Pleias

Pleias builds a data layer and energy-efficient LLMs to enhance AI performance for enterprise applications using ethical, non-copyrighted data.

Description

Pleias is a web application project focused on building a specialized data layer to enhance AI performance, particularly for enterprise applications. The project develops energy-efficient large language models (LLMs) tailored for information-intensive and highly-regulated industries. It offers three main components: STRATUM, which provides AI-native tooling for agents; SYNTH, which generates synthetic data for specialized training; and COMMON CORPUS, an open data resource for AI. The team comprises experts with backgrounds in AI research, data science, and engineering, and the project emphasizes ethical AI development by using non-copyrighted, public domain data to train models. Pleias aims to address challenges in AI data curation, toxicity filtering, and efficient tokenization, making AI more accessible and effective for specialized use cases.

Features

• STRATUM: AI-native tooling for agents to speed up agentic AI
• SYNTH: Synthetic data generation for specialized training needs
• COMMON CORPUS: Open data resource for AI, focusing on public domain content
• Energy-efficient LLMs designed for information-intensive and regulated industries
• Ethical AI development using non-copyrighted data to avoid legal issues
• Advanced data curation pipelines for toxicity filtering and safety
• Research-driven approach with innovations like Picky BPE for efficient tokenization
• Team of experts with PhDs and experience in AI, data science, and engineering

Benefits

• Enhances AI performance through specialized data layers and tooling
• Reduces legal risks by using non-copyrighted, public domain data for training
• Improves model efficiency and safety with advanced data curation and toxicity filtering
• Supports specialized use cases in regulated industries like healthcare and community services
• Promotes open-source AI development with accessible data and research contributions

Links

• Home: https://pleias.fr