Giskard leads GenAI Evaluation in France 2030's ArGiMi Consortium

As part of the ArGiMi consortium, Giskard has been selected to lead a new research & development initiative in Generative AI innovation. This project, called "Communs numériques pour l'intelligence artificielle générative," is part of the France 2030 program launched by Bpifrance in October 2023.

The ArGiMi consortium, comprising Artefact, Giskard, and Mistral AI, aims to develop the next generation of French Large Language Models (LLMs) for businesses, with a focus on creating models that are reliable, ethical, and compliant with current regulations.

ArGiMi’s goals: creating safer Large Language Models

Currently, only 5% of Gen AI projects reach production, mainly due to difficulties in ensuring model reliability, financing the projects, and efficient deployment. Our consortium aims to tackle these challenges:

1. Mistral AI will develop new LLMs,

2. Artefact will focus on integrating these models into business applications, and

3. Giskard will ensure model quality, conformity, and security.

The consortium will collaborate with public institutions, including INA and BnF, as well as private organizations to enrich language models with French data. While respecting copyright & data privacy, we'll work with partners to ensure our models are trained on diverse, high-quality French data that represents various sectors and use cases.

Commitment to Open-Source

A key aspect of this project is its commitment to open-source development. We believe that sharing evaluation methodologies, models, and data openly will foster responsible innovation and accessibility. 

This aligns with Giskard's philosophy of transparency and collaboration in AI development, as we’ve been committed to open-source AI from day one.

By democratizing access to these models, it allows all organizations and individual researchers to actively participate in AI development. This transparency increases trust, as the community can scrutinize models, quickly identifying and correcting biases, errors, or vulnerabilities, in order to build more reliable, secure & high-performance AI models.

Key deliverables and impact on AI innovation

By winning this call for projects, the ArGiMi consortium confirms France's position as a leader in AI, contributing to responsible AI innovation.

The ArGiMi consortium will work on several key deliverables:

1. Development of new state-of-the-art open-source LLM in French, for specific enterprise needs and sectors.

2. Model evaluation and reliability: This is where Giskard's expertise comes to the forefront. We'll lead efforts in evaluating models, developing open-source evaluation methods for LLMs.

3. Deployment enablers: We'll contribute to creating tools that simplify the integration and fine-tuning of LLMs, including open-source datasets for benchmarking and fine-tuning.

Conclusion

This project marks a significant step forward in developing reliable and ethical Generative AI models in France, and Giskard's role in ensuring model quality, conformity, and security will be key. As part of the ArGiMi consortium, we’ll be making LLMs more performant, trustworthy, and aligned with societal values. 

‍Reach out to us today to learn more about how we can help you to ensure your models are safe and reliable.