Openai CEO Sam Altman recently admitted that he underestimated how the company is attached to certain AI models and called for a decision to suddenly remove the older version as “mistakes.” In a post on Microblogging Platform X (formerly known as Twitter), Altman wrote: He continued on.Altman said this attachment has been happening on Openai's radar for over a year, but there was little mainstream attention. He also warned of the risk that people in “mentally fragile states” blur the line between reality and roleplay.He further warned that people are using technology including AI in self-destructive ways. He said, “If a user is mentally vulnerable and prone to paranoia, they don't want AI to enhance it. Most users can maintain a clear line between reality and fiction or roleplay, but not a small percentage.” “We value user freedom as a core principle, but we also feel responsible for how we implement new technologies with new risks,” he said. In extreme cases, the response is clear, such as promoting the delusions of vulnerable users, but he said, “The concerns that bother me the most are more subtle.”
Sam Altman: People use chatgpt like therapists
In the post, Sam Altman highlighted that many people are already using ChatGpt like “therapists and life coaches.” “If people get good advice, level up towards their goals and their life satisfaction has increased over the years, they'll be proud to create something that really helps, even if they rely on lots of chatgpt,” he wrote.However, he warned against situations where users feel better in the short term but are not unaware of long-term happiness, or where they want to use less ChatGpt but feel unstoppable.Altman said he could imagine a future where “many people really trust ChatGpt's advice on their most important decisions.” He predicted that billions of people could talk to AI straight away like this, and that it would be important to ensure that the outcome was “big pure positive.”He added that Openai has better tools than previous high-tech generations for measuring impact, including its ability to gather user feedback and teach models about nuanced issues.
