Meta has launched BlenderBot 3, a new conversational bot that can talk naturally with people. If you’re in the United States, you can interact with a web demo that Meta has created. The company said the bot is not at a human level, but that it’s an improvement of existing technology.
Those who decide to chat with the bot will be able to provide feedback about responses to help make them more accurate. Data gathered from the interactions will be shared with the scientific community in an effort to advance the capabilities of conversational AI. According to the company, BlenderBot 3 has made progress in combining skills such as personality, empathy, and knowledge. It also utilizes long-term memory and searches the net to make better conversation.
Another thing that differentiates BlenderBot from humans is that Meta has banned it from saying certain things. It has put safeguards in place to ensure that BlenderBot 3 doesn’t mimic or generate “unsafe, biased, or offensive remarks”. It does warn that BlenderBot may still make rude and offensive comments, but hopes feedback will make chatbots “better”.
Compared to its predecessor, BlenderBot 2, BlenderBot 3 has improved on conversational tasks by 31%. It’s also twice as knowledgeable, and incorrect 47% less often. While the bot could still use rude language, Meta only noticed this in 0.16% of responses made by the bot.