The emergence of the International Trillion Parameter Consortium
This week, Barcelona becomes the epicenter of the Artificial Intelligence revolution in science with the European launch of the International Trillion Parameter Consortium (TPC). This global consortium of scientists, made up of more than 850 participants from more than 100 organizations around the world, has as its main objective to enhance the application of Artificial Intelligence in scientific research through supercomputing.
The TPC was formed with the idea of employing large-scale foundational models for scientific discovery. These foundational models are neural networks that have revolutionized deep learning, a method of Artificial Intelligence that processes data in a similar way to the human brain. Instead of developing AI from scratch, scientists use these foundational models as a starting point to develop new applications quickly and cost-effectively.
The impact of foundational models in science
Foundational models are significantly changing the deep learning lifecycle. Although developing a foundational model from scratch can be expensive, in the long run it is more efficient to use already trained models to develop new deep learning applications. This represents a challenge to scientific progress, as only a few organizations have the resources necessary to build large-scale foundational models.
To overcome this limitation, TPC fosters collaboration between different institutions and research teams. They share strategies, architectures, and high-quality datasets to jointly build state-of-the-art large language models (LLMs). These language models are deep learning algorithms capable of performing a variety of natural language processing tasks, such as recognizing, translating, predicting, or generating text.
Challenges and opportunities for AI in science
The TPC faces several challenges in its goal to advance AI in science. These challenges include developing scalable model architectures, training strategies and organizing scientific data, and optimizing AI libraries for exascale computing platforms. However, there are also important opportunities to accelerate scientific advances in AI, especially through the predictive power of large language models.
The European launch of the TPC in Barcelona brings together industry leaders, researchers and leading professionals in the field of AI and science. During the event, the transformative potential of Generative AI in scientific-technical applications will be discussed, as well as the participation of the European research communities in AI, high-performance computing and disciplinary sciences.
The future of AI in science
The TPC is in a formation phase and its objective is to continue growing and supporting the professional development of AI experts. Additionally, it seeks to reduce the duplication of efforts necessary to accelerate scientific advances in AI. Large language models have become reference methods in many scientific domains and their adoption in various fields continues to increase.
The TPC represents an important milestone in the Artificial Intelligence revolution in science. Through collaboration and the use of supercomputing, this consortium is expected to drive scientific discoveries and technological advances that benefit society as a whole.
