Author HubMatthias Bastian
Matthias is the co-founder and publisher of THE DECODER, exploring how AI is fundamentally changing the relationship between humans and computers.
Ad
Ad
Ad
Ad
Ad
Ad
Ad
Artificial Intelligence: News, Business, Research
Google has published a free white paper that outlines how AI agents work and what they can do. The company explains that AI agents are systems that can watch their environment, make choices, and act on their own to reach specific goals - without constant human input. The paper describes three main parts that make up AI agents. First, there's the AI model at the core. Second, the agent needs tools, which are APIs that allow it to interact with external systems. Third, Google includes what it calls an "orchestration layer" that manages the planning and logic operations. According to Google, these autonomous agents could significantly expand the capabilities of current language models.
According to internal documents obtained by TechCrunch, Google has been benchmarking its Gemini AI model against Anthropic's Claude. Google contractors are given up to 30 minutes per prompt to evaluate which model produces better outputs, focusing on criteria like truthfulness and comprehensiveness. Claude tends to be more safety-conscious in its answers compared to Gemini, according to Techcrunch. A Google DeepMind spokesperson confirmed that they do compare results across models, but stressed that they don't use Anthropic's models to directly improve Gemini, which would go against Anthropic's ToS. Also, this kind of competitive benchmarking is common in the AI industry - companies regularly benchmark their models against competitors to understand where they stand. Moreover, Google is an investor in Anthropic.