Google at the I/O 2022 consumer keynote on Wednesday announced the launch of LaMDA 2 as its new conversational artificial intelligence (AI) model that can help make conversations via computing devices easier and more efficient. Touted to be the most advanced conversational AI, LaMDA 2 is the successor to the original LaMDA — shortform of Language Models for Dialog Applications — that was launched at the I/O developer conference last year. Alongside LaMDA 2, Google announced AI Test Kitchen as its dedicated app to help test AI models including LaMDA with people outside of the company.
Similar to the original model that debuted last year, LaMDA 2 is aimed to help develop better conversational applications.
Google CEO Sundar Pichai at the keynote said that LaMDA 2 carried “incredible conversational capabilities” that can help make conversations with computers more natural. The new AI model uses a huge bulk of datasets from various sources to communicate with users on different topics.
Pichai showcased the scope of LaMDA 2 with an onstage demo in which a user asked the AI model to describe a scene in the Mariana Trench, which is located in the western Pacific Ocean. The model responded by giving details about creatures that living in the trench. It even generated follow-up questions and touched upon the topics, such as submarines and bioluminescence, for which it had specifically not been trained to answer.
“You can take the conversation anywhere you want. Maybe you’re curious about how smell works and you want to dive deeper. You’ll get a unique response for that too,” Pichai said.
He acknowledged that although LaMDA 2 has been designed with improved safety, the model might still “generate inaccurate, inappropriate, or offensive responses.” Google has for all these reasons built the AI Test Kitchen app that would be accessible to experts such as AI researchers, social scientists, and human rights experts to help bring improvements to the ongoing AI developments including LaMDA.
The AI Test Kitchen app comprises three modes, namely ‘Imagine It’, ‘Talk About It’, and ‘List It’. The ‘Imagine It’ mode allows LaMDA to describe a place — real or imaginary. The ‘Talk About It’ mode, on the other hand, allows the AI model to talk about a topic by detailing its various aspects, while the ‘List It’ mode allows users to give any task or topic to the model to get its breakdown into relevant and useful bullet points. The app can also generate tips for the points it provides that will be related to the given topic, the demonstration showed.
Google has also provided an option to send feedback through the AI Test Kitchen app to let its engineers improve the AI model.
“One day, we hope we can answer questions on more topics in any language you speak, making knowledge even more accessible, in Search and across all of Google,” said Pichai.
The AI Test Kitchen will initially be limited to the US and will not be available to all users. The purpose of it is essentially to improve the models that Google has built for testing. The app is also set to continue adding other “emerging areas of AI” to bring improvements to various existing and new models.
All this may eventually help the Mountain View, California-based company enhance products including Google Search, Google Assistant, and others that are using software algorithms to understand user preferences and delivery a conversational experience with machines that we have in our rooms (read computers, laptops, smart screens, and smart speakers) or even in our hands (read smartphones).
Google is also developing a technique called chain-of-thought prompting to help describe multi-step problems as a series of intermediate steps to help AI models answer even tough questions such as how many hours are there in a month.
Pichai said that when Google’s new technique is blended with its recently launched AI model Pathways Language Model (PaLM), it could bring “promising results” to users.
“In chain-of-thought prompting, we give the model a question-answer pair, but this time, an explanation of how the answer was derived,” he said.
Alongside the new conversational AI-focussed and language model developments, Google at this year’s I/O consumer keynote announced the launch of its machine learning hub for Google Cloud customers. It is the world’s largest, publicly-available machine learning hub, Pichai claimed.
The machine learning hub carries eight Cloud TPU v4 pods that are custom-built on the same networking infrastructure that is backing Google’s largest neural models.
“They provide nearly nine exaflops of computing power in aggregate — bringing our customers an unprecedented ability to run complex models and workloads,” Pichai explained.
The purpose of the machine learning hub is to enhance experiences across fields such as medicine, logistics, and sustainability, among others.
Google is indeed not alone in the world of AI and machine learning updates as its competitors including Amazon, Meta, and Microsoft are also making strong efforts to grow their businesses in these emerging fields. However, the new approaches taken by Google are aimed to be different from the competition. The company also has its range of products available to gain massive datasets and convince a large number of people that would eventually help it test the ongoing developments in an efficient way.