Just one year after its creation, the NRB – ULiège Research Chair on Artificial Intelligence is already delivering promising results. Researchers from the University of Liège and NRB unveil TreeFinder, an innovative method that makes language models (LLMs) traceable, verifiable and auditable. This is a major step forward for more transparent and responsible AI.
Signed in July 2024, the NRB – ULiège AI Chair was born out of a shared ambition: to accelerate the adoption of ethical and effective Artificial Intelligence for the benefit of businesses, the public sector and society.
Created within the Faculty of Applied Sciences at ULiège, this chair is dedicated to the research and development of language models (LLMs) applied to software engineering.
The objective is to transform scientific advances into concrete solutions capable of optimising business processes, strengthening the competitiveness of companies and encouraging innovation.
This chair between NRB and ULiège is undoubtedly an exemplary model of partnership between industry and academia. It fuels fundamental research and enables NRB to conquer new markets.
The result of this collaboration, TreeFinder addresses a major concern: the traceability of responses generated by large language models.
Until now, LLMs have functioned as ‘black boxes,’ making it difficult to understand how they produce their responses. This is a crucial issue in sensitive sectors such as healthcare, law and finance. Today, LLMs miss key phrases in long contexts, are distracted by noise and produce answers that are difficult to verify.
TreeFinder is a game changer.
This simple, independent (model-agnostic) method identifies the specific phrases in a long document that actually influenced the model's response.
The result: more reliable, auditable and certifiable question-answering systems that can explain why they answer the way they do.
LLMs offer impressive performance, but there are still significant challenges to overcome to make them auditable and trustworthy. Thanks to our collaboration with NRB, we have been able to develop a method that bridges the gap between theoretical research and the concrete needs of businesses.
The TreeFinder algorithm is based on two key principles:
By combining these signals, TreeFinder isolates the sentences that ‘really matter’ in generating a response, while eliminating noise.
This hierarchical, fast and accurate approach makes it possible to:
In practical terms, in the medical field, TreeFinder can justify a clinical summary with the exact sentences from a patient's file.
In the legal sector, it can link a response to specific passages in a contract or court ruling.
And for businesses, it secures internal AI engines by explaining which sources influenced the response given.
Working with NRB has been a real source of inspiration. It has enabled us to identify concrete problems encountered in the deployment of AI and to provide answers based on fundamental research. A real synergy has been created between researchers and engineers to transform these challenges into innovations.
Major technological, economic, educational and ethical challenges can only be met by joining forces. This is the whole point of this chair, which demonstrates the power of open ecosystems and sustainable collaborations between research and business.
This dynamic illustrates the value of cross-collaboration between universities and businesses: a win-win model, where research benefits from real-world use cases and industry draws on rigorous scientific expertise to build trusted solutions.
’Contributive Attribution for Question Answering via Tree-based Context Pruning'
Work carried out by Lize Pirenne, Gaspard Lambrechts, Norman Marlier, Maxence de la Brassinne Bonardeaux, Gilles Louppe and Damien Ernst, with the support of Wallonia and the NRB Research Chair on LLMs.
The code is available in open source on GitHub: github.com/Pangasius/TreeFinder
Because trust in science depends as much on the quality of discoveries as on rigorous practices, a network of scientific integrity ambassadors has been launched as part of the FWB-CoARA project.
MARE-WIND – Biodiversity and Offshore Wind: Understanding Transformative Impacts on Marine Ecosystems – has been selected for funding under the Biodiversa+ BiodivTransform call.
This prize and grant are awarded to them for their work in popularising science with a view to improving the dissemination of knowledge.
