The manuscript likely does not represent China's second 'DeepSeek moment.'
Manus, a Chinese-origin artificial intelligence tool, has rapidly gained popularity. However, this technology has several limitations and errors.
Manus, an "agent" artificial intelligence platform that was previewed last week, has generated a lot of buzz, comparable to the excitement of a Taylor Swift concert. The product director of Hugging Face described Manus as "the most impressive AI tool I have ever tried." Meanwhile, AI policy researcher Dean Ball labeled it "the most sophisticated computer that uses AI." In just a few days, its official Discord server reached over 138,000 members, and it has been reported that invitation codes to use Manus are being sold for thousands of dollars on the Chinese app Xianyu. However, it is still unclear whether all this fervor is justified.
Despite the significant attention it has received, Manus was not developed entirely from scratch. The platform reportedly uses a combination of existing and fine-tuned artificial intelligence models, including Claude from Anthropic and Qwen from Alibaba, to perform tasks such as drafting research reports and analyzing financial documents. On its website, La Efecto Mariposa, the Chinese company behind Manus, provides striking examples of the platform's capabilities, ranging from buying properties to programming video games.
In a viral video on X, Yichao "Peak" Ji, the head of research for Manus, suggested that this platform is superior to other similar tools like OpenAI's deep research. According to Ji, Manus outperformed deep research on a popular benchmark for AI assistants known as GAIA, which evaluates an AI's ability to perform tasks through internet navigation and software use. Ji stated that "[Manus] is not just another chatbot or workflow. It is a fully autonomous agent that bridges the gap between conception and execution... We see it as the next paradigm of human-machine collaboration."
Nevertheless, some early users have expressed doubts about Manus. Alexander Doria, co-founder of the AI startup Pleias, reported encountering error messages and endless loops during his testing. Other users noted that Manus makes mistakes on factual questions, does not consistently cite its sources, and often omits information that can easily be found online.
My own experience with Manus has not been very positive. When I asked it to order a fried chicken sandwich from a well-rated restaurant in my delivery area, Manus crashed after ten minutes. On a second attempt, a menu that matched my preferences appeared, but the platform could not complete the ordering process or provide a link for payment. Additionally, when I tried to book a flight from New York to Japan, despite giving clear instructions, all it managed to do was redirect me to various fare pages on different airlines, many of which did not work.
When trying to reserve a table at a nearby restaurant and develop a video game inspired by Naruto, Manus froze again. After three days of testing, my opinions about Manus are mixed. On the positive side, I highlight its ability to research online and generate reports, as well as its capability to execute tasks through scripts. However, it has technical issues that prevent it from fulfilling its promises.
A representative from Manus claimed that their small team is focused on improving the platform and addressing the issues reported during the closed beta. While Manus seems destined for a great future, the success of its initial launch has been driven by factors such as invitation exclusivity and enthusiasm in Chinese media. However, misunderstandings about its capabilities have arisen, including a video that did not accurately represent Manus in operation. Currently, although the company claims to be working on scalability and resolving bugs, Manus appears to be more of a promise for the future than a fully functional reality.
