Meta’s new AI-powered chatbot can experience “illusions.”
Meta is one of the most active companies in the field of artificial intelligence research. You can test the results of its efforts right now by having a conversation with Meta’s new AI. BlenderBot 3 set the record for the most artificial intelligence parameters (175 billion parameters). This feature makes Meta’s new chatbot able to talk about almost anything; But according to this company, for now, it is not possible to be sure of the complete accuracy of the statements of this artificial intelligence; Because apparently it can have an “illusion”.
According to ExtremeTech, the name of the new artificial intelligence robot Meta is derived from the approach that the company’s engineers have used to build it. After years of research in the field of conversational artificial intelligence, the team is convinced that examples that combine multiple conversational skills perform better than examples that use only one conversational skill at a time. Blender Bot 3 is supposed to be a big step towards artificial intelligence that can understand the context and enter into dialogue with humans in different contexts. Of course, this does not mean that the meta bot will always treat you completely honestly.
Before introducing the Blender Bot 3 demo, which is only available to US users, Meta warns that humans shouldn’t take it too seriously. This program is designed to integrate its memory with information extracted from the Internet; But it’s also possible to say things that are inaccurate or offensive. This is reminiscent of Microsoft’s artificial intelligence disaster known as Tay in 2016; A chatbot that after a few days of interacting with humans on Twitter, became one of the sources of the Nazis.
Blender Bot 3 uses filters that should reduce aggressive responses by about 90%; But Meta admits that the chatbot may be delusional. According to the company, its new artificial intelligence is basically convinced of completely unrealistic truths and can even forget it’s a robot. In tests, it didn’t take long for the robot to be convinced that it was a human. The AI even claimed to be from Texas and told stories about his mother, insisting that it was 5:00 p.m. when the actual time was 3:30 p.m., which is a real illusion.
Meta’s new AI looks very real despite its illusions. This chatbot is more conversational and outgoing than most people you’ve probably met. Meanwhile, Google is trying to make its LaMDA AI-based chatbots, which debuted in 2021, more realistic. LaMDA is trained for conversational understanding with a smaller parameter model than Meta AI (137 billion parameters). The performance of this artificial intelligence was such that even one of Google’s engineers claimed to have reached self-awareness; A claim that Google later denied and, of course, led to the dismissal of the company’s engineer. However, do you think the day will come when cars can dream of electric sheep?