Cloud Next Google is rolling out a bunch of generative AI models and tools across its Workspace apps and Cloud, including an expansion of the promised Duet AI, its personal virtual assistant.
The ad giant earlier this year teased it was adding ChatGPT-like Duet to its Workspace suite of cloud-based office applications – the collection formerly known as G Suite: we were told the chatty bot can be used to generate material for Google Slides, organize rows and columns in Sheets, and generate backgrounds in Meet, for instance.
Now it’s threatening to inject into other areas of Workspace Duet’s generative functionality – you ask it to do or make something, and it tries its best.
Google finds itself in the peculiar situation of having spent years and years researching, building, and deploying machine-learning systems – it basically developed transformers, the neural network architecture powering today’s boom in generative AI – but is perceived, at least, as having fallen behind Microsoft and OpenAI, which have been hyping up their AI products and billing themselves as pioneers.
Google has found out the hard way that no matter how much interesting stuff you come up with in a lab, for your own systems or for research, if you don’t openly and obnoxiously productize and market it early on, you’ll be seen as a chaser rather than a leader by pundits.
And so earlier this year, Google rushed to release its chatbot Bard amid fears Microsoft’s revamped generative-AI-based Bing web search might eat into its own search engine business. OpenAI also grabbed all the attention with its gone-viral ChatGPT bot and a companion API service, and text-to-image software DALL-E. Google has ground to make up, in the public’s view.
On the one hand, this was arguably Microsoft and OpenAI’s win: they pushed products that caught the public and media’s attention, and Google didn’t capitalize enough on its research work beforehand. On the other, Google perhaps didn’t see the value in deploying mindless bots that confidently make up stuff, though now it has to be seen to do something or the shareholders will revolt.
At its Cloud Next ’23 event this week, the Big G announced Duet is coming to enterprises costing $30 per month. Pricing for normal folk remains unannounced.
IT departments interested in trying Duet AI can request a trial. Powered by the tech titan’s PaLM 2 large language model, we’re told the tool can generate text in Google Docs or Gmail given an input prompt or query. It can also be used to drill into Google’s security-related services, such as its Mandiant Threat Intelligence, if you have access.
It can also, as promised earlier, produce images and captions for Google Slides or organize data on Google Sheets. Now we’re told Duet AI can or will be able to automatically take notes in Google Meet, translate speech for 18 different languages, and talk to users as a chatbot via Google Chat.
Other updates announced at the marketing event include beefing up PaLM 2’s capabilities to support 38 languages and extending its context window – the amount of text it can process in its input prompt – to 32,000 tokens, enough words to fill up about 85 pages of text. Google has also improved the quality of its code-generating model Codey and its text-to-image software Imagen, or so the biz claims.
Researchers at Google DeepMind have meanwhile developed SynthID, a tool that watermarks images generated by Imagen. The pixels have been altered in a way that allows software to identify the watermark while being invisible to the human eye.
The beta version of SynthID is being released to a limited number of customers who use Imagen on Google’s Vertex AI cloud platform, to help them better detect fake AI-generated images.
Google also teased as a preview its AlloyDB AI, which is supposed to help developers build their own generative AI apps using data held in AlloyDB for PostgreSQL.
The Silicon Valley biz is also hosting other text-based ML models on Vertex AI, including Anthropic’s Claude 2, the open-source Falcon LLM Technology developed by the Technology Innovation Institute, and Llama 2 from Meta. On top of the AI software announcements, Google also refreshed its hardware chops with its fifth-generation TPU AI accelerator, new virtual machine instances powered by Nvidia’s H100 GPUs, and more. You can watch a replay of the Cloud Next ’23 keynote below. ®
