ZurichNLP #13

Besta Maciej (ETH Zurich) on "Effective and Efficient LLM Ecosystems with Graph of Thoughts and Beyond": Graph of Thoughts (GoT) is a framework that advances reasoning and prompting capabilities in large language models (LLMs) beyond those offered by paradigms such as Chain-of-Thought or Tree of Thoughts (ToT). The key idea and primary advantage of GoT is the ability to model the information generated by an LLM as an arbitrary graph, where units of information ("LLM thoughts") are vertices, and edges correspond to dependencies between these vertices. This approach enables combining arbitrary LLM thoughts and whole reasoning chains into synergistic outcomes, preserving the experience from failures instead of discarding it, distilling the essence of whole networks of thoughts, or enhancing thoughts using feedback loops. We illustrate that GoT offers advantages in the outcome accuracy and token cost over state of the art on different tasks such as keyword counting, NDA merging, or sorting. We then discuss how to augment GoT with more powerful capabilities that include tool usage and RAG using an abstraction called Topologies of Reasoning (ToR) that forms a blueprint for building an effective and efficient LLM ecosystem. We finalize with an overview of our recent developments within ToR, which include CheckEmbed (a method for robust verification of LLM outcomes and hallucination detection) and Multi-Head RAG (a design that enhances RAG when dealing with multi-aspectual data).

Riccardo M. (ETH Zurich) on "Accelerating Industrial Programming with AI: Bridging LLM and Industrial Automation": Despite the growing sophistication of industrial automation, the complexity of programming production lines remains a bottleneck for engineers. In addition, automation engineers are becoming very rare, and the current number cannot satisfy the high demand, leading to costly delays for System integrators. At Xelerit, we are pioneering a solution that leverages LLM to simplify and expedite the process of generating code for industrial robots and PLC. By bridging the gap between human instructions and machine code, our platform empowers engineers to focus on innovation and design rather than tedious coding tasks, speeding up the process and reducing the number of engineers needed for projects. This talk will explore the role of LLM combined with techniques such as RAG, COT and close-loop in transforming industrial programming, the challenges we are overcoming, and the potential this technology holds for revolutionizing industrial automation.