DTW24-Ignite: Global Telecom AI Alliance lays out vision for telco LLMs

The five telcos behind the Global Telecom AI Alliance (GTAA) – SK Telecom, Deutsche Telekom, e& Group, Singtel and Softbank – took the stage with their hyperscaler partners at DTW24-Ignite to spell out their business vision for telco-specific LLMs.

The GTAA members want to use LLMs to identify and solve telco-specific business issues using local languages, as well as industry-specific terms and their own internal business jargon –none of which can be done quickly or easily with generic LLMs.

“Simply adding engineering prompts with a generic AI LLM … it’s pretty painful. It requires a lot of tuning to get to consistent outcomes and accurate results, and especially with very complex use cases,” said William Woo, Group Chief Information Officer / Chief Digital Officer, Singtel.

Singtel finds that “a customized telco LLM definitely outperforms a general LLM for telco-related tasks,” according to Woo.

Another aim of the alliance is to ensure that communications service providers (CSPs) benefit from future AI business models built using telco data.

“We have talked [in the past] about the hyperscaler and the … destructive role they have played in telco infra,” said Harrison Lung, Group Chief Strategy Officer, e&.“In the world of AI, how do we ensure that the telcos have not just a seat at the table, but true ownership of the value pool that will be generated from it? We felt it was best to collaborate in building this telco LLM [in which] we each have our unique data set and assets.”

The GTAA’s founders announced at the event a joint venture agreement to co-develop and launch a multilingual Telco Large Language Model, which will work for Korean, English, German, Arabic, and Bahasa among other languages. Together they will make equal investments in the initial working capital required to develop it.

The JV will look at deploying innovative AI applications tailored to the needs of the founding parties in their respective markets, which cover a global customer base of approximately 1.3 billion across 50 countries.

Aiming for personalization

Customer care has been one of the first areas where telcos have used GenAI and is an initial focus for GTAA members.

“An industry-specific LLM equipped with comprehensive understanding of our diverse product range will help us present a personalized offer that aligns with the preferences of our customers,” said Singtel’s Woo.

But the CSPs are also setting their sights further.

For example, e& is examining how to “embed AI in all of our existing businesses and business models,” according to Lung.

“We are a telecom company; we’re also a technology company. How do you overlay these AI capabilities in those assets so that you have [for example] a better recommendation engine for the Netflix equivalent that we run?” Lung added. "How do you embed AI into Careem, which is a local super app … so the next time you open it, it is super tailored and customer to you.”

SK Telecom, meanwhile, believes the personalization of services enabled by GenAI will drive the development of third-party ecosystems, which then will feed into AI data center ambitions. (SKT is working on LLMs with Anthropic and invested $100 million in the startup last year).

“We expect to provide users with their more expensive, personalized experience while simultaneously increasing the impact within the … ecosystem by having various third parties join,” explained SG Chung, Chief AI Global Officer, SK Telecom. “On the enterprise side, we are expanding our infra business with AI data centers. We believe the significant number of end users passing through our … platform will drive demand for our AI data center infrastructure.”

Chung added: “SK Telecom is actively clustering the synergy between these two business areas. So, I’d like to emphasize that we can generate revenue through AI data center itself, and also SK Telecom [can run their] … AI businesses more efficiently by reducing operational cost.”

Enabling automation

Other telco-specific uses for GenAI include improving employee productivity, field operations and network operations.

Referring to TM Forum’s autonomous network levels, Sameer Vuyyuru, Director Global Business Development for Telecom, AWS, said: “We think generative AI can get us to Level 4 automation two to three years sooner … than using traditional ML [machine learning] methods. And so, we’re heavily focused on improving network operations and field operations.

In the meantime, AWS is working with BT to provide coding assistance to approximately 1,200 network engineers.

“So far, we’ve automated about 12% of the tedious, non-value-added tasks … and written over 100,000 [lines of] code in the first four months,” said Vuyyuru.

AWS foresees the industry developing domain-specific, language-specific and task-specific LLMs.

“The future is going to be model-to-model interaction as much as possible,” Vuyyuru said. “So, you’re going to have a model for network operations talking to customer operations. That is where we’re headed.”

And it’s not just telcos that are investing in tailored LLMs.

“We’re probably working with about 20 different sovereign governments right now to build specific LLMs for that country or region,” said Vuyyuru.

Both Microsoft and Google, which took part in the presentation, are looking forward to further telco GenAI development.

“As an industry, we still are only scratching the surface of what’s possible over here,” said Ankur Jain, VP Telecom, Distributed Cloud, Google Cloud.