China’s AI Models Struggle to Bring Transformative Use, Executive Says

00:00

00:00/00:00

Listen to this article 1x

Baidu CEO Li Yanhong gives a keynote speech at the 2024 World Artificial Intelligence Conference in Shanghai on Thursday. Photo: VCG

Industry executives attending the World AI Conference in Shanghai expressed concerns about recurring technological issues in the development of large language models (LLMs) and offered possible solutions, as demand for more reliable artificial intelligence (AI) models designed for sector-specific software grows stronger.

The issues also come as China’s tech titans have kicked off a new round of price cuts for their AI models, extending a fierce price war in an effort not to lead the industry but to gain more users for their cloud businesses — a key resource for AI developers.

You've accessed an article available only to subscribers

Subscribe today for just $.99.

VIEW OPTIONS

Unlock exclusive discounts with a Caixin group subscription — ideal for teams and organizations.

Subscribe to both Caixin Global and The Wall Street Journal — for the price of one.

Share this article

Open WeChat and scan the QR code

DIGEST HUB

Digest Hub

Back

Explore the story in 30 seconds

Executives at the World AI Conference in Shanghai raised concerns about large language models' technical issues and proposed solutions.
Chinese tech companies are engaging in a price war for AI models to attract cloud business users, with significant cuts from firms like Baidu and ByteDance.
Chinese AI models have higher error rates than OpenAI's ChatGPT-4, and solutions include embedding knowledge graphs to reduce inaccuracies.

AI generated, for reference only

Who’s Who

SenseTime Group Inc.: SenseTime Group Inc., led by CEO Xu Li, is a Chinese facial recognition and AI specialist. At the World AI Conference in Shanghai, Xu Li stated that current AI models in China mainly function as data memory devices and struggle to support vertical applications. These applications are designed for specific sectors like banking or healthcare, and the existing limitations impede sector-wide transformative changes.

MiniMax: At the World AI Conference in Shanghai, Yan Junjie, CEO of Chinese AI startup MiniMax, highlighted that homegrown large language models (LLMs) have an average error rate exceeding 60%, nearly double that of OpenAI’s ChatGPT-4. Yan emphasized the need to reduce this error rate to around 3% to meet user expectations for more accurate AI-generated content.

Ant Group Co. Ltd.: Ant Group Co. Ltd.'s Chief Technology Officer, He Zhengyu, suggested embedding large language models (LLMs) with knowledge graphs to address LLM hallucinations. This approach grounds the models by allowing them to reference accurate data, which is especially crucial for complex industrial applications like finance and healthcare.

Tencent Holdings Ltd.: Tencent Holdings Ltd.'s head of intelligence, Wu Yunsheng, stated at the World AI Conference that the company's multimodal AI models have rapidly improved by continually adding diverse data and computing resources during research. This highlights Tencent's commitment to advancing its AI capabilities. Additionally, the company is actively participating in the ongoing price war within the AI industry, which aims to expand its cloud business rather than merely competing for users.

Baidu Inc.: Baidu Inc. is one of the first Chinese tech firms to launch a local alternative to ChatGPT. During the World AI Conference, Baidu announced significant price cuts for its LLMs, Ernie 4.0 and Ernie 3.5. A company source indicated that the ongoing price war in the AI sector is expected to continue for at least another year, targeting firms with strong cloud infrastructure and a large user base.

ByteDance Ltd.: ByteDance Ltd., the owner of TikTok, initiated a steep price cut in May for its large language model (LLM) Doubao, setting off a price war among Chinese AI companies. This move prompted competitors like Alibaba, Baidu, and Tencent to reduce their LLM prices as well. The main aim of this price reduction is to expand their cloud businesses rather than simply attract users away from rivals.

Alibaba Group Holding Ltd.: Alibaba Group Holding Ltd. responded to ByteDance’s pricing strategy by slashing prices of their large language models. This move was part of a broader price war among Chinese tech firms aiming to expand their cloud businesses. The primary aim of these price cuts is not to capture competitors' users, but to enhance their own cloud infrastructure and user base.

AI generated, for reference only

What Happened When

May 2024：: ByteDance Ltd. announced a new price level for its LLM Doubao, initiating a price war in the Chinese AI industry.

Thursday, July 4, 2024：: Xu Li, CEO of SenseTime Group Inc., gave a keynote speech highlighting the limitations of existing AI models available in China during the opening day of the World AI Conference.

Thursday, July 4, 2024：: Yan Junjie, CEO of MiniMax, discussed the high error rate of homegrown LLMs compared to OpenAI’s ChatGPT-4 at the World AI Conference.