ChatGPT has quickly become the reference when it comes to AI models, at least in the West. But as competition between large language models (LLMs) heats up, we're starting to discover that Open AI's model could have some serious competition in China, particularly a program known as ChatGLM, a chatbot that comes close to ChatGPT in some capabilities and surpasses it in Chineseaccording to its developers.
“Basically, ChatGLM is an alternative to ChatGPT,” explains Jie Tang, a computer scientist at Tsinghua University in Beijing, during a talk presenting ChatGLM's capabilities at the International Conference on Learning Representations (ICLR 2024). ChatGPT, like most similar models, can respond in a variety of languages, but they are created by American companies and use English as the primary language. In contrast, ChatGLM is bilingual and designed to work in Chinese and English.
Tsinghua University and its spin-off company Zhipu AI, valued at more than 2 billion dollars, according to Tang, developed ChatGLM and the underlying model GLM, which stands for General Language Model. More than 700 researchers and engineers from Zhipu AI and about 100 students from Tsinghua University are working on AI language models.
One reason to focus on China's advancement is that ChatGPT is not available in this country. But that is not the only reason to build local alternatives. China-oriented LLMs produce results that better reflect the needs and preferences of the country's inhabitantssays Tang, including, for example, country-specific financial or educational information.
This expert compares his program to training a language model in a Chinese social media app instead of a Western one. “Basically, WeChat knows more about the people of China than Snapchat – adds Tang -. Models that adapt to different languages avoid oversimplifying or neglecting the specific characteristics of certain languages and cultures.”
At ICLR 2024, Chat GLM developers shared data stating that the highest-performing version of the underlying ChatGLM model, GLM-4, is is within 90% of the scores achieved by GPT-4 of OpenAI on various benchmarks. These include tests of general knowledge, common sense and mathematics. ChatGLM also outperforms GPT-4 in a Chinese LLM optimization benchmark.
To this we must add that dozens of other LLMs are being developed in China: More than 100 AI language models were made public in 2023 alone. Technology giants like Baidu or Alibaba have their own artificial intelligence chatbots, for example.
Giving AI systems human-level capabilities when it comes to a wide range of tasks – a milestone known as artificial general intelligence, or AGI – will require them to be embedded in the world. Could ChatGLM to be the first artificial intelligence system to achieve AGI? “I have no idea,” Tang concludes. “I hope we are first, but we are competing with everyone else.”
The next obvious question then is how close computer scientists are to developing an AGI model and whether achieving this breakthrough is desirable. That is one of the biggest dilemmas we face.