CERN and Mistral have signed a statement of intent to establish a long-term partnership developing AI foundation models for scientific discovery, the organizations announced on October 15.
Their initial ambition is a pretrained Large Physics Model, trained on billions of simulated proton-proton collisions at the Large Hadron Collider.
The planned model would support several tasks, including reconstructing collisions, identifying particles, detecting unexpected signals and assessing future detector designs.
Today, collision reconstruction uses a sequence of specialized algorithms to turn detector signals into objects scientists can analyze. Machine learning already supports parts of that process. The proposed foundation model would explore learning across the reconstruction chain and serving multiple tasks within one system.
CERN would contribute physics expertise, scientific data and large-scale computing, while Mistral would bring model-training capabilities.
The announcement establishes an intended collaboration rather than a completed model. The release gives no budget or completion date.
Its scientific promise will depend on whether a shared model can meet the precision and validation requirements of the individual tasks it is designed to support.
Sources: https://home.cern/