Upwork's CTO and Head of AI, Andrew Rabinovich, discusses the company's use of AI, particularly the Upwork Mindful AI (UMA) meta-agent, to facilitate interactions between clients and freelancers. Upwork employs multiple fine-tuned models and RAG (Retrieval Augmented Generation) systems to match freelancers with jobs, create job posts, and answer questions. Rabinovich highlights digital work as an ideal environment for training AI agents, allowing them to learn from mistakes with human feedback. He introduces Reinforcement Learning from Experience (RLEF), a new approach combining classical RL with LLMs, to enable agents to surpass human limitations. Looking ahead, the goal is for UMA to deliver results directly to clients by orchestrating both human and machine resources, with third-party agents integrated into the platform.
Sign in to continue reading, translating and more.
Continue