March 2024 Summaries

A customer support agent was designed to match the understanding of a human customer, with a focus on speed and iteration. The system used explicit intents for well-defined flows and a RAG architecture for product questions that change often. However, performance diverged from initial results after deployment, with out-of-domain questions performing poorly. To address this, techniques such as augmenting training data, using an LLM hybrid system, or creating more specific intents were explored. After iterating on the training dataset, validation accuracy improved by 23%, and further optimizations on the evaluation set resulted in a 14% accuracy gain. Adding more specific intents increased performance by capturing key concepts from the datasets and reducing out-of-domain classification errors.

The impact of prompt changes on the performance of large language models (LLMs) was studied, with a focus on intent classification for conversational AI. Researchers noticed that upgrading to a newer version of an LLM resulted in significant degradation in performance, highlighting the importance of careful prompt design. Through iterative testing and analysis, they identified specific changes to the prompt that improved model performance, including adding an indication of importance, changing from descriptions to actions, and incorporating one-shot examples. The study demonstrated the significance of prompt optimization for improving real-world performance of LLMs.

Our new brand represents endless scalability, perpetual power, and everlasting fun, reflecting the need for constant change and collaboration in technology. We're empowering developers to customize and extend AI Agents to meet unique team needs, focusing on building an intuitive and personalized experience. Our refreshed look reintroduces Voiceflow as a full stack platform beyond conversation design and no-code tooling. Recent updates include AI Knowledge Base, Content Management System, OpenAI model integrations, Developer APIs, web chat widget, and webchat extensions. We'll continue to drive innovation, connecting people across teams, organizations, and the industry.

Your simple Q&A agent should be accessible to most businesses, even small and low-volume companies, with tools like customGPT, Microsoft CoPilot, ChatBase, and GPT Builder helping teams launch agents quickly. A simple Q&A agent can automate basic tasks, freeing up customer success, sales, and operations teams from menial tasks. To get started, just upload a few documents and start asking questions, embedding the agent within a customer's existing search flow using Voiceflow APIs. The key to good responses is understanding how to format data well, using technology called Retrieval Augmented Generation (RAG). Adding actions to your simple Q&A agent can automate tasks such as routing support tickets, generating leads, and providing personalized content. Complex AI agents leverage customer data, drawing from internal databases, CRM systems, and external APIs to gather relevant information. They typically route inquiries to the most appropriate pathway, perform sophisticated tasks, and complete advanced actions like multi-step account changes or problem-solving support queries. Sophisticated AI agents can dynamically offer services, manage complex accounts, troubleshoot issues, automate decision making, personalize recommendations, and provide high-quality responses. To take your AI agent from baby steps to marathons, expand it with thoughtfulness and planning, considering the evolving nature of AI and human oversight requirements.