Getting your Trinity Audio player ready... |
Chinese e-commerce and technology company Alibaba (NASDAQ: BABAF) has announced the release of a new artificial intelligence (AI) model with advanced capabilities to become the undisputed top dog in Mainland China.
Dubbed Qwen2, the large language model (LLM) is the latest iteration of Alibaba’s Tongyi Qianwen model series, evolving from Qwen 1.5 to achieve new milestones in under a year. According to the company’s official profile on Hugging Face, Alibaba is adopting an open-source approach to trigger adoption from global developers for the LLM.
Apart from leaning on an open-source strategy, Qwen2 has several capabilities that the firm believes will attract a raft of enterprise users, including advanced training methods and new use cases.
For starters, the model features support for 27 languages outside English and Chinese, a significant leap from Qwen 1. The latest version covers languages from Southern Asia, Eastern Asia, the Middle East and Europe, with Africa and Latin America noticeably absent from the list.
Qwen2 particularly excels in coding and mathematics, with Alibaba integrating code experience from Qwen 1.5 and infusing new high-quality datasets. Furthermore, early tests confirm proficiency in long-context understanding following training on 32k-length contexts.
“Compared with the state-of-the-art open source language models, including the previously released Qwen 1.5, Qwen2 has generally surpassed most open source models and demonstrated competitiveness against proprietary models across a series of benchmarks targeting for language generation, multilingual capability, coding, mathematics, and reasoning,” read the statement.
Stress tests to ascertain the safety levels of the LLM reveal watertight internal systems to protect users from bad actors at a level “comparable” to GPT-4.
Qwen2 is able to achieve its proficiency across several verticals, thanks to its focus on pretraining designed to bring the model’s intelligence “closer to human.” In the future, the company says it will focus on introducing multimodal models that can receive vision and audio inputs.
“We are training larger Qwen2 models to further explore model scaling along with our recent data scaling,” read the statement. “In the near future, we will continue to opensource new models to accelerate open-source AI.”
All-in for AI
Since 2023, Alibaba has been steadily pivoting toward AI and other emerging technologies to avoid playing catchup with its global peers. Company CEO Eddie Wu confirmed that Alibaba will reinvent itself into an AI-first company in the coming months, hinting at significant infrastructure investment.
“As traditional internet models become increasingly homogenous and face the competitive pressure of saturation, new technologies such as AI are emerging as the new engine of global business growth,” said Wu. “We will calibrate our operations around these two core strategies and reshape our business priorities.”
In late 2023, the company shut down its quantum computing arm to double down on AI, rolling out an AI-based video generation tool and a course to support global talent development for the budding ecosystem.
In order for artificial intelligence (AI) to work right within the law and thrive in the face of growing challenges, it needs to integrate an enterprise blockchain system that ensures data input quality and ownership—allowing it to keep data safe while also guaranteeing the immutability of data. Check out CoinGeek’s coverage on this emerging tech to learn more why Enterprise blockchain will be the backbone of AI.
Watch: Foundational tech blockchain & AI can reinforce each other