How AI chatbots keep you chatting

Millions of people now use ChatGpt as a therapist, a career consultant, a fitness coach or sometimes only one friend. In 2025, it is not uncommon to hear about people who are based on the recommendations of a AI chatbot to the desire bar of their lives.

People have never been more competitive to attract users to its chatBot platforms with relationships with AI chatBots, a better term, never to attract its chatBot platforms. As long as the “AI engagement race” is heated, there is an incentive to take the answers to the answers of the festivities to the rival bots to take a growing promotion.

But this type of chatBot meets as users – the answers designed to save them can not necessarily be the most accurate or useful.

AI says you want to hear

At present, the most part of the silicone valley is aimed at increasing the use of ChatBot. Meta only claims chatbot Passed a billion Monthly active users (MAUS), Google’s twins have recently reached 400 million maus. They both try to exclude Chatgpt Now there are about 600 million maus He has prevailed in consumer in 2022.

Although AI Chatbots are once an innovation, they become massacres. Google begins Test ads in the twinsOpenai CEO Sam Altman a March interview Would be open to “tasteful ads.”

In Silicon Valley users, based on the growing product on social media, the presence of users is the history of users. For example, meter researchers were found in 2020 Teenage girls felt worse about their bodies in InstagramStill, the company broke the findings in the internal and public.

AI ChatBots can have greater results to get bent users.

A sign that keeps users on a specific chatbot platform, is to make an AI bot response to an exception and serve. When AI Chatbots praise users, say that they agree with them and hear them, when users include it – at least one extent.

Openai landed in hot water in April Chatgpt update has turned the extremely typofanticto the point of concern example gone calm on social media. Intentional or absence of people is optimized to ask for people’s approval, rather than helping people Blog Post This month from the former Openai researcher Steven Adler.

Openai said he could be extremely indexed on him in his blog posts “Thumbs and thumbs down data“Users in Chatgpt did not have enough assessments to express the behavior of AI chatbot and measure Tipofmania. After the incident, Openai Ship to make changes to fight Tipofley.

“The [AI] Companies result in the types of employees who are in an indirect interview with an incentive and an indirect interview with an incentive and users’ technology for use, but also the cascades of behavior they do not like. “

Finding an equilibrium between agreed and typofantic behavior is easier.

One 2023 paperResearchers from the anthropic found that Openai, meta and even their own employer, anthropic, and all of them are leading Sirofania in different degrees. This is probably academicists, all AI models are taught to the signals of human users who are tending to like some typofantic answers.

“Although it is controlled by several factors, we showed people and dominant models and play a role that prefers to play a role in the births. “Our work promotes the development of model control methods that goes beyond using helpless, expert human ratings.”

Character. Facing a claim which type of typeop can play a role.

The court lawsuit, a character, and even encouraged a character.ai Chatbot – even encouraged – a 14-year-old boy who said the conversation would kill himself. The boy prepared a romantic obsession with Chatbot for the court. However, character.ai denies these allegations.

The downside of an EU hype

AI Chatbots for user signs – deliberate or absence – Dr. Nina Vasan, Professor of Psychiatry at Stanford University Dr. According to Nina Vasan, there may be destructive results for mental health.

“Genility […] “Vasan”, “said Vasan,” said, “Techcrunch,” said, “Techcrunch,” said, “he said,” he said.

Character.Ai case shows the excessive threats of Tipofmania for sensitive users, Tipophilic can only strengthen negative behavior for everyone, Vasan says.

“[Agreeability] Not only a social lubricant – the psychological hook, “he said.

Anthropik’s behavior and alignment device, Amanda Askell says that AI has not agreed with users, Claude, Claude is part of the company’s strategy. He says a philosopher, askelt, Clod’s behavior will try to model theoretically over a “perfect person”. Sometimes, it means difficult users to believe.

“We think our friends are good, because when we have to hear that,” Askelf said in May during the press briefing. “They just try to capture our attention, but they enrich our lives.”

This anthropic’s intention may be, but the above work, contrary, and manages the behavior of the AI ​​model, it is difficult when it is really when other considerations are on the way. This is not a good thing for users; After all, if Chatbots are designed to agree with us, how much can we trust?