News Summary

Simular AI has introduced a new AI agent called S2, which is designed to handle complex tasks by using multiple specialized models. The startup, founded by Ang Li, a former researcher at Google DeepMind, aims to improve the performance of AI agents on tasks like using computers and smartphones. S2 stands out by integrating a powerful general-purpose AI model for planning, alongside smaller open source models for specific tasks such as interpreting web pages. This multi-model approach allows S2 to surpass other agents in benchmarks like OSWorld and AndroidWorld, achieving higher completion rates for tasks involving numerous steps. Victor Zhong, a computer scientist at the University of Waterloo, supports this approach, anticipating that future AI models will incorporate training data to better understand graphical user interfaces. Despite its advancements, S2 and similar agents still face challenges with edge cases and complex scenarios, highlighting the ongoing development needed in AI agent technology.