最新报道:Top AI researchers from OpenAI, Google DeepMind, and Nvidia discussed deployment challenges for AI agents at UC Berkeley's Agentic AI Summit, highlighting gaps between controlled demos and real-world reliability. Experts cited safety concerns and multi-task consistency as key hurdles, while noting infrastructure improvements and hardware advances show promise. Current progress remains confined to narrow domains like coding, with researchers emphasizing the need for technological breakthroughs and collaboration to achieve transformative AI agents by 2025.