Facebook Augments BlenderBot 2.0 Chatbot With Internet Access
Facebook has released BlenderBot 2.0, a major upgrade to the BlenderBot it shared with the world a little over a year ago. BlenderBot 2.0 augments the conversational ability of its predecessor with long-term memory and the ability to search online to ensure its responses are not obsolete.
The open-source BlenderBot 2.0 is the first to combine internet data retrieval with long-term memory, according to Facebook. The combination means it won’t sound out-of-touch or like someone who forgets every conversation as soon as it ends. BlenderBot 2.0 continues Facebook’s plans for a chatbot that can combine multiple conversational skills. The original version was built on up to 9.4 billion parameters and trained using 1.5 billion examples of conversation, according to Facebook. The ultimate goal is an AI that converses with the contextual and emotional nuance of a mature human.
Facebook claimed the first version of BlenderBot was the best in the world, with tests specifically comparing it favorably to Google’s then-new Meena chatbot. The upgraded version outperforms a newer much-hyped language model, OpenAI’s GPT-3. The memory and updated information lapses put BlenderBot ahead, according to the developers, who frequently cited Tom Brady’s move to Tampa Bay and Super Bowl win as something BlenderBot 2.0 would know about because it can use internet search engines to update its own long-term memory. Compared to its first iteration, BlenderBot 2.0 hit 17% higher in how engaging people found talking to it and scored 55% better in using previous conversations while being accurate with facts 12% more often.
“While existing systems can ask and answer basic questions about things like food, movies, or bands, they typically struggle with more complex or freeform conversations, like, for example, discussing Tom Brady’s career in detail,” Facebook’s researchers explained in a blog post about their work. “But technology based on BlenderBot 2.0 could one day become a useful part of everyday life by being able to have multisession conversations on any topic that can last days, weeks, or even months, and by adding to what it knows and can talk about as the conversation evolves. That’s because it’s the first chatbot capable of generating internet search queries, using and building knowledge over time and referring back to previous ideas. These advances, including the ability to build long-term memory and augment conversations with information from the internet, overcome some shortcomings of current systems.”
Making BlenderBot 2.0 is part of how Facebook acknowledges that the chatbot still needs work and that public experimentation may improve how well it performs, even if it is the best of what’s around at the moment. The boom in conversational chatbots fueled partly by the COVID-19 pandemic will likely help draw even more interest from developers, especially in healthcare AI and enterprise service. New chatbots and platforms proliferated, such as XAPP AI’s Conversational Self Service for Contact Center Intelligence. Startups saw huge funding rounds as well, with significant rounds announced regularly, including $14 million for higher-ed chatbot platform AdmitHub in February, Rasa’s $26 million round last June, and the $100 million raised by Gupshup earlier this month.
“Research into language model generation is moving quickly, and as an industry, we have better tools than ever before for significantly expanding chatbots’ conversational abilities,” Facebook’s researcher explained.”The model innovations we introduce here make important advances over current state-of-the-art systems, and we’re excited to see how other researchers advance on BlenderBot 2.0’s new abilities to build long-term memory and add to its knowledge by searching the internet.”