Stanford's Hottest "Town" Open-Sourced: 25 AI Agents Date! Tutorial Included

Stanford University has open-sourced a project of 25 AI agents in a "town". These agents live, work, socialize, make friends, and even fall in love in the virtual world. Each agent has a unique personality and backstory. The Stanford Agent Town is considered one of the most exciting AI Agent experiments in 2023, sparking discussions about multiple AI agents. The project address is https://github.com/joonspk-research/generative_agents.

Nobel Laureate Says AI Is Weakening the Competitiveness of Graduates from Top Universities

Michael Levitt, a Nobel Prize winner in Chemistry, stated at the 2025 Sustainable Global Leaders Conference that artificial intelligence is disrupting the education system, and the importance of academic degrees will decline. He pointed out that AI has democratized access to knowledge, and previously education was the gateway to knowledge, but now anyone can easily access knowledge through AI.

A Single Sentence Can Change AI's Creative Potential: Study Finds Simple Prompts Can Significantly Improve Output Diversity

A team from Stanford and other universities proposed the 'language sampling' method, which improves the creative diversity of generative AI by asking the model to generate five responses and their probabilities in the prompt. This method applies to both language and image models, and can stimulate richer creative outputs.

Stanford Research Team Releases AgentFlow: A Next-Generation Reinforcement Learning Framework for Modular, Tool-Using AI Agents

Stanford University releases the trainable intelligent agent framework AgentFlow, enhancing AI decision-making through a modular design. The framework includes four core modules: planner, executor, validator, and generator, which work together through explicit memory. The planner sets sub-goals and selects tools, the executor calls the tools, the validator checks the continuity of the process, forming a closed-loop intelligent decision-making system.

Gemini 2.5 Computer Use: Google Launches an AI Agent Model That Understands Web Pages

Google launched the Gemini 2.5 Computer Use AI model, which gives AI agents the ability to interact with web pages. This model can navigate, fill out forms, and perform complex operations in a browser through visual understanding and reasoning, breaking through the traditional interface limitations that require human intervention, opening up new frontiers in AI agent applications.

Former Microsoft Executive Founded AI Company Maximor to Break the Dependence on Excel for Financial Management

Excel remains the core tool for financial management, with many companies relying on it for tasks such as closing accounts, data verification, and audits. A former Microsoft executive founded the company Maximor, developing an AI agent to replace traditional spreadsheets, optimizing the workflow of financial teams and improving efficiency.