AI race: Alibaba, Tencent quickly adopt Meta's new Llama 3.1 model amid excitement

Chinese technology giants Alibaba Group Holding and Tencent Holdings have rushed to offer the latest artificial intelligence (AI) model from Facebook owner Meta Platforms into their cloud services, as the debut of Llama 3.1 drew widespread attention this week.

Alibaba Cloud, the e-commerce giant's online computing platform, was among the first to include the latest open-source Llama family of large language models (LLMs) - the technology underpinning generative AI products such as ChatGPT - by integrating it into its Bailian model training platform, the company said on Tuesday in a post published to its official WeChat account. Alibaba owns the South China Morning Post.

Alibaba said it is offering one month of free compute resources that can be used for training and inferencing tasks with Llama 3.1, which launched on Monday.

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.

Meta Platforms CEO Mark Zuckerberg discusses his company's powerful new AI model, Llama 3.1, with Bloomberg. Photo: Bloomberg alt=Meta Platforms CEO Mark Zuckerberg discusses his company's powerful new AI model, Llama 3.1, with Bloomberg. Photo: Bloomberg>

Shenzhen-based video gaming giant Tencent quickly followed with its own announcement the same day, saying Llama 3.1 is now available on its cloud platform. It includes a number of fine-tuning and inferencing tweaks to ensure Meta's open source models are usable in a range of areas, from intelligent conversation, text generation and writing tasks, according to the company.

Meta is offering the latest version of Llama in three different sizes - 8B, 70B and 450B - that are named after the billions of parameters they include. The number of parameters generally corresponds with an LLM's level of sophistication.

In its unveiling, Meta hailed Llama 3.1 as "the first frontier-level open-source AI model". Early tests have compared it favourably with leading closed-source models such as OpenAI's GPT-4o.

Widely considered the best-performing open-source LLM family, Llama models have become a default foundation model that many Chinese start-ups use to build their own generative AI products, typically with custom tweaks to change or improve performance.

Chinese industry insiders have cautioned against an over-reliance on Meta's models among domestic generative AI initiatives. Last year, Chinese AI unicorn 01.AI, founded by venture capitalist Lee Kai-fu, generated some controversy when it was found to be building on top of Llama models without being transparent about it.

Responding to criticism that advanced open-source models benefit China, Meta CEO Mark Zuckerberg said in a statement announcing the new models on Monday that "decentralised and open innovation" advantages the US.

"Our adversaries are great at espionage, stealing models that fit on a thumb drive is relatively easy, and most tech companies are far from operating in a way that would make this more difficult," he said.

While open-source models have typically lagged behind the most advanced in the industry, Zuckerberg said he expects to close that gap soon. "Starting next year, we expect future Llama models to become the most advanced in the industry," he wrote in his statement posted to Facebook.

LMSYS, an AI model research organisation supported by the University of California, Berkeley, currently ranks closed-source models from OpenAI, Anthropic, and Google as the top as the best performing. OpenAI's GPT-4o ranks No 1, with the three companies' other models filling out most of the other top 10 spots.

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.