Google Accidently Confirms Advanced AI Assistant 'Jarvis' That Autonomously Retrieves Information

Internal preview of the product was mistakenly posted on the Chrome Web Store.

Google has inadvertently confirmed the development of an advanced AI system named "Jarvis," following a leak that revealed the tech giant's latest innovation in AI.

Jarvis is designed to streamline web browsing and automate tasks.

According to The Information, the accidental leak occurred when an internal preview of the product was mistakenly posted on the Chrome Web Store. Although the information was quickly removed, it circulated widely, offering a glimpse into Google's forthcoming AI project.

Some users managed to download it, although without functional access due to specific permission requirements.

The leaked information reveals that Jarvis is not just any AI but an agent with the capacity to browse the internet autonomously, retrieve live information, and carry out actions typically restricted to humans, such as making online purchases (with user permission).

The listing described Jarvis as "a helpful companion that surfs the web with you."

Unlike current AI tools, which are confined to preloaded databases and cannot access new information, Jarvis is reportedly designed to perform real-time searches across the web, bypassing traditional search engines. This could empower users to retrieve up-to-date information, tailored to their needs and queries, directly from the source.

The leak follows a string of high-profile AI announcements from other leading companies. OpenAI, for example, experienced its own leak just days earlier, revealing an "O1" model capable of analyzing images, conducting searches, and using data-analysis tools.

OpenAI has also recently launched ChatGPT Search for Plus subscribers.

Last month, Anthropic, another major player in AI, unveiled its "computer use" feature, enabling its Claude AI to interact directly with users' computer screens. By interpreting on-screen content, the AI can perform tasks such as web browsing, clicking buttons, and inputting text.

During a demonstration, the AI showcased its capabilities by planning a hike near the Golden Gate Bridge, checking sunrise times, and creating detailed calendar invites.

Early adopters, including Canva, Asana, and Replit, have already begun integrating the technology into their platforms for graphic design, project management, and coding.

Alongside the computer use feature, Anthropic has introduced Claude 3.5 Sonnet, an upgraded language model with enhanced coding and reasoning capabilities. This model, along with the improved Claude 3.5 Haiku, offers significant improvements in various tasks.

Anthropic says Claude 3.5 Sonnet excels at complex coding tasks, outperforming previous models in terms of accuracy and efficiency. The model demonstrates stronger reasoning abilities, leading to better decision-making and problem-solving. It can also automate web-based workflows, saving time.

This article originally appeared on our sister site Computing.