SPIE accelerates its roll-out of generative artificial intelligence

Malakoff, 23 May 2024 – SPIE ICS, a French digital services company and a subsidiary of SPIE, the independent European leader in multi-technical services in the areas of energy and communications, aims to accelerate the roll-out of generative artificial intelligence to its employees and customers. SPIE ICS is convinced that there are two ways for companies to use “GenAI”^[1]. First, “Generative AI for all”: a general-purpose daily assistant for all employees. This is followed by “Generative AI for everything”: specialised “business area” assistants focusing on highly specific tasks.

Generative AI for all

In order to give all its employees a “daily assistant”, SPIE ICS has provided them with an interface allowing access to a general-purpose large language model that ensures that company data is secure. This interface allows users to analyse or generate text or code, and it takes attached documents into account. A quarter of SPIE ICS employees have already used this “AI for all”, with a total of 400 unique monthly users, or 12% of the workforce.

Around one hundred employees also have access to a more advanced interface. This allows users to choose their model, including multimodal models, to trigger Internet searches and to create images and diagrams.

Building on this experience, SPIE ICS then trained nine customers in generative AI during innovation committee meetings. Two of these customers are currently testing its “AI for all”.

Generative AI for everything

In order to rise to the challenges of digital transformation and optimised performance for its customers, SPIE ICS is now developing “specialised business area assistants”. Most companies first attempt to connect generative AI to their knowledge bases for consultation. The assistant then relies on pre-prompting and Retrieval-Augmented Generation (RAG). SPIE ICS has successfully tested one of these assistants at a customer’s premises and has produced them for two further customers.

Generative AI, both sovereign and cost-effective 

Companies and public authorities want to retain control of their data, which should not be sent to free AI platforms due to the risk of data becoming public. This data must be protected as much as possible from extra-territorial regulations and, where possible, remain in the European Union or even in France. SPIE ICS works with Mistral AI and has already rolled out their models for its own use and for one of its customers.

SPIE ICS has the ability to integrate open-source “on premise” models using “Edge” infrastructure^[2] dedicated to GenAI. Finally, SPIE ICS is closely monitoring Cloud computing operators that are developing inference capabilities in France, which are an essential building block for sovereign end-to-end products.

The environmental impact of the most powerful language models is considerable, in terms of both energy and water consumption. It is therefore essential to select the right model based on requirements.

“Generative AI’s emergence is a new step in the digital transformation of businesses. It represents another lever alongside current automation technologies (predictive AI, NLP^[3] etc.). For organisations, this is an opportunity to question the benefit of automating the processes at the heart of their value creation.” François Guéno, Director of Innovation and R&D at SPIE ICS.

Arthur Mensch, CEO and co-founder of Mistral AI, adds: "Mistral AI is delighted to provide its cutting-edge technologies to SPIE. By using our customisable and portable AI models, SPIE is helping to make cutting-edge AI both pervasive and accessible to all developers, a mission that Mistral AI holds dear."

_{[1] Generative Artificial Intelligence}