“Control your greed,” counsels Xiaomi Co-founder and Senior Vice President at EMBA-Global Asia 2019 Recognition Ceremony

“Control your greed,” counsels Xiaomi Co-founder and Senior Vice President at EMBA-Global Asia 2019 Recognition Ceremony

Mr. Feng HONG, Co-founder and Senior Vice President of Xiaomi, the world’s fourth-biggest smartphone company, addressed graduates from class of 2019 at the EMBA-Global Asia Recognition Ceremony held on 22 June.

After graduating from Shanghai Jiao Tong University with a degree in computer science and engineering, Mr. Hong received a Master’s Degree in computer science from Purdue University. After a career in Google, Mr. Hong initially spearheaded the development of Xiaomi’s MIUI platform, the smartphone’s operating system. He now oversees Xiaomi Finance, the firm’s financial services arms.

Mr. Hong congratulated the new graduates and noted that the EMBA-Global Asia is a very respected programme before sharing some lessons from his nine years at Xiaomi.

While the retail price of most products is usually six times the manufacturing cost, and 8 to 10 times for clothing, Xiaomi’s mark-up is typically just 10 percent. The company’s logic is to look to the bigger picture and the longer term, explained Mr. Hong. Selling at a lower price enables the company to sell more products, which further reduces material costs, creating a positive cycle.

“However, we do have a lot of arguments on the pricing part, especially when we enter some areas with traditionally very high margins,” he admitted.

Setting the price for the company’s air purifiers a few years ago was one such occasion. These products typically retail for about 3-4000 RMB and they cost about 900 RMB to produce.  “So how to price them? We thought about half the price was already good enough, [but] our CEO decided to set the price at 899 RMB,” Mr. Hong recalls.

The production team was not happy, with one member complaining that the target price was an insult to his work. Mr. Hong tried to convince the CEO to increase the price at least to 999 RMB, but the CEO told him: “We need to be disciplined at making just a small, reasonable profit on everything we do, and we need to do it religiously. I want to build trust with our customers so one day when they go to our store, they don’t need to look at the price tags, they just know they’re getting a pure product at the best price in town. Control your greed.” The discussion was over.

Over time, as sales revenues have multiplied, Mr. Hong has come to understand and appreciate the CEO’s strategy.

Xiaomi has also pursued innovative product development strategies to produce its many smart devices. Most of these products are built by the firm’s eco-system companies, explains Mr. Hong.

In 2014, the firm saw a great opportunity with smart devices, “We believed all those devices would be upgraded with connecting capabilities and they would be connected all together in the years to come. The problem was how to capture this trend.”

The company eventually decided to invest in incubation.

“We identified teams with good product engineering skills and provided all the other support,” he says. That support included providing all the start-up funds, sharing sales and marketing support, and providing supply chain support connections to allow these companies to get volume discounts very quickly.

The model has been a great success and Xiaomi now counts more than 100 companies in its eco-system. Between them, the companies are now launching almost one product every four days.

“Sometimes, it may be the right decision to step back, to take less, and do so willingly,” Mr. Hong advised the graduates. “After all, good companies make profits, but great companies win the hearts and minds.”

Other Events
港大經管學院最新多模態AI圖像生成能力排名出爐 部份中國人工智能模型表現突出
2025 | 學院新聞
港大經管學院最新多模態AI圖像生成能力排名出爐 部份中國人工智能模型表現突出
港大經管學院今日發表《人工智能模型圖像生成能力綜合評測報告》,針對15個「文生圖模型」及7個「多模態大語言模型」進行全面評估。研究顯示,字節跳動的即夢AI和豆包,以及百度的文心一言,在新圖像生成的內容質素及圖像修改的表現突出;而早前引起全球關注的DeepSeek最新推出的文生圖模型Janus-Pro,則在新圖像生成方面表現欠佳。研究亦發現部分文生圖模型雖然在內容質素方面表現優異,卻在安全與責任方面的表現強差人意。整體而言,與文生圖模型相比,多模態大語言模型整體表現較佳。
人工智能圖像生成評測成績單公布: 字節跳動百度表現亮眼,DeepSeek Janus-Pro表現欠佳
2025 | 研究
人工智能圖像生成評測成績單公布: 字節跳動百度表現亮眼,DeepSeek Janus-Pro表現欠佳
如今,人工智能領域的前沿模型技術已經從文本處理拓展至視覺信息的深度理解與生成。這些模型既能精准解讀圖像語義,又能根據文字描述創作出兼具真實感與藝術性的視覺內容,展現出令人驚嘆的跨模態理解與創作能力。本研究聚焦全新圖像的生成和基于現有圖像的圖像修改兩大核心任務,提出了一套系統性的人工智能模型圖像生成能力評測框架。我們基于多維測試集的構建與專家評審,對15個專業文生圖模型和7個多模態大語言模型的圖像生成能力進行了全面評估。結果顯示,字節跳動的即夢AI和豆包以及百度的文心一言在新圖像生成的內容質量與修改任務中表現突出,位列第一梯隊。對比不同類型的AI模型,我們發現,相對于專業文生圖模型,多模態大語言模型整體表現更佳。