Knowledge Graphs, Theorem Provers & Language Models — Vijay Saraswat & Nikolaos Vasiloglou | KGC '24 | The Knowledge Graph Conference

This podcast episode delves into the transformative landscape of computer science brought about by large language models (LLMs) and their integration with knowledge graphs, illustrating how these technologies can reshape programming methodologies and tackle complex real-world problems. The speakers explore the implications of advancements like in-context learning and many-shot learning, the challenges faced in programming with LLMs, and the potential for LLMs to bridge the gap between optimization and language processing. Ultimately, they envision a promising future where LLMs serve as a universal platform for reasoning and optimization, driving innovation while acknowledging the hurdles that still remain.

Outlines

Sign in to continue reading, translating and more.

Open full episode in Podwise

Knowledge Graphs, Theorem Provers & Language Models — Vijay Saraswat & Nikolaos Vasiloglou | KGC '24

The Knowledge Graph Conference

Setting the Stage for LLM Advancements

The Rise of LLMs: A New Era in Computer Science

Understanding the Power of Transformers and Large Language Models

The Challenge of Programming with LLMs: Finding the Right Inputs

LLM Advancements: Increased Context Size and In-Context Learning

The Promise and Challenges of In-Context Learning

LLMs as Universal Computers: Exploring Computational Abilities

Deciphering LLM Computation: Bounded Input, Bounded Output, and Time Complexity

The Impact of LLMs on Information Consumption and the Challenge of Authenticity

Extracting Knowledge Graphs from Text using LLMs

The Limits of LLMs in Complex Domains: Consistency and Domain-Specific Understanding

Bootstrapping Knowledge Graphs with In-Context Learning: Building Coherent Ontologies

The Power of GraphRAG for Retrieval-Augmented Generation

Unveiling the Mechanism Behind In-Context Learning: Latent Variables and Pattern Recognition

ICL as Implicit Fine-Tuning: Attention as a Mechanism for In-Situ Updates

ICL for Regression and Beyond: Exploring LLMs' Abilities for Computational Tasks

The Power of Many-Shot Learning: Unleashing LLMs' Potential for Complex Tasks

Can LLMs Handle Combinatorially Hard Problems? Exploring the Boundaries of LLM Computation

Self-Discovery in LLMs: Automating Reasoning and Problem Solving

Programming with LLMs: Exploring Techniques for Reasoning Module Selection and Specialization

The Importance of Multi-Step Prompting for LLMs: Breaking Down Complex Tasks

LLM-Based Task Execution: Generating Schedules and Addressing Constraints

Addressing LLM Unreliability: Fact-Checking and Self-Improvement Techniques

Using LLMs for Repair and Iteration: Improving LLM-Generated Plans

The Role of Knowledge Graphs in LLM Verification and Repair

Using LLMs for Distance Estimation: Leveraging Relational Information for Efficient Computation

The Limitations of Graph Databases for Constraint Satisfaction in Complex Planning Problems

Combining Knowledge Graphs and LLMs for Solving Combinatorial Optimization Problems

LLMs for Solving the Diet Problem: Bridging the Gap between Optimization and Language Models

The Future of Gen-AI Programming: A Universal Platform for Reasoning, Optimization, and Knowledge Graphs

Confessions of a Gen-AI Programmer: The Challenges and Opportunities of Programming with LLMs

Knowledge Graphs, Theorem Provers & Language Models — Vijay Saraswat & Nikolaos Vasiloglou | KGC '24

The Knowledge Graph Conference

00:00Setting the Stage for LLM Advancements

Setting the Stage for LLM Advancements

02:10The Rise of LLMs: A New Era in Computer Science

The Rise of LLMs: A New Era in Computer Science

06:17Understanding the Power of Transformers and Large Language Models

Understanding the Power of Transformers and Large Language Models

09:21The Challenge of Programming with LLMs: Finding the Right Inputs

The Challenge of Programming with LLMs: Finding the Right Inputs

10:01LLM Advancements: Increased Context Size and In-Context Learning

LLM Advancements: Increased Context Size and In-Context Learning

14:38The Promise and Challenges of In-Context Learning

The Promise and Challenges of In-Context Learning

16:57LLMs as Universal Computers: Exploring Computational Abilities

LLMs as Universal Computers: Exploring Computational Abilities

20:11Deciphering LLM Computation: Bounded Input, Bounded Output, and Time Complexity

Deciphering LLM Computation: Bounded Input, Bounded Output, and Time Complexity

21:34The Impact of LLMs on Information Consumption and the Challenge of Authenticity

The Impact of LLMs on Information Consumption and the Challenge of Authenticity

23:32Extracting Knowledge Graphs from Text using LLMs

Extracting Knowledge Graphs from Text using LLMs

28:01The Limits of LLMs in Complex Domains: Consistency and Domain-Specific Understanding

The Limits of LLMs in Complex Domains: Consistency and Domain-Specific Understanding

32:50Bootstrapping Knowledge Graphs with In-Context Learning: Building Coherent Ontologies

Bootstrapping Knowledge Graphs with In-Context Learning: Building Coherent Ontologies

33:35The Power of GraphRAG for Retrieval-Augmented Generation

The Power of GraphRAG for Retrieval-Augmented Generation

40:14Unveiling the Mechanism Behind In-Context Learning: Latent Variables and Pattern Recognition

Unveiling the Mechanism Behind In-Context Learning: Latent Variables and Pattern Recognition

44:21ICL as Implicit Fine-Tuning: Attention as a Mechanism for In-Situ Updates

ICL as Implicit Fine-Tuning: Attention as a Mechanism for In-Situ Updates

46:41ICL for Regression and Beyond: Exploring LLMs' Abilities for Computational Tasks

ICL for Regression and Beyond: Exploring LLMs' Abilities for Computational Tasks

48:20The Power of Many-Shot Learning: Unleashing LLMs' Potential for Complex Tasks

The Power of Many-Shot Learning: Unleashing LLMs' Potential for Complex Tasks

50:10Can LLMs Handle Combinatorially Hard Problems? Exploring the Boundaries of LLM Computation

Can LLMs Handle Combinatorially Hard Problems? Exploring the Boundaries of LLM Computation

51:19Self-Discovery in LLMs: Automating Reasoning and Problem Solving

Self-Discovery in LLMs: Automating Reasoning and Problem Solving

52:11Programming with LLMs: Exploring Techniques for Reasoning Module Selection and Specialization

Programming with LLMs: Exploring Techniques for Reasoning Module Selection and Specialization

59:01The Importance of Multi-Step Prompting for LLMs: Breaking Down Complex Tasks

The Importance of Multi-Step Prompting for LLMs: Breaking Down Complex Tasks

1:00:27LLM-Based Task Execution: Generating Schedules and Addressing Constraints

LLM-Based Task Execution: Generating Schedules and Addressing Constraints

1:01:06Addressing LLM Unreliability: Fact-Checking and Self-Improvement Techniques

Addressing LLM Unreliability: Fact-Checking and Self-Improvement Techniques

1:03:11Using LLMs for Repair and Iteration: Improving LLM-Generated Plans

Using LLMs for Repair and Iteration: Improving LLM-Generated Plans

1:06:53The Role of Knowledge Graphs in LLM Verification and Repair

The Role of Knowledge Graphs in LLM Verification and Repair

1:10:42Using LLMs for Distance Estimation: Leveraging Relational Information for Efficient Computation

Using LLMs for Distance Estimation: Leveraging Relational Information for Efficient Computation

1:13:51The Limitations of Graph Databases for Constraint Satisfaction in Complex Planning Problems

The Limitations of Graph Databases for Constraint Satisfaction in Complex Planning Problems

1:15:01Combining Knowledge Graphs and LLMs for Solving Combinatorial Optimization Problems

Combining Knowledge Graphs and LLMs for Solving Combinatorial Optimization Problems

1:18:07LLMs for Solving the Diet Problem: Bridging the Gap between Optimization and Language Models

LLMs for Solving the Diet Problem: Bridging the Gap between Optimization and Language Models

1:23:28The Future of Gen-AI Programming: A Universal Platform for Reasoning, Optimization, and Knowledge Graphs

The Future of Gen-AI Programming: A Universal Platform for Reasoning, Optimization, and Knowledge Graphs

1:24:53Confessions of a Gen-AI Programmer: The Challenges and Opportunities of Programming with LLMs

Confessions of a Gen-AI Programmer: The Challenges and Opportunities of Programming with LLMs