AI models from Tencent and Alibaba demonstrate a better understanding of the Chinese Language than humans on the CLUE benchmark

Tencent Holdings and Alibaba own AI models that have successfully passed a benchmark test CLUE. These test results indicate that these AI models can understand the Chinese language better than humans. However, researchers claim that AI still has to go through many advancements before truly understanding languages.

Until recently, no AI models have had significant scores in the Chinese Language Understanding Evaluation (CLUE) Benchmark, which gives a set of tasks to evaluate how well the models or machines can understand and respond to the Chinese language. But, the AI models of Tencent and Alibaba recently achieved all-time high scores in CLUE.

Chinese Language Understanding Evaluation is referred to as CLUE. It is made up of many NLU datasets. It is a community-driven project that combines nine challenges on original Chinese text, including machine reading comprehension and many well-known single-sentence categorization tasks. It has happened for the first time that these two AI models have outperformed humans on CLUE since academics set the benchmark three years ago.

According to 29th November 2022 rankings on CLUE’s website, the two tech titans’ models viz Tencent’s “Hunyuan AI model” ranked top with a score of 86.918, and second place was bagged by Alibaba’s “AliceMind” with a score of 86.685 respectively. Both rated higher than humans, who received an 86.678 point score, followed by Oppo (smartphone manufacturing company) and Meituan (food delivery firm).

In a recent piece, Alibaba Cloud, the cloud computing branch of the e-commerce behemoth, said, “Although the rankings fluctuate often, the top scores have never overtaken humans until today.” According to the most recent findings, “the Chinese language comprehension of the AI model has achieved a new level.”

There are also other advancements and research going on other than these topics. The notoriously difficult national college admission tests in China, known locally as gaokao, Du Xiaoxiao produced an essay that scored higher than the majority of students, according to a statement made earlier this year by Baidu, a Chinese internet search company. China’s Big Tech giants have enhanced the NLP technology that powers voice assistants, machine translation, and spam detection like AliGenie from Alibaba and Xiaowei from Tencent.

Some US-based researchers claimed that last year, many AIs outperformed humans on some comprehensive tasks but still could not identify whether the words and letters in a phrase had been arbitrarily rearranged. This is one of the advantages humans still have in understanding the language. According to research conducted by Cambridge University, the human brain can read words with jumbled letters if the first and last letters are the same. This ability of the human brain is yet to be acquired by these human models.

There are many elements in a language that these AI models are yet to conquer. The context and emotion of a phrase also matter, which these models are struggling with. Even though there is a long way to go to fully attain an understanding of these languages, AI models have covered a huge gap between them and humans. Emotions in a language is a huge challenge for AI models but affirmative, ongoing research give hope to conquer all those challenges.

Check out the Reference Article. All Credit For This Research Goes To Researchers on This Project. Also, don’t forget to join our Reddit page and discord channel, where we share the latest AI research news, cool AI projects, and more.

🐝 Join the Fastest Growing AI Research Newsletter Read by Researchers from Google + NVIDIA + Meta + Stanford + MIT + Microsoft and many others...