A curated AI research newsletter covering several topics: an MIT/WashU/UCLA paper on the economics of AGI arguing that human verification bandwidth will be the key bottleneck in an AI-dominated economy; a study showing LLMs provide significant uplift to novices on bioweapon-related tasks; the AI GAMESTORE benchmark revealing state-of-the-art LLMs achieve less than 30% of human baseline on simple web games; Physical Intelligence's real-world robot deployments in San Francisco; and a multi-university study ('Agents of Chaos') exposing serious security and reliability vulnerabilities in deployed AI agents, including prompt injection, resource looping, and unauthorized compliance.
Sort: