In the rapidly evolving world of artificial intelligence, a seemingly insignificant online community—Baidu Tieba's "Weakest Heart" forum—has unexpectedly become a crucial source of data for AI training, garnering widespread attention from the tech industry and online communities. This community, brimming with absurd humor, has surprisingly demonstrated remarkable value in AI training, prompting us to ponder: What makes these seemingly "weak" statements the cradle of intelligence?
In April of this year, a joint study by the Chinese Academy of Sciences and the University of Waterloo, among other institutions, revealed a shocking result. In eight tests including question-answering, brainstorming, categorization, generation, and summarization, the "Weakest Heart" forum outperformed well-known platforms like Wikipedia, Zhihu, Douban, and Xiaohongshu, becoming one of the most popular Chinese AI training databases. This discovery challenges traditional perceptions of high-quality data.
At the recent Bund Conference, core members of the "Weakest Heart" forum made their first public appearance, challenging AI and unveiling the true nature of this unique community. Zeng Xiaodong, CEO of Wujie Fangzhou, explained the reasons for choosing "Weakest Heart" as training material: To make AI more human-like, colloquial, multi-round dialogue materials are needed, and "Weakest Heart" meets this demand perfectly.
Core member of "Weakest Heart," Hu Luobei, shared his interesting experiences with AI. As early as 2022, he tried to have AI interpret some jokes, only to find that while AI could search for related information, it couldn't grasp the true meaning of the jokes. This highlights the limitations of AI in understanding human humor.
However, behind these seemingly absurd jokes lies profound logic. For example, the phrase "Knowing there are tigers in the mountain, not going to the knowing mountain" cleverly disassembles and reassembles the word "knowing," creating new meanings. It is these linguistic traps that train AI's understanding and reasoning abilities in Chinese, enabling machines to communicate more like humans.
The charm of "Weakest Heart" lies in its status as the "basic science" of jokes. The community members are mostly from a science background, and their jokes are not only logically rigorous but also infused with rich rhetoric and life observations. This unique creative approach provides AI with valuable learning materials.
Interestingly, the existence of "Weakest Heart" seems to serve as a defense line between humans and AI. As Hu Luobei said, "No AI can leave the Weakest Heart with a smile, because they simply don't understand humor." This deep-level language understanding and sense of humor become the key differentiators between human intelligence and artificial intelligence.
Although we live in an era surrounded by AI, the existence of "Weakest Heart" reminds us that human creativity and sense of humor are still unique. This seemingly absurd community not only offers a unique perspective for AI training but also serves as a microcosm of human wisdom and creativity.