Salesforce AI Research has launched eVerse, a simulation environment that enhances the training of AI agents through synthetic data generation and reinforcement learning. This innovative tool is a step towards achieving enterprise general intelligence (EGI), aimed at creating reliable AI systems that excel in business applications. As stated by Silvio Savarese, the need for accuracy hovers around 99%, given that 90-95% accuracy is often unacceptable in practical applications.
One of the central challenges is addressing ‘jaggedness,’ where AI performs well on complex tasks but falters on simpler ones. To tackle this, eVerse enables learning from real-time feedback in a controlled environment, allowing for iterative training and testing. This approach eschews merely scaling model sizes in favor of experiential learning. Salesforce has successfully utilized eVerse to develop Agentforce Voice, a tool designed for voice conversations in real-world scenarios. The simulation replicates various complexities, including background noise and accents, enabling thorough training before deployment.
Additionally, UCSF Health has piloted eVerse to enhance healthcare billing agents, increasing their capability to handle inquiries effectively. Initially able to manage 70% of questions, the AI’s performance improved to 80% due to iterative learning facilitated by eVerse. This transformative tool is set to redefine how AI interacts within different enterprise contexts, especially in tackling complex tasks.
👉 Pročitaj original: CIO Magazine
