Google is working on new artificial intelligence features for its flagship search product, according to CEO Sundar Pichai. He told investors that it would roll out aspects of its massive LaMDA large language model into search results “very soon”. During the earnings call he also confirmed the AI would be coming to Google Cloud, Gmail and Docs.
Pressure is growing from Microsoft which has taken a $10bn investment in ChatGPT-maker OpenAI with plans to embed the natural language chatbot into its Bing search engine, Azure cloud services and other consumer and business products.
Google has two large language models, LaMDA with 137bn parameters and PaLM with 540bn. In comparison, GPT-3.5, the model from OpenAI used to train ChatGPT has 175bn parameters. Both Google LLMs are designed around natural language processing but LaMDA, short for Language Model for Dialogue Applications, has already been shown to be capable of engaging in a free-flowing “seemingly endless” conversation on any number of topics.
Pichai says the rollout of AI into Google’s products would begin with LaMDA but both models would be made available so that “people can engage directly with them”. This, he says “will help us continue to get feedback, test and safely improve them” suggesting a ChatGPT-style interface could be on the horizon.
“AI is the most profound technology we are working on today,” said Pichai during the company’s earnings call on Thursday. “Our talented researchers, infrastructure, and technology make us extremely well-positioned as AI reaches an inflection point.”
He has been pushing Google as an AI-first company since 2017, when the company published a ‘field-defining’ transformer research project and work in diffusion models that form the basis of many generative AI applications including image generation tools. “Translating these kinds of technical leaps into products that help billions of people is what our company has always strived on,” said Pichai.
Is Google ready to take on ChatGPT?
Reports suggest Google was taken off guard by the success of ChatGPT, which became the fastest-growing consumer application in history, reaching 100 million active monthly users in January. It was only launched at the end of November as a research project to test the potential for dialogue-based AI.
It has been testing a number of features on employees as part of a plan to respond to the success of the chatbot, and cement its position as an AI-driven company before Microsoft embeds AI into Bing. This includes a chatbot called “Apprentice Bard” built on LaMDA and possible redesigns of its search results page to show AI responses.
“Very soon, people will be able to interact directly with our newest, most powerful language model as a companion to search, in experimental and innovative ways,” said Pichai. It isn’t clear how soon but he suggested it would be in the coming weeks and months as a beta product, getting feedback and retraining based on responses.
AI was a central focus of the Alphabet fourth-quarter earnings call, calling it the most “profound technology” the company is working on today. It comes amid growing pressure from Microsoft and others in the search, advertising and cloud space.
Pichai said as well as consumer products, Google would be making APIs and developer tools available to allow them to build with its AI products and language models and “empower them to discover new possibilities with AI”.
Gmail and Google Docs to get AI assistance
This will likely see aspects of LaMDA and PaLM made available through Google Cloud in the same way Microsoft plans to offer multiple OpenAI APIs through its Azure cloud service. Pichai said the tools it would be offering are “particularly amazing for composing, constructing and summarising” but that they would scale slowly as he views them as being “early day” projects.
“Our AI is a powerful enabler for businesses and organisations of all sizes, and we have much more to come here,” said Pichai. “There are a few flavours of this. Google Cloud is making our technological leadership in AI available to customers via our Cloud AI platform, including infrastructure and tools for developers and data scientists like Vertex AI. We also offer specific AI solutions for sectors like manufacturing, life sciences, and retail and will continue to roll out more.”
He said Google would also bring the benefits of large language models into Gmail and Docs, as well as other parts of its Workspace suite of collaboration tools, and provide advertisers with AI research to improve campaigns.
Google also owns DeepMind, a British AI company it acquired for £400m in 2014. As part of the AI-first push, DeepMind will be further incorporated into Google. It has been working closely with other Google-based AI teams and will be reported as part of Google, rather than the company’s “Other Bets” division, from the first quarter of 2023.
This change in reporting “reflects the strategic focus in DeepMind to support each one of our segments” said Ruth Porat, Alphabet’s CFO during the earnings call. “We consolidate Other Bets into Google only when that bet supports products and services within Google or Alphabet broadly” she explained. This has previously included cybersecurity company Chronicle being rolled into Google Cloud.
DeepMind has also been working on natural language models, including the 280bn-parameter Gopher, with a focus on dialogue and natural language. It is also exploring ways to ensure it aligns with human values. As part of this, it has developed its own chatbot named Sparrow, trained to reduce the risk of unsafe and inappropriate answers, and have the ability to browse the internet to confirm its responses are accurate.