Meta AI's BlenderBot 3: 5 things to know about the chatbot that learns from interactions

Meta AI, the New York City-based artificial intelligence laboratory of Meta Inc, has built and released a new state-of-the art chatbot, called BlenderBot 3 (BB3), and is letting members of the public interact with the system to get a feedback on its capabilities.

While the live demo is available here (at the moment only in the US), Meta AI said that it is ‘committed to openly sharing participating de-identified organic conversational data collected from the live demo as soon as we have collected enough data and assessed quality, safety and other issues.’

Here are 5 things to know about Meta AI's BlenderBot 3:

(1.) A publicly available 175B-parameter chatbot, BB3 has been released with model weights, code, datasets, and model cards.

(2.) The BB3 searches the internet to chat about any topic. It has been designed in such a way that it learns to improve its skills through natural conversations and feedback from people ‘in the wild.’

(3.) It is available in three model sizes – 3B, 30B and 175B – and, according to the company, experiments have shown that more the people interact with the model, the more it can learn.

(4.) To make the chatbot learn from interactions and feedback, the team combined two recently-developed machine learning techniques, SeeKer and Director.

(5.) To help BB3 avoid who are trying to trick it into toxic or unhelpful responses, the company has devised new techniques that enable it to learn only from those genuinely helping.

You can read more about BlenderBot3 here or here.