Search results for UC Berkeley
AI & Enterprise
Google DeepMind unveils roadmap to monitor and block AI agents
Google DeepMind has set out a control framework that treats AI agents as potential insider threats rather than software tools, Axios reported. In an \"AI Control Roadmap,\" it proposed raising monitoring and intervention as agents become more capable, including real-time access blocking or forced shutdowns. DeepMind applied real-time monitoring to its Gemini Spark agent after analysing about 1 million coding-agent tasks. It said most detected issues were unintentional.
Industry
Nvidia unveils ENPIRE self-training robot framework, achieves 99 percent success rate with 8 robots
Nvidia unveiled ENPIRE, a framework that uses AI coding agents to train real robots without human supervision. The company published a paper with Carnegie Mellon University and UC Berkeley. ENPIRE begins with a one-time human setup of reset routines and reward functions, then agents search papers, choose learning methods and rewrite code to test on robots. Nvidia ran eight dual-arm robots that shared results via Git and reported 99 percent success across four real-world tasks.
AI & Enterprise
Tech CEOs warned against overestimating AI automation as layoffs mount
Criticism is growing in the tech industry that chief executives are overestimating how much work AI can actually automate. Box founder Aaron Levie said many CEOs appear collectively caught in an “AI delusion” and often mistake early prototypes for real job replacement. Studies cited in the article question whether AI adoption leads to measurable productivity gains, while research suggests human-level performance remains limited in many tasks and management approvals could become new bottlenecks.
