The GPT-3 and BlenderBot 1.0 models are extremely forgetful, but that’s not the worst of it! They’re also known to “hallucinate” knowledge when asked a question they can’t answer.
It is no longer a matter of whether or not machines will learn, but how. And while many companies are currently investing in so-called “deep learning” models that focus on training ever larger and more complex neural networks (and their model weights) to achieve greater levels of sophistication by making them store what they have learned during the course/training process, it has proven difficult for these large models to keep up with changes occurring online every minute as new information continually floods into its repository from all over the internet.
Facebook AI is excited to announce that they are releasing a new open source chatbot, BlenderBot 2.0 by way of their research platform ParlAI. BlenderBot 2.0 accesses data directly from any number of sources at once rather than relying solely on storing everything it learns within its own weight matrix – this allows for quicker response times when presented with dynamic. With its ability to access memory and reduce hallucination, BlenderBot 2.0 builds on the original version of BlenderBot — the first chatbot to blend a diverse set of conversational skills including empathy, knowledge, and personality together in one system – with exciting prospects for what’s next! BlenderBot 2.0 is a significant update to its predecessor, which was open-sourced in 2020.
BlenderBot 2.0 is better at conducting more extended, more knowledgeable, and factually consistent conversations over multiple sessions than the existing state-of-the-art chatbot. BlenderBot’s improved conversational abilities have made it a serious contender for artificial intelligence research.
The AI model takes the information it gets from conversations and stores them in long-term memory. The knowledge is stored separately for each person they speak to, which ensures that new information learned in one conversation can’t be used against another.
This model can read and respond in real-time, making it an excellent tool for keeping up with current events. It can scan the internet for new information to have a more up-to-date conversation.
Facebook AI Research is releasing the complete model, code, and evaluation set up to help advance conversational AI research. The Facebook team combined human conversations with internet searches that have been bolstered for training purposes. They also advanced their work by creating a multisession chat system where previous sessions are referenced so other researchers can reproduce this work as well!
Paper 1: https://github.com/facebookresearch/ParlAI/blob/master/projects/sea/Internet_Augmented_Dialogue.pdf
Paper 2: https://github.com/facebookresearch/ParlAI/blob/master/projects/msc/msc.pdf