NB 30 - Addressing the AI Hallucination Problem

Special guest Eyelevel.ai CEO Neil Katz joins Geoff and Greg to discuss generative AI's hallucination problem. Large language models' (LLMs') propensity to hallucinate in their answers represents one of the biggest barriers to enterprise adoption. Eyelevel boasts a 95% accuracy on private instance responses by using its APIs and tools to prepare proprietary data for LLM consumption. The trio dives into the LLM marketplace, including discussions about why brands choose to implement a private instance, how the LLM market has evolved, and what causes the hallucination problem. Then, they discuss the enterprise data problem and how retrieval augmented generation (RAG) techniques still need additional help to strengthen LLM responses. Chapters include: 0:00 Start 4:40 Private instance versus licensing enterprise editions of LLMs 7:32 Eyelevel’s Air France implementation achieving 95% success rates 12:29: The need for enterprise data preparation 18:14 The hallucination problem with LLMs and RAG approaches 28:23 How governance can or cannot help enterprises 32:32 Why some use open source versus proprietary LLMs 39:12 The future of AI and an incredible vision Learn more about Eyelevel at https://www.eyelevel.ai/ or contact Neil Katz via LinkedIn at https://www.linkedin.com/in/neilkatz/ Learn more about your ad choices. Visit megaphone.fm/adchoices