Former Google and Apple Researchers Launch a Startup to Build AI’s Missing Feedback Loop (www.wired.com)

A group of former AI researchers from top tech companies, including Google, Apple, and OpenAI, has announced the launch of Trajectory, a startup aimed at revolutionizing how AI models learn from real-world user interactions. The company seeks to address the longstanding challenge of creating AI systems that can improve continuously post-deployment, a need underscored by experts like Turing Award winner Richard Sutton. Trajectory has successfully raised $15 million in seed funding, with notable investors including Google DeepMind’s chief scientist, Jeff Dean, and AI pioneer Fei-Fei Li. Trajectory's platform utilizes a post-training approach, where models are regularly updated based on user feedback and specific business needs, improving the performance of AI tools in various domains beyond coding. While current implementations reflect an early version of continual learning—with models updating weekly—the team envisions a future where updates can occur multiple times a day. This ambition aims to alleviate the dependency on embedded engineering teams for maintaining AI systems, thus making continual learning more accessible for businesses, including Fortune 500 companies.