Advertisement
UK markets open in 7 hours 8 minutes
  • NIKKEI 225

    39,341.54
    -325.53 (-0.82%)
     
  • HANG SENG

    17,716.47
    -373.46 (-2.06%)
     
  • CRUDE OIL

    81.89
    +0.15 (+0.18%)
     
  • GOLD FUTURES

    2,338.20
    +1.60 (+0.07%)
     
  • DOW

    39,164.06
    +36.26 (+0.09%)
     
  • Bitcoin GBP

    48,688.35
    +607.22 (+1.26%)
     
  • CMC Crypto 200

    1,283.73
    +17.59 (+1.39%)
     
  • NASDAQ Composite

    17,858.68
    +53.53 (+0.30%)
     
  • UK FTSE All Share

    4,460.27
    -20.39 (-0.46%)
     

Shanghai AI start-up founded by ex-Microsoft engineers bets on 'scaling law' to boost AI capabilities

A Shanghai-based artificial intelligence (AI) start-up, founded a year ago by a former vice-president at Microsoft, is betting on so-called "scaling law" to enhance its AI capabilities, despite the difficulties in obtaining advanced chips, an executive said.

China's demand for higher computing power will be never ending amid the generative AI boom, Zhu Yibo, head of systems at AI start-up StepFun, or Jieyue Xingchen in Chinese, said at a media briefing in Shanghai on Thursday. "Computing power, systems, data, and algorithms are the cores in the pursuit of the scaling law," he added.

StepFun, founded by Microsoft Asia Research Institute's former chief scientist Jiang Daxin in 2023, is one of the many Chinese AI start-ups trying to catch up with US peers in AI computing power. It has launched the Step-1V multimodal large language model (LLM) with over 100 billion parameters, and it is testing the Step-2V model that boasts over one trillion parameters.

Do you have questions about the biggest topics and trends from around the world? Get the answers with SCMP Knowledge, our new platform of curated content with explainers, FAQs, analyses and infographics brought to you by our award-winning team.

ADVERTISEMENT

The push for greater computing power stems from a belief in the scaling law, which states that as the size of a model and its training data increases, the model's performance improves. However, Chinese AI start-ups face the common challenge of restricted access to the most advanced AI chips from US supplier Nvidia, making their quest more difficult.

China's domestic market has spawned at least 200 LLMs, many claiming to be among the top-performing in the world. Photo: Shutterstock Images alt=China's domestic market has spawned at least 200 LLMs, many claiming to be among the top-performing in the world. Photo: Shutterstock Images>

StepFun's computing centre under development in Shanghai is expected to be one of China's leading AI facilities, according to the company. When asked about the impact of Washington's export restrictions on AI chips to China, Zhu said the challenges were "manageable", without elaborating.

The company has become one of the most recognised AI players in China, partly thanks to the experience of its founder. Jiang worked at Microsoft for 16 years, leading projects such as the Bing search engine, the intelligent voice assistant Cortana, Azure cognitive services, and natural language understanding systems for Microsoft 365.

The other founders, Zhu and Jiao Binxing, shared working experiences at Microsoft.

China's domestic market has spawned at least 200 LLMs, the technology behind generative AI services like ChatGPT. Many of these claim to be among the top-performing in the world, including offerings from unicorns Zhipu AI, Baichuan, Moonshot AI and Minimax. However, the path to commercialisation of these products remains unclear.

Apart from its two LLMs, StepFun has launched two AI applications: a ChatGPT-like personal assistant called Yuewen, and Maopaoya, a character AI product with gaming features. According to an interview with Jiang in local media outlet 36kr, StepFun will focus on consumer-facing applications.

This article originally appeared in the South China Morning Post (SCMP), the most authoritative voice reporting on China and Asia for more than a century. For more SCMP stories, please explore the SCMP app or visit the SCMP's Facebook and Twitter pages. Copyright © 2024 South China Morning Post Publishers Ltd. All rights reserved.

Copyright (c) 2024. South China Morning Post Publishers Ltd. All rights reserved.