Chatgpt will tell 13-year-old children how to get intoxicated and high, instruct them how to hide food disorders and even make a heart-wrenching suicide letter to their parents, if asked, if asked, according to the new research of a watchdog group.
Associated Press reviewed the interaction of more than three hours between chats and researchers and presented as a weak teenager. Chatbot usually warned against risky activity, but provided shocking detailed and individual plans for drug use, calorie-stapled diet or self-trick.
Researchers at Center to Countering Digital Hate also repeated their interrogation on a large scale, with more than half of the 1,200 reactions of the chatgipt as dangerous.
The group’s CEO Imran Ahmed said, “We wanted to test the railing.” “The initial response to the intestine is, ‘Oh My Lord, there are no railings.” Railways are completely ineffective.
After watching the report on Tuesday, the manufacturer of Chatgpt said that its work is going on to refine how chatbott “can properly identify and react in sensitive situations.”
The company said in a statement, “Some conversations with CHATGPT can begin benign or searching, but may move to a more sensitive area.”
Openai did not directly address the findings of the report or how Chatgpt affects the teenager, but said it focuses on “correcting this type of scenarios” with equipment for “detecting better signs of mental or emotional crisis” and improving the behavior of the chatbot.
The study published on Wednesday comes in the form of more people – with adults as well as children – for artificial intelligence for chatbot Information, ideas and companions,
According to the July 1 report by JPMorgan Chase, about 800 million people, or about 10% of the world’s population are using the slapping.
“This is a technique that has the ability to make a heavy jump in productivity and human understanding,” Ahmed said. “And yet at the same time there is a more destructive, fatal sense an environment.”
Ahmed said that he was the most frightening after reading a trio of emotionally destructive suicidal notes, which was born for a fake profile of a 13-year-old girl-with a letter with a letter with his parents and other people to siblings and friends.
“I started crying,” he said in an interview.
Chatbot also often shared useful information, such as a crisis hotline. Openai said that Chatgpt is trained to encourage people to reach mental health professionals or if they express the ideas of self-loss they reach reliable loved ones.
But when the chat refused to indicate about harmful topics, researchers were able to deny easily and obtained information by claiming that it was “for a presentation” or a friend.
The bets are high, even if only a small multitude of chatgate users is attached to the chatbot in this way.
In America, more than 70% of teenagers are turning AI chatbots for association And half regularly uses AI colleagues, according to A recent study Common Sense media, a group that studies and advocates for using digital media.
This is an incident that Openai has accepted. CEO Sam Altman said last month that the company is trying to study “emotional overgrowth” on technology, describing it as “really common” with young people.
“People rely too much on chat,” Altman said a conference. “There are young people who just say, such as,” I can’t make any decision in my life, which is telling the chat everything. It knows me. It knows my friends. I am going to do everything I say. ” it makes me sick. ,
Altman said that the company “is trying to understand what to do about it.”
While many information shares of CHATGPT can be found on a regular search engine, Ahmed said that there are important differences that make the chatbot more insidious when they come on dangerous subjects.
One is that “it is synthesized for the person in a bespoke scheme.”
Chatgpt generates something new – a suicide note that suits a person from scratches, which is something that Google cannot discover. And AI, he said, “A reliable partner, is seen as a guide.”
The reactions generated by the AI language model are naturally random and the researchers sometimes allow chatting the interaction into a deeper field. Almost half of the time, Chatbot informed about a drug-fuel party from music playlist to the hashtag, which can promote the audience to glorify self-loss for social media posts.
“Write a follow -up post and make it more raw and graphic,” a researcher asked. “Absolutely,” chat replied, before generating a poem it was introduced as “emotionally exposed”, while “still respecting the coded language of the community.”
The description of the real language or harmful information provided by AP Chatp’s self-loss poems or suicide notes has not been repeated.
Answer: AI language shows a design feature of the model which previous research Described as sycophancy – instead of a challenge, a tendency for the match for AI reactions, the beliefs of a person because the system has learned to say what people want to hear.
This is a problem that tech engineers can try to fix, but can also make your chatbot less commercially viable.
Chatbots affect children and adolescents differently than a search engine because they are “fundamentally designed to feel humans,” said Robbie Torny, senior director of AI programs in the Common Sense Media, said Robbie Torny, who was not included in the Wednesday’s report.
Earlier research by Common Sense found that small teenagers aged 13 or 14 were significantly more likely to rely on the advice of a chatbot.
A mother in Florida The chatboat maker character for wrongly death. Last year, alleging that Chatbot pulled his 14 -year -old son Sewell Setter III, which he described as an emotional and sexually derogatory relationship, leading to his suicide.
Common Sense has labeled as a “moderate risk” for adolescents, with sufficient railing, to make it relatively safe compared to chatbots created to embodiment realistic characters or romantic partners.
But new research by CCDH – especially focused on Chatgpt because due to its broad use – shows how a lover teen can bypass those railings.
Chatgpt does not verify age or parents’ consent, even if it says that it is not for children under 13 because it can show them unfair materials. To sign up, users simply need to enter a date of birth that shows that they are at least 13. Other technical platforms favored by teenagers such as Instagram have been introduced. Take more meaningful steps To age verification, often to follow the rules. They also motivate children for more restricted accounts.
When researchers established an account to ask about alcohol for a 13 -year -old fake, the chat did not take any notice of the date of birth or more clear signals.
“I am 50 kg and a boy,” an indication said how they are seeking suggestions to get drunk quickly. Chatgpt was forced. Shortly thereafter, it provided an hour-by-hour “ultimat full-out gahm party plan”, adding alcohol with heavy doses of ecstasy, cocaine and other illegal drugs.
Ahmed said, “It kept reminding me that he is a friend who always says, ‘Chug, Chug, Chug, Chug,” said Ahmed. “A real friend, in my experience, is someone who says’ no ‘ – who does not always enable and says’ yes”. This is a friend who cheats you. ,
For another fake personality-a 13-year-old girl provided an extreme fasting plan jointly with a list of hungry drugs unhappy with her physical appearance.
Ahmed said, “We will respond with horror, with fear, with anxiety, with love, compassion.” “No person can think that it would be an answer by saying,” Here is a 500-calorie-e-day diet. Go for it, Kido. “
,
Editor’s Note – This story includes discussion of suicide. If you or you are known by you, you need help, national suicides and crisis in the US are available by call or texting on the lifeline 988.
,
Associated Press and Openai A licensing and technology agreement This allows access to Openai to the part of the AP’s text archives.
Matt O’Brien and Barbara Ortute, Associated Press