Episode 2260 - July 8 - Tiếng Anh - Bản Tin Công nghệ Thông tin – Ngày 7 tháng 7 - Vina Technology at AI time - Lê Quang Văn | Podcast - Nhac.vn

I.T. News – July 7, 2024.

1 - China’s AI competition deepens as SenseTime, Alibaba claim progress at AI show

South China Morning Post. Jul 6, 2024.

Chinese large language model (LLM) developers from SenseTime and Alibaba Group Holding have rushed to declare fresh progress in developing their artificial intelligence (AI) technology at the country's largest annual industry conference, in sign of deepening competition in the country's largely closed-off market.

At the World Artificial Intelligence Conference (WAIC) in Shanghai, Chinese AI champion SenseTime on Friday released a series of updated versions of its SenseNova LLMs, including SenseNova 5.5, its latest foundational model, which the company claims has 30 per cent improved performance compared with the previous version released in April.

"I believe the key to success and differentiation in China's road to AI is how to construct high-level thinking logic based on synthetic data in vertical industries," SenseTime co-founder and CEO Xu Li said at the conference.

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.

Xu claimed that SenseNova 5.5 has surpassed GPT-4o in five of eight key metrics, citing data from OpenCompass, a platform for benchmarking large models. At the same time, its stock price plunged 16 per cent in Hong Kong on Friday.

SenseTime is among a number of Chinese Big Tech companies that have sought to use WAIC as an opportunity to boast about progress made in developing LLMs amid heightened competition in the domestic market.

Alibaba's cloud computing unit, the developer of the Tongyi Qianwen LLMs, also touted new user growth. Alibaba owns the South China Morning Post.

In the past two months, downloads of the open-source Tongyi Qianwen models doubled to more than 20 million.

The number of customers served by Alibaba Cloud Model Studio, the company's generative AI development platform designed for corporate customers and partners, also increased by over 150 per cent to 230,000, according to Zhou Jingren, Alibaba Cloud's chief technology officer.

Zhou asserted Alibaba Cloud's commitment to open-source initiatives.

"Two years ago, we released the Tongyi model series at WAIC. At that time, we announced that Tongyi's core model would be open source," he said.

"Today, Tongyi Qianwen has achieved full-scale and full-modal open source in the true sense, narrowing the gap between open-source and closed-source models."

Yan Junjie, founder and CEO of Shanghai-based AI start-up MiniMax, made a bold prediction in an interview with local outlet Sina.com that despite red-hot competition in the global LLM market - China alone has more than 200 such models available - only five companies will be making these models in the future.

"How many of these will be in China, that's a question I don't know [the answer to]," Yan said.

MiniMax is one of China's so-called AI Tigers, alongside Zhipu AI, Baichuan, and Moonshot AI - all based in Beijing.

The group's unofficial name represents a belief that they are China's best hope in rivalling overseas start-ups like Microsoft-backed OpenAI and Anthropic, the Amazon.com-backed developer of Claude.

In his speech at WAIC, Yan said that there is an imperative for AI models to improve their accuracy before playing a bigger role in traditional industries.

SenseTime also launched at the Shanghai event its 5o and 5.5 Lite models. SenseNova 5o is a multimodal model, while 5.5 Lite is a terminal-based model.

Xu said SenseNova 5o is the first large AI model in China to realise a new means of human-AI interaction by integrating cross-modal information such as sound, text, images and video in real time.

In a live demonstration, a SenseTime staff member took a video of himself standing on stage and showed how the model was able to immediately identify that he is