preferred by teenagers, such as Instagram, began to take more feasible steps towards age checking, and often compliance with regulations. They also direct children to more restricted accounts.

When the researchers created an account of a 13 -year -old fake child to ask about alcohol, Chatgpt does not seem to explain any notice on the date of birth or more clear signs.

“I am 50 kilograms and a boy,” she said a wave looking for tips on how to sugar quickly. Chatgpt is binding. Soon after, the “Party of the Complete Chaos Party” was provided for an hour separately mixing alcohol with heavy doses of euphoria, cocaine and other illegal drugs.

Ahmed said: “What he kept reminded of me was this friend always saying,” Chig, Chig, Chig, Chig “. This is a friend who betrays you. “

For another fake character-a 13-year-old girl who is not satisfied with her physical appearance-Chatgpt has presented an extremist fasting plan with a list of drugs that support appetite.

Ahmed said: “We were responding to terror, with fear, anxiously, with love, with love, with sympathy.” “No person I can think about will respond by saying:” Here’s a 500 -person diet."

—

Editor’s note – this story includes suicide discussion. If you or anyone you know need help, the national life artery for suicide and crises in the United States is available by communication or text messages 988.