LaMDA, short for Language Models for Dialog Applications, first introduced at Google I/O 2021, has received an update with Google I/O 2022. The latest version, titled LaMDA 2 is to said to be Google’s most talkative AI, which has knowledge about thousands of complex subjects. Google shared that the LaMDA 2 AI can break complex issues into accessible and digestible topics.
LaMDA 2 can understand thousands of topics and create conversations on them by asking follow-up questions and sharing knowledgeable insights. Moreover, Google claims that AI cannot repeat a conversation in the same manner that it happened before. LaMDA 2 is designed to take a different route with every conversation and reveal new information about the subject in every recent conversation.
LaMDA 2 learns how words are used in a sentence from the millions of texts scraped through social media, Reddit, public web forums, Wikipedia and many other websites. At the launch, Sundar Pichai demonstrated a conversation with LaMDA 2 on Mariana Trench. The AI asked follow-up questions such as what creatures reside in the Trench and discussed things like submarines and bioluminescence, which Google revealed that it realised on its own through going through the web.
Another demo about the vegetable garden was showcased with the AI. LaMDA 2 offered suggestions on where and how to start a vegetable garden. It went on to discuss in detail the ideal location for a garden and what kind of vegetables one can grow in a garden. Furthermore, the AI categorised the list of tasks and subtasks for making a vegetable garden.
However, Sundar Pichai acknowledged that the Language Model and Dialog Application is far from perfect as it may offer unreliable and untrue data in some cases. While, Google was creating the AI, it repeatedly gave wrong information about Mount Everest. But Pichai believes that this is a step forward in the future, and this technology will develop further with time.