Machine learning algorithms are integral and one of the fundamental parts of Google products. And within the framework of the Google I/O 2021 conference taking place these days, they talked about several new very promising developments in this direction.
Perhaps the most prominent AI-related announcement is LaMDA (Language Model for Dialogue Applications). It is an advanced neural network conversational model. Google calls the technology a breakthrough in natural language understanding. It was created and trained for light and constructive conversations on a variety of topics, rather than short, stereotyped answers to specific questions on strictly defined scenarios. The algorithm is able to switch from one topic to another, as humans usually do, but machines cannot yet.
So far, LaMDA only works with text, but the ultimate goal is to create a universal system that can also understand images and audio, states ITC. LaMDA is currently being tested internally by Google, but the company plans to implement this algorithm in Google Assistant and Search in the future. In the latter case, LaMDA should replace or complement other similar BERT technology.
As part of the presentation, Google showed a demo with two examples of the use of technology, notes NIX Solutions. In the first one, the algorithm played the role of Pluto and answered questions about itself, and in the other, it turned into a paper airplane.
Also, Google announced the transition to the fourth generation of tensor processors for data centers, it is twice as fast as the previous one.