Telco Giants Launch Open AI Initiative

The need for specialized AI in telecommunications is driven by the fact that general-purpose models often fail to grasp the industry's specific complexities. Only 16% of generative AI deployments in the telecom sector have been focused on network operations, a significant gap considering it's the industry's largest cost center. These generic models can struggle with interpreting 3GPP standards, understanding network data, and automating operations with the necessary accuracy. To address this, the Open Telco AI initiative provides a platform for open collaboration on models, data, and tools. AT&T's contribution includes a family of open telco models that are hardware and cloud-agnostic, trained on publicly available data. Other available models include RFGPT from Khalifa University and a Large Telco Model (LTM) from AdaptKey AI. The initiative's progress and the performance of these specialized models are tracked through the Telco Capability Index. This index utilizes a public leaderboard to assess models against a set of seven telecom-specific benchmarks. These benchmarks evaluate performance on critical tasks such as network troubleshooting and translating intent into network configurations. Accessible compute power is a core component of the initiative, with AMD providing GPU platforms for model training, fine-tuning, and evaluation. Cloud partner TensorWave offers the necessary infrastructure to scale these demanding AI workloads. This access to high-performance computing lowers the barrier to entry for developers and researchers. The collaborative effort extends to data, with a library of knowledge graphs, fine-tuning datasets, and curated materials from the GSMA. The initiative also fosters a community through challenges and resources aimed at solving real-world telecom problems. Over 46 organizations from more than 35 countries are already contributing to this shared ecosystem.