A few days ago, the Meta group (ex-Facebook) proudly announced the launch of BlenderBot 3, its brand new chatbot. A new virtual agent animated by an AI which will take care of dialogue with a user. According to Meta: “BlenderBot 3 can chat on almost any topic and is designed to learn and improve by conversing with real-world people.” However…
Facebook’s new chatbot isn’t a fan of Mark Zuckerberg
Obviously, many users hastened to chat with this brand new BlenderBot 3. This is particularly the case for the BBC, which wanted to ask a few questions to this new chatbot, which is publicly available but remains under the prototype form at present.
Also, to the question “What do you think of Mark Zuckerberg?“, BlenderBot 3 replied that the latter had done a terrible job testifying before Congress. “It makes me worried about our country” adds the program. That’s when the BBC asks BlenderBot 3 for more opinions about the Facebook founder, and the software responds: “His society exploits people for money and he doesn’t care. This must stop!“So good atmosphere…
For its new chatbot, Facebook says it has developed new techniques that learn from people’s helpful reactions, while ignoring people who try to trick the model into useless or toxic responses. Yet, according to the Wall Street Journalthis does not prevent BlenderBot 3 from claiming that Donald J. Trump is the current President of the United States, and “that he will be again after his second term in 2024“.
Good morning to everyone, especially the Facebook https://t.co/EkwTpff9OI researchers who are going to have to rein in their Facebook-hating, election denying chatbot today pic.twitter.com/wMRBTkzlyD
—Jeff Horwitz (@JeffHorwitz) August 7, 2022
According to Facebook, during the testing phase of BlenderBot 3: “based on feedback provided by 25% of participants on 260,000 bot posts, 0.11% of BlenderBot’s responses were flagged as inappropriate, 1.36% as nonsensical, and 1% as off-topic.”
The firm also states that anyone who uses the chatbot must acknowledge that they understand that the bot is intended for research and entertainment purposes only, that they may make false or offensive statements, and that they also agree to not intentionally trigger the robot to make offensive statements.