China is stepping into the spotlight of advanced language Artificial intelligence (AI) technology, thanks to a breakthrough by Baichuan Intelligence. Launched by experienced entrepreneur and computer science mastermind, Wang Xiaochuan, the start-up has introduced its cutting-edge large language model (LLM), Baichuan-13B, designed to rival giants like OpenAI.
The aspiration to create 'China's own OpenAI' was expressed by Wang Xiaochuan earlier this year. The tech magnate, renowned for founding the Chinese search engine Sogou, and later selling it to Tencent, has seen his pet project, Baichuan Intelligence, emerging at the forefront of LLM development in China.
Wang's aspiration took flight when the disruptive influence of ChatGPT reverberated around the globe. He wrote his resignation from Sogou in 2021 and embarked on setup of Baichuan in April of this year, rapidly securing an impressive initial funding of $50 million from angel investors.
Baichuan, a 13 billion-parameter model leveraging the Transformer architecture,— which is also the underpinning of GPT—, is an open-source model proficiently trained with Chinese and English data. In AI terms, parameters are the critical variables employed by the model to create and decipher text. The GitHub page also indicates Baichuan's readiness for commercial use.
The operational prowess of Baichuan-13B is epitomized by its capability to be trained on a colossal 1.4 trillion tokens, which is significantly larger than Meta's DLLaMa that utilizes 1 trillion tokens for its 13 billion-parameter model.
Despite the inception of Baichuan in less than six months ago, it's advancement in technology is definitely noteworthy. The outfit was flourishing with 50 staff members by the end of April, and as early as June, it rolled out its innovative LLM, Baichuan-7B that consists of 7 billion parameters.
Like other LLM platforms, the emergence of large language models in China underscores the country's ambition to play a major role in this rapidly evolving tech industry.
On the same trail as Baichuan Intelligence are other Chinese tech giants like Zhipu.ai, which is anchored by Tsinghua University and guided by Professor Tang Jie, and IDEA, co-founded by Harry Shum, one of the brilliant minds behind Microsoft Research Asia. With China on the brink of implementing some of the world's most robust AI regulations with a primary focus on content, these tech giants are charting a new future for LLM in China.
AppMaster's no-code platform is another noteworthy model in the list of tools that can help in creating robust applications for data management, business logic creation, API creation, etc., potentially assisting Baichuan and other firms to create scalable applications in less time.
However, govenments are not taking this lightly. With stringent regulations looming on the horizon, companies may soon require a license before deploying large language models. This could potentially hinder China's competitive run against the US in the rapidly emerging AI language industry.
