When it comes to speech recognition, Google is widely regarded to have some of the most advanced technologies that put it a step ahead of other companies. Google is now aiming at extending that lead with their latest research breakthrough in conversational abilities, called LaMDA.
LaMDA stands for “Language Model for Dialogue Applications” and it is based on the same neural network architecture that is used by Google’s earlier language models such as BERT and GPT-3. What makes LaMDA unique is that it is trained on dialogue, which allows the model to pick up on different nuances that are characteristic in open-ended conversations.
Some of the nuances that LaMDA focuses on include “sensibleness” and “specificity”. Google is also researching into making conversations that have more “interestingness”, while ensuring that LaMDA remains factually correct.
Google did not mention where or when they will be integrating LaMDA’s capabilities. The model is still in its early stages of development, and we can expect Google to use it soon with the Google Assistant, Google Search and other products.