LG CNS, in collaboration with AI startup Cohere, has developed a large language model (LLM) optimized for inference.

It features over 111 billion parameters and supports 23 languages, including Korean, English, Chinese, Hebrew, and Persian.

This announcement follows the release of a smaller Korean-specialized model with 7 billion parameters in May.

The new LLM is designed to process sensitive data securely within clients’ infrastructure, particularly in finance and public services.

It employs model compression technology, enabling it to operate on just two GPUs, compared to the usual four needed for models of this size.

LG CNS used its AI expertise alongside Cohere’s enterprise-focused Command LLM to develop this system.

The companies plan to continue working together to enhance the model’s capabilities further.

🔗 Source: The Korea Times

🧠 Food for thought

1️⃣ Sovereign AI movement gains commercial momentum amid geopolitical tensions

LG CNS’s inference-optimized LLM represents a growing trend of nations and companies developing independent AI capabilities to reduce reliance on dominant foreign models.

This development aligns with a broader global movement where countries like India, Japan, Denmark, and Italy are investing billions in domestic AI infrastructure to ensure technological independence 1.

The emphasis on processing sensitive data within clients’ own infrastructure directly addresses one of sovereign AI’s key drivers: data privacy concerns in sectors like finance and government 2.

By enabling on-premises deployment with reduced hardware requirements (two GPUs instead of four), LG CNS is lowering the significant infrastructure barriers that typically challenge sovereign AI initiatives 3.

This approach helps organizations maintain control over their data while complying with increasingly strict regulations like the EU AI Act, which emphasizes transparency and accountability 4.

2️⃣ Cultural and linguistic diversity emerging as competitive differentiator in AI development