Meta builds AI model that can virtually chat with people & learn from conversations

Meta AI

Meta AI has built and released BlenderBot 3, a publicly available chatbot complete with model weights, code, datasets, and model cards, and has deployed it in a live interactive conversational AI demo here.

BlenderBot 3 by Meta AI is capable of searching the internet to chat about virtually any topic, and it’s designed to learn how to improve its skills and safety through natural conversations and feedback from people “in the wild”.

Meta combined two recently developed machine learning techniques, SeeKeR and Director, to build conversational models that learn from interactions and feedback. According to Meta, initial experiments show that as more people interact with the model, the company will be sharing organic conversational data collected from the interactive demo system and model snapshots in the future.

Meta AI’s progress in building conversational AI systems with BlenderBot and its successor, BlenderBot 2 initiated the development of the first unified system trained to blend different conversational skills, such as personality, empathy, and knowledge, to have long-term memory, and to search the internet to carry out conversations.

As a step in this direction, Meta built and deployed a live demo of BlenderBot 3, the conversational agent that can converse naturally with humans, who can then provide feedback to the model on how to improve its responses. The demo is only available in the US. The company will be sharing data from these interactions and has shared the BlenderBot 3 model and model cards with the scientific community to help advance research in conversational AI.

Also Read: Meta introduces new AI tool that transforms text into visual art

BlenderBot 3 is built from Meta AI’s publicly available OPT-175B language model, approximately 58 times the size of BlenderBot 2. The model has a modular design, which is a subsequent version of the recently introduced SeeKeR architecture. The company’s initial experiments show that they can make the models better by enabling them to learn from their experience.

The AI model is built with all the skills of its predecessors, which include internet search, long-term memory, personality, and empathy. To improve upon its engagingness, Meta AI collected a new public dataset consisting of over 20,000 human-bot conversations predicated on over 1,000 skills and trained BlenderBot 3 to learn from conversations to improve the diverse body of skills that people find most important, from talking about healthy food recipes to finding child-friendly amenities in the city.

When the conversational response of the bot is unsatisfactory, feedback is collected from the conversationalist. Using this data Meta can improve the model, so it does not repeat its mistakes, and can then redeploy it for continued conversation, iterating the approach to search for more mistakes, and eventually improving it further.

The approach uses a new learning algorithm called Director, which generates responses using two mechanisms: language modeling and classification. Language modeling provides the model with the most relevant and fluent responses (based on training data) and then the classifier informs it of what is right and wrong (based on human feedback). To generate a sentence, the language modeling and classifier mechanisms must agree.

The deployment of BlenderBot 3 and the accompanying program of continuous data collection can provide a path that eventually leads to production applications, such as virtual assistants.


Comments