Journal

Print PDF

In my experience, Asian students are not used to asking questions. This is likely because many are never encouraged or even allowed to ask questions in a variety of settings, including classrooms. On the other hand, Western students likely ask more questions. But honestly, not all questions are helpful. Perhaps this book sheds some light on how to ask good questions, or what good questions are.

A curious note is that the word for question 問題 also means problem in modern Chinese. So a “person with questions” also means a “person with problems.” To a lesser extent, such usage is also found in English, e.g., a “questionable person.” But conflating questions with problems may be one of the reasons why Asians do not like to openly ask questions, which are perceived to be problematic and confrontational.

This conflation did not seem to exist in classical Chinese. I wonder if it is a transfer from the Sino-Japanese “mondai” 問題 in the early 20th century.

https://book.douban.com/subject/37154260/

Print PDF

Did DeepSeek plagiarise ChatGPT?

Certain media outlets give the impression of an accusation: DeepSeek, like certain Chinese goods, is a cheap imitation that violates intellectual property rights, threatening US economy and security.

But such reports are in fact missing the point and are misleading. Of course DeepSeek learned from ChatGPT. In fact, as a new product, it learned from many other older models. The point is that it is doing better and is more efficient, with a different training approach.

What is quirky about DeepSeek is its Chinese censorship. As many have observed, DeepSeek would churn out replies and that one can see that sensitive answers get deleted or redacted in real time!

There is in fact no hard evidence for plagiarism since the data source for all LLMs ultimately belong to everyone on the internet. But plagiarism of codes does happen. The case of Llama3-V, developed by a small team of Stanford students, stealing the codes of Tsinghua’s MiniCPM in June 2024 surprisingly went unnoticed and unreported in Western media. The creators Aksh Garg and Siddharth Sharma, shocked the open-source community with their Llama3-V, a powerful multimodal AI, with an even more shocking budget: $500! Garg and Sharma initially denied the allegation, but at the end apologised publicly. How were they busted? Their AI model was able to analyse ancient Chinese bamboo slip texts, a function that was uniquely developed by the Chinese team, producing the same answers and even mistakes. They blamed it on a certain Mustafa who wrote the codes and now disappeared. Christopher Manning, Director of StanfordAILab tweeted on X condemning the plagiarism, sort of downplayed the case as something “seems done by a few undergrads” and that he knows “nothing”…

DeepSeek is just a private Chinese AI company and one of the many. Somehow the media portrayed it as if it were a “Chinese” scheme with some sinister motives.

Coming to think about it. Printing, compass, gunpowder were all first invented in China. As Needham pointed out, Europeans never realised that and at least since Francis Bacon’s Novum organum 1620, most people thought that they were invented in the West. The Protestant missionaries (and possibly the Jesuits also) taught this to the Chinese as late as the nineteenth century and some even believed it (as the case of Shaou Tĭh in my earlier post). Strangely, Chinese never accused Europeans for stealing their technology. Believe it or not, Chinese who know history should be glad to see how ideas developed by their ancestors spread and had contributed to the progress of human societies and civilisation in general.