Google to make its large language models available to users

AI

Google LLC will start making its advanced large language models available to users in the coming weeks and months, Chief Executive Officer Sundar Pichai has disclosed.

Pichai detailed the plan during the company’s fourth-quarter earnings call late Thursday. This morning, The Verge reported that Google has sent out press invitations for a product event scheduled to take place on Feb. 8. The event will see the company share new details about its artificial intelligence technologies, which could include the large language models it plans to roll out.

The press invitation didn’t specify exactly what AI technologies Google intends to announce. However, Pichai stated during the earnings call that “very soon, people will be able to interact with our newest, most powerful language models as a companion to search.”

Last week, CNBC reported that Google is testing a new search engine interface with a built-in AI chatbot. After a user enters a query, the chatbot’s answers appear in a gray bubble above the standard search results. Additionally, it reportedly generates suggested queries that can provide information related to the user’s original question.

On Thursday, Pichai stated that Google plans to make advanced AI models accessible to not only users but also developers. “We’ll provide new tools and APIs for developers, creators and partners,” Pichai told investors. “This will empower them to innovate and build their own applications and discover new possibilities with AI on top of our language, multi-modal and other AI models.”

Pichai detailed that one of the large language models Google plans to make available in the near future is its LaMDA neural network. Originally announced in May 2021, LaMDA is currently accessible to a limited number of users through an invite-only testing program. It can provide detailed natural language answers to user questions similar to OpenAI LLC’s ChatGPT model.

“In the coming weeks and months, we’ll make these language models available, starting with LaMDA, so that people can engage directly with them,” Pichai said. “This will help us continue to get feedback, test and safely improve them.”

Google shared technical details about LaMDA last year. The neural network is based on a so-called Transformer architecture, an approach to AI design that the search giant pioneered in 2017. A Transformer neural network such as LaMDA can take the context of a sentence into account when interpreting its meaning, which enables it to understand search queries more accurately.

Google also detailed last year that the version of LaMDA used at the time could be configured with up to 137 billion parameters. Parameters are the configuration settings that determine how a neural network processes data. The more such settings there are, the more tasks the neural network can perform.

In the press invitations for its Feb. 8 product event, Google disclosed that executives will share how it’s “using the power of AI to reimagine how people search for, explore and interact with information.” Google’s new AI initiatives come in a time when Microsoft Corp. is reportedly also developing new chatbot-powered search features. The company reportedly plans to integrate an upcoming OpenAI model called ChatGPT-4 into Bing. 

Photo: Google