Preeti
Pleais, a start-up, has developed smaller language models for specific bureaucratic and administrative purposes, focusing on transparency, ethics, and copyright compliance. Co-founder Pierre-Carl Langlais explains the need for dedicated solutions in light of a shortage of GPU infrastructure in Europe. The company aims to provide customers with models that meet strict regulatory requirements.
Langlais was inspired by a Chinese project, QWEN, and aims to provide affordable language models that can run on GPUs or CPUs for public services, banking, and healthcare sectors. The company plans to sell related products like integrated research tools rather than charging for the models themselves. Their public service model, Albert, is designed to simplify administrative language while maintaining ethical standards.
Pleais trains their models on open corpora of out-of-copyright texts to avoid competition obstacles in the AI field. They believe that smaller models can be highly effective in specialized contexts and targeted uses like document analysis and administrative procedures. The company anticipates that regulated professions and public services will make greater use of their models due to the European AI Act’s principle of responsibility for generative AI and the need for transparency in both the private and public sectors.
Source link
Source link: https://hellofuture.orange.com/en/p-c-langlais-pleais-our-language-models-are-trained-on-open-corpora/
GIPHY App Key not set. Please check settings