What big Korean telecom's investment in Anthropic AI means
The telecom giant revealed that it will invest $100 million in the AI research vendor. The vendor will build a telecom-specific LLM to support the Global Telco AI Alliance.
As the generative AI market matures, foundation model providers continue to receive financial support from global investors.
Earlier this week, South Korea-based telecommunications giant SK Telecom revealed it would invest $100 million in AI safety and research vendor Anthropic. The investment comes three months after the AI startup raised $450 million in Series C funding led by Spark Capital. It also comes a month after software multinational SAP invested in Anthropic and Anthropic competitor Cohere.
Anthropic AI strategy
The investors' interest in Anthropic and others is not surprising, especially since Anthropic's LLM strategy involves more of an enterprise perspective than OpenAI's, according to Futurum analyst Mark Beccue.
"Their approach is to be more enterprise-grade, so I think we might be seeing more investments in different LLMs," he said.
Like OpenAI with ChatGPT, Anthropic boasts a generative AI system, which it named Claude. The latest version, Claude 2, was introduced last month.
But unlike OpenAI, Anthropic and Cohere use a "bring your own data" approach, which is valuable to enterprises because it can help them avoid using public data or experience leaks of their own data.
The bring-your-own approach means that while the vendor has its own LLM model, enterprises can use their own proprietary data in their applications of the model.
Anthropic also employs an enterprise approach, avoiding some of the problems enterprises face with LLM hallucinations, Beccue said.
In May, Anthropic introduced research on what it calls constitutional AI. Constitutional AI is a method the AI startup uses to help its LLM determine which queries it will answer and which it won't.
Constitutional AI uses AI feedback to evaluate the outputs of LLMs. The system then uses a set of principles -- analogous to how countries' constitutions set forth basic values and principles for governing -- to assess its results.
This focus on accuracy is also appealing to enterprises and others using AI systems because they don't have to worry that the system will regurgitate wrong information, Beccue said.
Meanwhile, one interesting aspect of the investment from SKT is that the telecom service provider giant is also entering the generative AI market.
SKT and AI
Anthropic is expected to build an LLM specific to telecom service providers as part of the SKT investment, which was revealed on Aug. 13. The LLM will help the Global Telco AI Alliance formed by SKT, Deutsche Telekom, e& and Singtel.
We might be seeing more investments in different LLMs.
Mark BeccueAnalyst, Futurum Group
The international alliance was formed last month to create new customer service experiences in telecommunications applications using AI. But the group could face some hurdles.
"When you look at alliances at [telecom service providers], they don't work very well," Beccue said. "The ones that do tend to circulate standards … to do these other ideas where you're just going to build these LLMs that everyone can use and maybe scale to that. I don't think it's going to work."
It may not work because bigger telecom service provider players, such as AT&T, Verizon and CenturyLink, are probably already thinking about AI and don't need to scale. Smaller companies are the ones that need to scale, Beccue added.
Moreover, SKT needs to clarify who will sell the LLM that Anthropic will build.
"I'm wondering if that's tied to the $100 million investment," Beccue said. "It's a positioning that I think kind of sets up [the alliance] to sell to non-Tier 1 players to try to be a resource."
Esther Ajao is a news writer covering artificial intelligence software and systems.
