Divide & Classify, Conquering Enterprise LLMs, Andrew Ahn & Dr Robin Bloor, Hosted by Eric Kavanagh

Divide & Classify, Conquering Enterprise LLMs was a webinar recorded in December 2023 with The Bloor Group Briefing Room, with Host Eric Kavanagh, CEO of The Bloor Group, Andrew Ahn, Founder & CEO of Praxi along with Dr Robin Bloor, Co-Founder & Chief Analyst of The Bloor Group.

This conversation explores the use of large language models (LLMs) in the enterprise and the challenges and solutions associated with them. The discussion covers topics such as ground truth, hallucinations in LLMs, embedding strategy, data curation, and automation. The importance of specific use cases and the potential impact of LLMs on claims processing are also highlighted. The conversation emphasizes the need for a focused approach to LLM projects and the potential benefits of vector databases in enhancing LLM performance. Overall, the conversation provides insights into the evolving landscape of LLMs in the enterprise. This conversation explores the possibilities and implications of modifying language models (LLMs) and using them for specialized use cases. It discusses the potential of LLMs to analyze corporate data and provide insights for decision-making. The conversation also touches on the feasibility of running LLMs on-premises and the potential reduction in size and cost of LLMs. Additionally, it explores the concept of fine-tuning LLMs to ensure they pull facts from trusted repositories. The conversation concludes by discussing the impact of LLMs on business intelligence and analytics and the overall revolution of language models.

Takeaways
Ground truth is crucial in training LLMs and ensuring accurate results.
Embedding strategy and metadata play a key role in training and utilizing LLMs effectively.
Specific use cases and targeted LLMs can provide more accurate and relevant results.
Data curation and automation are essential for maximizing the benefits of LLMs in the enterprise.
The potential of LLMs in claims processing and other areas can greatly impact efficiency and accuracy. Language models can be modified and customized to suit specific use cases and industries.
LLMs have the potential to analyze corporate data and provide valuable insights for decision-making.
Running LLMs on-premises is possible, but it may require a larger solution and infrastructure.
Fine-tuning LLMs can ensure they pull facts from trusted repositories and provide accurate information.
LLMs have the potential to revolutionize business intelligence and analytics by enabling interactive and real-time data analysis.

00:00 Introduction and Future Proof Webinar Series
00:49 Conquering LLMs for the Enterprise
01:45 The Concept of Ground Truth
02:11 Giving LLMs True Truths
03:07 The Issue of Hallucinations in LLMs
04:02 Innovation to Augment LLMs and Mitigate Hallucinations
04:29 Training LLMs and Raising Children
05:24 The Importance of Embedding Strategy
06:16 The Algebra of Data and Ludwig Wittgenstein
07:36 The Impact of LLMs on Enterprise Technology
08:03 The Power of LLMs in Customer Service and Information Architecture
09:28 Data Curation and Automation with Praxi Data
10:52 Praxi Data's Objectives and Missions
11:22 Focus on Specific Use Cases and Specialized LLMs
13:49 Enterprise Challenges and Solutions
17:04 Targeted LLMs and Collaboration
18:02 Relating Non-Structured and Structured Data
19:01 Addressing Regulatory and Security Requirements
20:26 The Missing Link: Data Curation
22:21 The Foundational Stack of PraxyData
23:46 Derivative Objects and Metadata
26:41 Using LLMs to Analyze Corporate Data
27:41 The Cost and Feasibility of LLM Projects
31:08 Choosing the Right LLM and Avoiding Mistakes
36:10 The Rise of Interactive User Interfaces
37:35 The Potential of LLMs in Unearthing Patterns
38:05 The Role of Vector Databases
42:29 Determining the First Prototype for an LLN Project
46:13 The Use of Vector Databases in LLN Projects
51:37 The Importance of Specific Use Cases for LLNs
53:46 The Potential Impact of LLNs on Claims Processing
54:26 The Role of Metadata and Lineage in LLN Projects
54:55 Modifying Language Models
57:19 Specialized Use Cases
57:46 Using LLMs for Corporate Data
58:43 Connecting LLMs to Proprietary Data
01:00:08 Running LLMs On-Premises
01:02:07 Reducing LLM Size and Cost
01:03:03 Fine-Tuning LLMs for Trusted Repositories
01:04:57 Unknown Costs of LLMs
01:08:24 Impact on Business Intelligence and Analytics
01:10:48 The Revolution of Language Models