))
Preventing Hallucination in Agentic AI
Manage episode 449763508 series 3598876
コンテンツは Vivun によって提供されます。エピソード、グラフィック、ポッドキャストの説明を含むすべてのポッドキャスト コンテンツは、Vivun またはそのポッドキャスト プラットフォーム パートナーによって直接アップロードされ、提供されます。誰かがあなたの著作物をあなたの許可なく使用していると思われる場合は、ここで概説されているプロセスに従うことができますhttps://ja.player.fm/legal。
Most AI knows how to respond—but does it know how to solve real SE problems?
In this episode, CEO Matt Darrow sits down with Vivun’s Sr. Machine Learning Engineer, Chen Liang, to explore why most AI just isn’t cut out for sales engineering—yet.
While language models like ChatGPT are powerful, they lack the insider knowledge and finesse that make sales engineers irreplaceable. Matt and Chen dive into Vivun’s unique approach to building AI agents that don’t just predict answers but deliver reliable, actionable insights grounded in real-world experience.
Discover how structured knowledge framework and top-down procedural graph prevent AI hallucinations and keep AI agents aligned with SE needs.
In this episode, you’ll learn:
- Structured Knowledge for Smarter AI: Crafting a reliable AI system begins with a structured approach to knowledge, ensuring AI agents operate with accuracy and industry insight.
- Preventing AI Hallucinations: Large language models often generate misleading responses, but with domain-specific guidance, AI can deliver more trustworthy results.
- Real-Time Improvement Through User Feedback: Tracking user interactions, like the “frustration index,” helps AI continuously evolve and meet real-world demands.
- AI as a Collaborative Team Player: Rather than just a tool, AI can become a partner in sales engineering, supporting complex decision-making with structured, actionable insights.
Things to listen for:
(00:00) Why AI alone isn’t enough for sales engineering
(05:09) How structured knowledge reduces AI hallucinations
(07:44) Using a knowledge graph to keep AI on track
(10:28) The importance of evaluation for AI reliability
(13:25) Real-world examples of AI’s limitations without domain knowledge
(15:23) Quantitative vs. qualitative methods for AI evaluation
(16:36) Real-time feedback and the “frustration index”
(18:46) Human-in-the-loop and automated quality checks
(20:05) Enhancing AI with ongoing data and real-time adjustments
24 つのエピソード
シェア
