ALETHEIA is an ambitious research and innovation project focused on training large-scale foundation models that integrate cognitive AI principles with human behavior modeling. The project develops a novel AI architecture—comprising over 50 specialised cognitive modules organised into a biologically-inspired organ system—that aims to bridge the gap between current large language models and human-like reasoning capabilities.
The research approach combines advanced distributed training techniques across heterogeneous accelerator environments (TPUs and GPUs) with a unique training data pipeline that synthesises high-quality chain-of-thought reasoning, multi-domain knowledge, and cognitive behavioral patterns. The project addresses a critical innovation gap: while current LLMs excel at pattern matching, they lack the structured cognitive framework necessary for genuine understanding and reasoning.
ALETHEIA proposes a paradigm shift by embedding cognitive protocols—including metacognition, emotional valence processing, and adaptive learning—directly into the model architecture and training methodology. The requested compute allocation will enable training of 7B-parameter foundation models on curated datasets exceeding 500,000 diverse, high-quality training samples, with systematic evaluation against established benchmarks.
This work has significant implications for European AI sovereignty, advancing the state-of-the-art in ethical, transparent, and cognitively-grounded artificial intelligence systems.
Principal Investigator, Company and Country
Abdulmalek Saket, Royal Fenice kft, Hungary