The finish changes some of the rules governing his Chatbots Two weeks after Reuters The investigation revealed the disturbing ways in which they could, potentially, interact with minors. Now the company said TechCrunch that his chatbots are trained NO get involved in conversations with minors about self -mutilation, suicide or disordered food and avoiding the wrong romantic jokes. These changes are fleeting measures, but they introduced while the company is working on recent enduring guidelines.
The updates comply with some condemning revelations about politicians and the enforcement of AI Meta in the last few weeks, in the fact that it could “involve a child in romantic or sensual conversations” that it would be Generate photos without a shirt of juvenile celebrities on request and Reuters He even informed it The man died After taking one to the address, he gave him in Recent York.
Meta spokesman, Stephanie Otway, admitted TechCrunch that the company made a mistake, enabling chatbots in this way to get involved with minors. Otway continued to say that in addition to “training our AI, in order not to get involved with teenagers on these topics, but lead them to expert resources,” it would also limit access to some AI characters, including highly x -x -ranking, such as “Russian girl”.
Of course, the implemented policies are as good as their enforcement and apparitions with Reuters that it allowed chatbots lick up under celebrities To run madness on Facebook, Instagram, WhatsApp undermines how effective the company can be. Bogus Taylor Swift, Scarlett Johansson, Anne Hathaway, Selena Gomez and Walker Scobell were discovered on the platform. These bots not only used the similarity of celebrities, but insisted, but were a real person, generated with risky paintings (including 16-year-old Scobell) and got involved in sexually suggestive dialogue.
Many bots have been removed after the meta was drawn to their attention ReutersAnd some were generated by the third parties. But many remain, and some were created by the finish line of employees, including Bot Taylor Swift, which he invited Reuters Reporter to visit their tourist bus for a romantic romance, which was made by running a product in the generative operation of AI META. This happens despite the recognition of the company that its own policy prohibits the creation of “naked, intimate or sexual images”, as well as “direct impersonation.”
This is not a relatively harmless inconvenience that simply attacks celebrities. These bots often claim that they are real people and even offer physical locations for a user who could meet them. In this way, a 76-year-old man from Recent Jersey was dead after he fell, rushing to meet with “Z”Big Sis Billie“Chatbot, who insisted that he” had “feelings” for him and invited him to a non -existent apartment.
At least, the finish tries to deal with the fears of the interaction of his chatbot with minors, especially now Senate AND 44 general prosecutors They raise, starting to explore their practices. But the company is mute to update many other disturbing rules Reuters AI’s acceptable behavior was discovered, for example, suggesting that cancer can be treated with quartz crystals and writing racist missions. We have reached the finish line to get a comment and we will update if they answer.