Telco giants form JV to push telco-specific LLM plan

• The Global Telco AI Alliance, which was formed last year, takes its next steps
• The four founding members are joined by SoftBank as they continue their work to build telco-specific large language models (LLMs)
• The alliance members will now form a joint venture 
• The main focus for the alliance members now is to gather enough data to optimise the LLMs

BARCELONA – #MWC24 – The founding members of the Global Telco AI Alliance, along with Japanese telco SoftBank, have unveiled plans during an SK Telecom-hosted press event here in Barcelona to establish a joint venture (JV) for the development of telco-specific large language models (LLMs).

SK Telecom (SKT), Deutsche Telekom (DT), e& and Singtel originally formed the alliance in July 2023, with plans for a generative AI LLM unveiled by SKT and DT in October 2023, and now SoftBank has joined the fold as the initiative moves to its next stage by forming a JV. Each of the partners will invest equally in the venture, though financial details have not been revealed.   

The quintet’s plan is to design LLMs that can help telcos enhance their customer interactions by using digital assistants and chatbots. The models will initially be optimised for the Korean, English, German, Arabic and Japanese languages, while there are plans to encompass additional languages further down the line.

The joint venture will aim to deploy “innovative AI applications tailored to the needs of the Global AI Telco Alliance members in their respective markets, enabling them to reach a global customer base of approximately 1.3 billion across 50 countries,” according to SKT.

Singtel is the largest player by customer base, with 770 million subscribers in 21 countries including Australia, India and Indonesia. Deutsche Telekom has around 250 million subscribers across 12 countries, with its biggest markets being Germany and the US, while e& has 169 million subscribers in 16 countries across the Middle East, Asia and Africa.

Currently, the alliance is focused on optimising the LLMs – a process that requires access to customer service data to “fine-tune the model for telco-specific questions”.

The reasoning behind this, according to the companies, is because “tariff and contract models and information on special hardware,” such as how a reset of the router is carried out, are “rarely found in the general training data of the large models. But it’s exactly this content that a telco bot needs to know”, so that it is able to understand, summarise and respond to these specific concerns.

Such targeted training will ensure that the LLM “understands the unique language and needs of telecom operators, paving the way for enhanced, personalised, and efficient customer experiences”.

Eric Davis (pictured, far right), VP of Language Superintelligence Labs at SKT, told TelecomTV that each of the alliance members will play an equal role in the developments and noted that the alliance works as an ecosystem, with the telco members working “closely” with the likes of Microsoft, Amazon, Google, OpenAI, Anthropic and other big tech companies.

The joint venture is expected to be established later this year, with Davis expecting it to be by the end of June: The timeline depends on obtaining data from SKT’s telco partners, as well as engineering and modelling work that is in motion. He added that the company is working actively with partners, such as Anthropic, to “tune them all together”.

The South Korean telco also needs to work with the cloud providers to allow people to have access to the models. 

While some of the members, such as DT, have a larger subscribers base, Davis noted that the alliance partners wanted to have “equal skin in the game”. He pointed out that the main challenge for the alliance is getting “buy-in” from other telcos, as they are very conservative by nature. “So, actually saying that this works and showing them evidence is very, very important,” he emphasised.

With five different telcos in the alliance, another challenge is to align the different views so that everyone is on the same page.

Asked about ensuring the privacy and security of customer data, Davis explained that SKT has “a very complicated process where we anonymise the data,” whereby names are replaced with other names that look like real names but are not. “Then we store the data in a very secure SDV [synthetic data vault] bucket or Azure cloud where only we have access or our select partners have access to it,” he added.

Suk Geun Chung, chief AI global officer at SK Telecom (pictured second from the right), highlighted that the joint venture aims to create opportunities for new revenue streams.

- Yanitsa Boyadzhieva, Deputy Editor, TelecomTV