SoftBank Corp has unveiled a large AI model for telcos that it says offers a 'generative AI foundation' for the industry. When adapted to basestation optimization it can cut task time from days to minutes.
SoftBank Corp has unveiled a large AI model for telcos that it says offers a "generative AI foundation" for the industry.
It says its Large Telecom Model (LTM) was built on its own network and operational data and enables "advanced inference in the design, management, and operation of cellular networks."
The company says it has developed specialized AI models by fine-tuning the LTM.
In particular, it configured it for basestation optimization and found it reduced task times "from days to minutes," achieving a similar level of accuracy as manual or partially automated approaches.
"These results demonstrate that by fine-tuning the LTM for specific use cases, it will become easier to develop dedicated AI models tailored to various operational scenarios in cellular networks," the company said.
The LTM combines inferencing capabilities to solve network issues with flexible responsiveness enabled by natural language processing, SoftBank said.
The company says it will continue developing the model and aims to implement it into its own operations.
But it did not say when it would make the model available to other telcos and if so on what terms. SoftBank is a founding member of the Global Telco AI Alliance, which aims to build LLMs, such as this one, tailored for operators.
New skill set
Ryuji Wakikawa, head of SoftBank's Research Institute of Advanced Technology, said the LTM transforms the process of designing, building and operating mobile networks.
"By fine-tuning the LTM, it's possible to build AI models specialized for various processes and deploy them as agents," he said.
A SoftBank white paper on AI, issued last month, said implementing telco AI models was similar to other AI models "but the scale and new skill set needed can be vastly different compared to the current telco practices."
According to GSMA, general LLMs such as Chat GPT score poorly against telecom industry benchmarks and often produce inaccurate or misleading recommendations. It has launched an industry initiative to improve the performance of LLMs in telco-specific applications.
SoftBank Corp's best-known other telecom AI initiative is its AITRAS AI-RAN solution, aimed at deploying to support edge scenarios.
SoftBank says it has just installed a trial version of AITRAS at NVIDIA's headquarters in Santa Clara. Using an LLM that runs on the network, researchers have built a "porter robot" that carries small items and follows visitors based on verbal instructions.
Nvidia is a partner of SoftBank and its parent in multiple AI projects, including AITRAS, along with the construction of Japan's most powerful AI supercomputer.
You May Also Like
