IRT SystemX announces the signing of a Memorandum of Understanding with LINAGORA, the French leader in open-source software. This partnership aims to address trust issues in large language model (LLM) technologies and promote innovative and secure solutions for the industries of tomorrow.
This collaboration will lay solid foundations for future joint research to improve the explicability, robustness and integration of LLMs through the development of open source solutions.
Shared objectives for significant advances
The two entities have decided to initiate their collaboration around the following objectives:
- Improve the explainability of LLMs to understand how they generate their responses.
- Strengthen their robustness against manipulations or malicious inputs.
- Develop reliable trust scores to evaluate the performance of models in various contexts.
- Optimize their hybridization with specific knowledge bases.
- Define performance and reliability criteria specific to the industrial sector (aerospace, automotive, etc.).
- Develop an evaluation framework to objectively compare different language models based on a European culture.
- Identify potential funding sources to achieve these ambitions.
Towards concrete and applicable results
As part of this partnership, both parties commit to producing:
- A comprehensive state-of-the-art review on LLM trust.
- Implemented solutions to enhance trust in these models.
- A guide of recommendations for integrating trust principles into the development and use of LLMs.
- A methodological framework for the evaluation and certification of LLMs in the industrial context.
- An open-source software tool for the automated evaluation of LLMs.
- Academic and industrial publications.
- A white paper on the challenges and best practices for certifying LLMs in critical industries.
