Making powerful AI more accessible to everyone

University of Waterloo researchers have developed a new technique to make powerful AI models cheaper, faster, and more environmentally friendly to train. The method, called SubTrack++, significantly speeds up the pre-training and fine-tuning of large language models (LLMs), the most computationally expensive part of building AI systems, by up to 50 per cent while maintaining top-tier accuracy. By reducing the cost and energy required to train these models, the approach could help democratize AI, making advanced tools more accessible to individuals, smaller organizations, and researchers outside large corporations. The work also aligns with broader efforts in responsible and accessible AI research and will be presented at the NeurIPS 2025 conference.