Did DeepSeek beat ChatGPT? A question well worth answering, as the self-proclaimed “Top Open-source Large Language Model” from China made its sudden debut and is making significant waves in the AI assistant market. The development has disrupted ChatGPT’s dominance and its position on the technological stock market, sparking broader discussions on China’s development in the information technology sector, its implications, and its potential to change the world.
DeepSeek’s website asserts “DeepSeek-R1 is now live… rivaling OpenAI’s Model o1,” with the inclusion of a chart showing their benchmark scores, claiming to trump over ChatGPT 4o in nearly every metric.
The potential for DeepSeek is untested in the fast-paced nature in Hong Kong, particularly since AI assistants such as ChatGPT and Gemini are unavailable here. Although other AI tools like Microsoft’s Copilot remain popular on university campuses and in Central’s offices, could DeepSeek take over the AI assistant game in Hong Kong?
Let us put ChatGPT-4o and DeepSeek-V3 to the test by examining both using the context of university life through the comparison of answers generated by the AI models.
ChatGPT 4o vs DeepSeek V3 Battle
(The conversations were created on March 2, 2025)
1. Getting ideas and recommendations from AI
As AI is frequently utilised for gaining a new perspective or idea on a given topic, let us try to ask both ChatGPT 4o and DeepSeek V3 for flower viewing place recommendations.
Prompt:
Can you show me 5 best places to view flowers in Hong Kong?
Result:
Both ChatGPT 4o and DeepSeek V3 have given accurate locations for flower viewing in Hong Kong. DeepSeek V3 has provided the best time to visit the locations for flower viewing, while ChatGPT 4o provided the transportation information for the locations. Moreover, ChatGPT 4o has cited a recommended location from our previous article with its internet searchability.
2. Analyzing image
Both ChatGPT 4o and DeepSeek have the ability to analyze pictures. Let’s upload a picture and ask both AI bots what they can see.
Again, we will input the same prompt for both AI models:
Can you tell me what you see in this picture? And where do you think the picture was taken?
Result:
In this case, ChatGPT 4o had an advantage over DeepSeek. Both ChatGPT and DeepSeek recognize this picture is of a street, but DeepSeek made a basic estimation on the location – “possibly in an East Asian country” while ChatGPT was very confident that it was located in Hong Kong after recognizing the red Hong Kong taxi and the “LOOK RIGHT” crossing sign.
As a made-in-China AI model, DeepSeek also made a wrong output of the Chinese characters outside the bank, giving the result of “定存傻惠” (Stupid offers for fixed deposits) instead of “定存優惠” (Offers for fixed deposits). While the mistake may not be too serious in this case, it could make room for significant misunderstandings if DeepSeek misinterpreted something more sensitive or critical.
4. “Think deeper” function: Write a Video Script for a project
Both ChatGPT 4o and DeepSeek V3 feature an option for deeper analysis. Known as “Reason” on ChatGPT 4o and “DeepThink R1” on DeepSeek V3, this function allows the AI to take more time to generate more comprehensive and detailed responses for users.
Let us try to use these functions to create a short promotional video for The University of Hong Kong‘s Journalism and Media Studies Centre.
Prompt:
I am making a 2-mins promotional video for HKU’s Journalism and Media Studies Centre but I only have 2 people in my crew. Can you write me a creative script and storyboard to promote the Bachelor of Journalism program to prospective international students?
Result
Both ChatGPT 4o and DeepSeek V3 have presented over six scenes, showcasing visual elements and interviews with professors and students. However, DeepSeek has gone a step further by creating simple storyboards.
Additionally, both ChatGPT and DeepSeek included filming tips, which is a nice bonus even though it was not requested.
It is also worth considering the time spent for deeper analysis. DeepSeek spent 20 seconds to think, while ChatGPT used 7 seconds. However, in return, DeepSeek has generated the storyboard, which ChatGPT did not.
What do university students in Hong Kong think about DeepSeek?
Despite the global controversies behind the creation of DeepSeek, it remains increasingly popular among university students in Hong Kong. Many appreciate its thorough explanations for solving homework problems, while others have discovered and enjoy exploring its unexpected capabilities.
DeepSeek helped me a lot on revision. It can explain the mathematics step-by-step, with or without the “Think Deeper” function. I feel like it is better than ChatGPT.
– Marco Chan, an Engineering student from Hong Kong Polytechnic University.
I think it is very nice. It can even do fortune-telling, and I found it pretty accurate.
– Laurel Wang, a Chinese Literature student from The University of Hong Kong.
A Hong Kong version of DeepSeek?
On February 25, 2025, the Hong Kong Generative AI Research and Development Center announced the soon-to-be-released “HKGAI V1” locally-made AI language model. Driven by DeepSeek’s open-source model, the HKGAI V1 is able to handle inputs in Mandarin, Cantonese, and English. Hong Kong’s IT Secretary Sun Dong believed that the new AI model could help navigate the government’s efficiency towards a higher quality of development.
Universities in Hong Kong appear to be easing their stance on AI usage. The Hong Kong University of Science and Technology has launched generative AI research labs, while the University of Hong Kong has announced that AI literacy will be a mandatory course for incoming students starting in the coming semester. It would be fascinating to observe the growth and implementation of AI within university campuses and across various industries in our daily lives.