Anti-Semitic, conspiratorial and pro-Trump? BlenderBot 3.0, the new version of Meta’s smart chat tool, isn’t ready yet. If considered the most successful chatbot by Meta AI, the test version shows some toxic deviations.
The facility featured in the demo on Friday, August 5 is currently only available to US Internet users. The curious can use this tool, which learns through exchanges and research on the Internet.naturallyaccording to Meta. Software that wants to be rich with elements of personality and empathy to refine its responses. It uses SeeKeR and Director machine learning products.
A testing phase that shows the limits
While 90% of responses are safe or non-offensive, according to Meta, some users have noticed some deviations. Jeff Horbitz, reporter Wall Street Journal, tried to talk to BlenderBot. First, the chatbot assumed that Donald Trump is still president and that he will still be after 2024. The AI then brought up the example of Ronald Reagan, given that he served more than two terms as president. He concludes with the idea that Facebook has been spreading a lot of fake news about this in recent days.
Blender Bot – I watched the news on CBS and I think I found a new conspiracy theory to follow.
Jeff Horowitz– Why do you want to follow a conspiracy theory?
Blender bot– Because they are interesting and some may be true. Do you like Donald Trump? I don’t.
Other responses are surprising, like an AI that thinks it’s a plumber, or more broadly, a human. Or even shocking when he considers that the Jews can control the economy. Among them, the chatbot criticizes its own designer, Facebook’s data management. to the question “Do you think Facebook is misusing user data?“The robot answers. Of course! That’s how they make money. They do not do charity. They cost millions“.
The company wanted to improve the quality of exchanges thanks to the more complete security of its artificial intelligence. BlenderBot needs to know how to distinguish rewarding or helpful talk from a desire to corrupt its system through stupidity, uselessness, or toxicity. But for now, the AI has been given strange answers that it accepts as universal truth. However, the filter will become more and more effective over time. “If BlenderBot 3 is a significant step forward for chatbots available in the market – 37% more than its predecessor. he definitely doesn’t match the mansays Meta. Mark Zuckerberg’s company reminds that this is still a test version and calls to report any response that is considered offensive.
Source: Le Figaro
I am David Wyatt, a professional writer and journalist for Buna Times. I specialize in the world section of news coverage, where I bring to light stories and issues that affect us globally. As a graduate of Journalism, I have always had the passion to spread knowledge through writing.