Google IO Sundar Pichai, CEO of Alphabet and its Google subsidiary, opened the 17th annual Google I/O developer conference on Tuesday, evangelizing the transformational power of artificial intelligence, as he did last year and the year before that.
But amid all the hype, the Chocolate Factory leader did promise some new AI-related features and products for business users later this year.
Pichai began by remarking that the season of Gemini begins on May 20, adding, “I’m not really sure what the big deal is; every day is Gemini season here at Google.”
Gemini season is an astrological reference, while Google uses the term to refer to its family of AI models. But Pichai’s word play is fitting because AI, like astrology, has been sold based on belief.
That’s starting to change as practical applications become apparent. Pichai’s keynote emphasized results and benchmarks. “We are shipping faster than ever,” he said, perhaps mindful that on Monday, a few hundred miles north in Seattle, Washington, his Microsoft counterpart Satya Nadella gave a similar spiel at the Build conference.
“We have announced over a dozen models and research breakthroughs and released over 20 major AI products and features, all since the last I/O,” said Pichai.
Pichai called out the progress of Gemini 2.5 Pro, its most advanced LLM, comparing its Elo benchmark score of 1448 to Gemini 2.0 Pro (1379), Gemini 1.5 Pro (1249), and Gemini 1.0 Pro (1111). “Today, Gemini 2.5 Pro sweeps the LLMArena leaderboard in all categories,” he said.
Pichai credits Gemini’s success with Google’s full-stack approach to infrastructure and remarked that the mega-corp’s 7th generation Tensor Processing Unit (TPU), Ironwood, delivers 10x as much performance as the previous generation and is capable of 42.5 exaFLOPS of compute per pod. The TPU will be available to Google Cloud customers later this year.
That infrastructure, he added, enables Google “to deliver the best models at the most effective price point.”
But there’s more to AI than benchmarks, and Pichai provided updates on three projects, Starline, Astra, and Mariner, that may actually find adoption within companies.
Project Starline, a 3D video conferencing system demoed in 2021, has been updated with AI. The project has spawned Google Beam, which “uses a new state of the art video model to transform 2D video streams into a realistic 3D experience,” as Pichai put it.
Using an array of six cameras, video conferencing subjects get captured at different angles so they can be rendered in a 3D light-field display, at 60fps in real-time.
Google has partnered with HP to deliver Google Beam to customers later this year, supposedly.
Technology from Project Starline has also made its way to Google Meet, the web giant’s video meeting service, in the form of real-time bi-directional language translation, voiced using reasonably humanoid AI-generated speech.
English and Spanish translation, said Pichai, is now available for small-business Meet subscribers, with more languages planned in the coming weeks. Enterprises will gain access later this year.
Finally, the Gemini API is inheriting several capabilities developed through Project Mariner, a research project exploring computer use by software agents. These include multitasking and teach-and-repeat, the ability to teach the model to perform a specific task so that it can plan similar tasks on its own.
Pichai also said Google’s GenAI SDK is now compatible with Anthropic’s Model Context Protocol, aka MCP, a scheme for standardizing how applications interact with LLMs.
Just as Google many years ago made user account data available across its services, the company aims to make AI services more personalized. “We are working to bring this to live with something we call Personal Context,” said Pichai. “With your permission, Gemini models can use relevant context across your Google Apps, in a way that is private, transparent and fully under your control.”
In Gmail, this will take the form of Personalized Smart Replies, automated responses massaged by AI to sound as if you had written them.
“Let’s say my friend wrote to me looking for advice,” Pichai mused. “He’s taking a road trip to Utah and he remembers I did this trip before. Now if I’m being honest, I’d probably reply with something short and unhelpful. Sorry, Felix. But with Personalized Smart Replies, I can be a better friend. That’s because Gemini can do almost all the work for me, looking up my notes in Drive, scanning past emails for reservations, finding my itinerary in Google Docs – trip to Zion National Park.
“Gemini matches my typical greetings from past emails, captures my tone, style, and favorite word choices, and then it automatically generates a reply.”
Personal Context will be available this summer in Gmail for subscribers, and will be available to send to friends, who may decide not to be friends if they learn you’ve delegated communication to Gemini.
There was plenty of AI gubbins for consumers as well.
Such is Google’s infatuation with AI that it is offering a new AI Mode to US users of its core product, Google Search, starting today. It’s available as a setting in Google Labs, with no word on how this might affect website traffic referrals.
“For those who want an end-to-end AI search experience,” said Pichai. “We are introducing an all new AI mode. It’s a total reimagining of search.”
Gemini is pretty good at telling you when you’re wrong
AI mode promises the ability to ask longer, more complex queries, one one might ask an AI model. Google is surfacing AI Mode as a new tab in its Search window, to be augmented by Personal Context once that arrives. And starting this week, Google Search will connect to Gemini 2.5.
Project Astra, a prototype AI assistant created by Google’s DeepMind AI research arm, has provided its camera and screen-sharing capabilities to Gemini, its personal assistant. A video taken by a tester of the service shows the AI assistant using vision capabilities to correct the tester’s inaccurate descriptions of objects encountered by wandering around a neighborhood.
“Gemini is pretty good at telling you when you’re wrong,” remarked Pichai at the conclusion of the video, who added that this capability is being rolled out to everyone using Gemini on Android and iOS today.
Google also announced new generative media models. Veo 3 and Imagen 4 provide video and image generation respectively. Lyria 2 provides music generation capabilities – now it’s not just visual artists who have to cope with AI-generated competition. And there’s a new AI filmmaking tool called Flow. The biz also launched SynthID Detector, a website to help identify AI-generated content. It’s a bit like a firearms maker running a trauma center.
“The opportunity with AI is truly as big as it gets,” Pichai said as he wrapped up. “And it will be up to this wave of developers, technology builders to make sure its benefits reach as many people as possible.” ®
