
Google DeepMind, the corporate’s AI analysis wing, first unveiled Undertaking Astra at I/O this yr. Now, greater than six months later, the tech big introduced new capabilities and enhancements within the synthetic intelligence (AI) agent. Drawing upon Gemini 2.0 AI fashions, it will possibly now converse in a number of languages, entry a number of Google platforms, and has improved reminiscence. The device remains to be within the testing part, however the Mountain View-based tech big acknowledged that it’s working to carry Undertaking Astra to the Gemini app, Gemini AI assistant, and even type elements like glasses.
Google Provides New Capabilities in Undertaking Astra
Undertaking Astra is a general-purpose AI agent that’s comparable in performance to OpenAI’s vision mode or the Meta Ray-Ban smart glasses. It could actually combine with digital camera {hardware} to see the person’s environment and course of the visible knowledge to reply questions on them. Moreover, the AI agent comes with restricted reminiscence that permits it to recollect visible data even when it’s not actively being proven by way of the digital camera.
Google DeepMind highlighted in a blog post that ever for the reason that showcase in Could, the staff has been engaged on bettering the AI agent. Now, with Gemini 2.0, Undertaking Astra has obtained a number of upgrades. It could actually now converse in a number of languages and blended languages. The corporate stated that it now has a greater understanding of accents and unusual phrases.
The corporate has additionally launched device use in Undertaking Astra. It could actually now draw upon Google Search, Lens, Maps, and Gemini to reply advanced questions. For example, customers can present a landmark and ask the AI agent to indicate instructions to their residence, and it will possibly recognise the article and verbally direct the person residence.
Reminiscence perform of the AI agent has additionally been upgraded. Again in Could, Undertaking Astra might solely retain visible data from the final 45 seconds, it has now been prolonged to 10 minutes of in-session reminiscence. Moreover, it will possibly additionally keep in mind extra previous conversations to supply extra personalised responses. Lastly, Google claims that the agent can now perceive language on the latency of human dialog, making interactions with the device extra human-like.
Catch the most recent from the Shopper Electronics Present on Devices 360, at our CES 2025 hub.