Pubblicata Lunedì 11 Novembre 2024 10:18
DeMaCS - 2024-11-13 LogicLM: robust AI application through Logic Programming
Research Seminars in AI "LogicLM: robust AI application through Logic Programming"
- Title: LogicLM: robust AI application through Logic Programming
Speaker: Evgeny Skvortsov
Partecipa online: https://bitly.cx/ZHag2F
Abstract
Advanced large language models have numerous applications in software engineering, and data analysis. In particular using LLM to translate user’s request from the natural language to a rigorous specification allows for highly efficient computer-human interaction. Where user would need to work through multiple menus or write a program in a specific language, now user can simply state what they would like to achieve. However this opportunity comes with a risk: there is never a guarantee that LLM will understand the user correctly. First of all, the ability of LLMs to hallucinate (though the proper term for this phenomenon is confabulate as Geoff Hinton points out) is widely known and discussed. And furthermore natural language is inherently ambiguous and it can be hard to detect an error in natural language.
LogicLM is an open source project to employ LLM’s capabilities in data analysis while keeping user in control of the process via use of logic programming approach. LogicLM allows you to describe your data in terms measures, dimensions and filters, which are defined in Logica, a logic programming language enriched with recursion, functions and other practical features.