Newshot
Account

Korean Carrier LG U+ to Enter AI Token Business Next Year, Helping Enterprises Cut AI Transformation Costs

Source: IT之家 Published: AI 中文 AILGU+Token
AI summary

LG U+ will enter the AI Token business next year, using its self-developed token optimization technology to help enterprises lower AI transformation (AX) costs; the service combines "Harness Engineering" for designing agent task execution paths, "intelligent routing" that assigns models by question type, and a "chip combination" approach using heterogeneous semiconductors to optimize inference. LG U+ will cut unnecessary repeated search and verification during AI agent task execution and match the most suitable model to each task's difficulty, and with model lightweighting and inference technology its self-developed on-device AI model reaches over 95% of the performance of general cloud models at one-third the size, while having small models generate answers quickly and large models verify them more than doubles token generation speed. On infrastructure, it pairs 英伟达, AMD GPUs and Korean company FuriosaAI's neural processing unit (NPU) to separately handle question analysis (Prefill) and answer generation (Decode), lowering infrastructure costs; LG U+ will both offer services jointly with partners and build its own, fully entering the B2B AI Token market from next year, which will

Translation(model translation, not the source's own text)

Korean telecom operator LG U+ to enter AI Token business next year, helping enterprises lower AI transformation costs

IT Home, November 15: According to a report today by Korean media Etnews, Korean telecom operator LG U+ will enter the AI Token business next year, using self-developed Token optimization technology to help enterprises reduce AI transformation (AX) costs.

NewsWorks
NewsWorks

According to the introduction, this service has three technologies: "Harness Engineering" for designing Agent task execution paths, "intelligent routing" for allocating models according to the type of question, and "chip mix" for using heterogeneous semiconductors to optimize the inference process.

IT Home learned from the original report that LG U+ will reduce unnecessary repeated searches and verification when AI Agents execute tasks, and match the most suitable model according to task difficulty. At the same time, by combining model lightweighting and inference technology, the company's self-developed on-device AI model, while one-third smaller in size, still achieves over 95% of the performance of a general-purpose cloud model. Through technology that lets a small model quickly generate answers and then has a large model verify them, Token generation speed is more than doubled.

On infrastructure, LG U+ adopts a chip-mix strategy, combining 英伟达, AMD GPUs, and the neural processing unit (NPU) of Korean local company FuriosaAI to handle question analysis (Prefill) and answer generation (Decode) tasks respectively. By separately handling tasks that require large-scale computing and tasks with higher memory bandwidth requirements, it reduces infrastructure costs.

LG U+ will adopt both joint service provision with partners and independent construction, and will fully enter the B2B AI Token market starting next year. With LG U+'s entry, all three major Korean telecom operators will enter the Token economy.

Related articles

  • AMD joins hands with FuriosaAI and MangoBoost to build a heterogeneous AI computing reference cluster
  • Korean chip startup FuriosaAI challenges 英伟达: valuation quadruples to $2.9 billion after new funding round
  • Korean AI chip company FuriosaAI discloses specs of next-generation inference accelerator: 32x compute, 32x memory bandwidth
  • LG U+ launches home security subscription service "Homecam Lite" in Korean market, integrating surveillance cameras and insurance
  • $7,000 burned in a day! OpenAI researcher uses AI to write code like crazy
  • Korean ASIC company FuriosaAI wins order for 7,000 chips from Swedish data center

The translation is machine-generated and individual terms may be off. Nothing is rewritten or cut; the original is authoritative.

Published by IT之家
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗