Deep Dive: Chinese AI firm DeepSeek challenges ChatGPT with new chatbot

Chinese artificial intelligence (AI) start-up DeepSeek released its chatbot app on January 20. Its performance has challenged the dominance of American tech giants like OpenAI.

The large language model (LLM) is called R1. DeepSeek released an earlier model called the V3 in December. According to the firm, V3 was built at a fraction of the cost and computing power that major US tech companies use to build their LLMs.

The low cost of DeepSeek called into question the billions of dollars US tech companies are spending on energy-hungry data centres. Some regarded it as a shocking realisation for the US AI industry, especially because DeepSeek boasts an open-source model.

The Chinese company said it had outperformed leading US developers’ products on various industry benchmarks, including for mathematical tasks and general knowledge.

Denis Simon is a non-resident fellow at the US think tank Quincy Institute. He said the most important lesson for the West was that “there were many paths to the same innovation target”.

He added that DeepSeek had followed an innovative approach to push “the frontier of the current thinking about AI technology”.

Should children be banned from using chatbots like Character.ai?

“By developing a lower cost, more efficient, and perhaps even more effective path to producing ‘artificial general intelligence’, DeepSeek has shown that it’s not all about scale and money,” Simon said.

“In fact, it is about cultivating talent and thinking more about software integration than it is about accumulating thousands and thousands of advanced chips.”

Led by DeepSeek founder Liang Wenfeng, the team is a pool of fresh talent. Many of them are young researchers and doctorates from top Chinese universities.

US President Donald Trump called DeepSeek a “wake-up call” after US stocks were affected amid fears the model could threaten American dominance in the technology sector.

DeepSeek was the most downloaded free app on the Chinese iOS App Store as of January 27. It also dethroned OpenAI’s ChatGPT to claim the top spot on the US iOS App Store.

DeepSeek has integrated the model into its chatbots’ web and app versions for unlimited free use.

OpenAI charges US$200 per month for unlimited access to its o1 models. A standard plan with limited access costs around US$20 per month.

Staff writers

1. According to News, which of the following is NOT a characteristic of DeepSeek?

(1) It offers unlimited use of its chatbot for free.
(2) It was more expensive to develop than ChatGPT.
(3) It was developed in the US.
(4) It was designed by Chinese researchers.

A. (1), (3) only
B. (2), (3) only
C. (1), (4) only
D. (3), (4) only

2. List TWO reasons why DeepSeek is causing a stir in the AI industry.

3. According to Denis Simon, what is ONE reason for DeepSeek’s success and what could US technology companies learn from it?

1. How does the graphic relate to the information presented in News?

2. Using News and the graphic, list TWO factors contributing to the growth of China’s AI industry.

• OpenAI says there is evidence the firm stole data using a practice common across the industry

• Italy’s Data Protection Agency has blocked access to DeepSeek over privacy concerns

Chinese artificial intelligence (AI) start-up DeepSeek has been accused of data theft through a practice that is common across the industry.

OpenAI said there is evidence that DeepSeek used distillation of its GPT models to train the open-source V3 and R1 models at a fraction of the cost of what Western tech giants are spending on their own, the Financial Times reported.

OpenAI and Microsoft, the ChatGPT maker’s biggest backer, have started investigating whether a group linked to DeepSeek exfiltrated large amounts of data through an application programming interface (API), Bloomberg reported, citing people familiar with the matter who asked not to be identified.

“Distillation will violate most terms of service, yet it’s ironic – or even hypocritical – that Big Tech is calling it out. Training ChatGPT on Forbes or New York Times content also violated their terms of service,” Lutz Finger, a senior visiting lecturer at Cornell University who has worked in AI at tech companies including Google and LinkedIn, said in an emailed statement.

DeepSeek has its own distilled models that use other open-source models, such as Meta Platforms’ Llama and Alibaba Group Holding’s Qwen. Alibaba owns the South China Morning Post.

However, OpenAI alleges that DeepSeek used API access to the closed-source GPT models to distil those in an unauthorised manner. DeepSeek has not admitted to using distillation in training its main models, V3 and R1.

More than just friends? Dangers of teens depending on AI chatbots for companionship

The concept of AI distillation has been around for years. In 2015, three researchers, including the “Godfather of AI” Geoffrey Hinton, published a paper titled “Distilling the Knowledge in a Neural Network”, illustrating how knowledge from large models could be transferred to smaller models that are easier to deploy.

The process is often likened to what happens in a classroom, with a “student” model learning from a “teacher” model. Through processes that involve text classification and answering questions, the student model learns appropriate responses to certain types of prompts. This training includes probabilities for all possible responses.

Meanwhile, Italy’s Data Protection Agency (GPDP) launched an investigation into DeepSeek last month, saying it had blocked the company from processing Italian users’ data.

The agency said it had decided to act after receiving “completely insufficient” answers to its questions about the firm’s use of personal data.

In December, the Italian watchdog fined OpenAI €15 million (HK$120,408,300 or US$15.6 million) over the use of personal data by its popular ChatGPT chatbot, but the US tech firm said it would appeal.

The investigation began in March 2023 when the GPDP temporarily blocked ChatGPT in Italy over privacy concerns.

Staff writers, Agence France-Presse

1. What is DeepSeek accused of doing with OpenAI’s technology?
(1) taking user data from OpenAI
(2) copying data from ChatGPT
(3) stealing computer chips from OpenAI
(4) training smaller models to mimic the behaviour of ChatGPT

A. (1), (3) only
B. (1), (4) only
C. (2), (4) only
D. (3), (4) only

2. What role did distillation allegedly play in the development of DeepSeek? Identify ONE potential benefit and ONE potential downside of this method.

3. Using Issue, list ONE reason why Italy’s Data Protection Agency has taken action against DeepSeek.

1. Why is ChatGPT being compared to DeepSeek?

2. What impact might the DeepSeek controversy have on the future of AI development and regulation? Explain using News, Issue, Glossary and your own knowledge.

large language model (LLM): a type of artificial intelligence that can understand and create human language. These models learn by studying huge amounts of text from books, websites, and other sources.

data centres: giant warehouses that house powerful computers and networking equipment, storing and processing the world’s data.

open-source: software made available to the public for free, allowing anyone to view, modify, and distribute its code.

advanced chips: specialised computer chips used to run the complex AI algorithms needed for natural language processing and generation.

distillation: a means of training smaller models to mimic the behaviour of larger, more sophisticated ones. The practice is common internally at many companies looking to scale down the size of their models while offering similar performance to users.

exfiltrating: secretly copying data from a computer or network and sending it somewhere else without permission.

1. B

2. The makers of DeepSeek say they spent less money and used less energy to create the chatbot than OpenAI did for ChatGPT. It is also open-source, meaning the software is free to the public. This allows anyone to view, modify, and distribute its code. (accept all reasonable answers)

3. Simon said DeepSeek focused on talent, hiring young professionals to work on its chatbot. American companies tend to prioritise acquiring advanced chips. American companies could learn from DeepSeek by shifting their focus to hiring talented researchers and software integration, saving them money. (accept all reasonable answers)

1. The graphic shows the US and China AI industries. ChatGPT and OpenAI are represented by the tree growing in America, and the one in China is DeepSeek. China’s industry is still small and growing compared to the US, as depicted by the smaller tree.

2. The graphic shows China’s industry receiving support in the form of technology and cash. Companies like DeepSeek are also focusing on building a pool of talented people to advance their technology rather than spending money on acquiring advanced chips. (accept all reasonable answers)

1. C

2. DeepSeek used distillation to train its models on existing AI models. One potential benefit is that it could reduce the number of advanced chips and data centres needed to train and improve AI models, but a potential downside is the legal and ethical issues that distillation creates, as it has been alleged that DeepSeek did it without permission.

3. Italy’s Data Protection Agency launched an investigation into DeepSeek’s chatbot and blocked it from processing Italian users’ data. This decision came after the agency received inadequate responses from DeepSeek regarding how it collects, stores, and uses personal information.

1. OpenAI’s ChatGPT is being compared to DeepSeek because DeepSeek’s cost-effective AI models are challenging the dominance of ChatGPT, with some claiming that DeepSeek outperforms OpenAI in certain areas, such as mathematical tasks and general knowledge, while being developed at a fraction of the cost. Both of these products caused a huge stir when they came out. These technological developments are also representative of the competitiveness between China and the US.

2. The DeepSeek controversy highlights key challenges in AI development, including ethical concerns over data usage, intellectual property rights, and international competition. If DeepSeek indeed used OpenAI’s models without permission, it raises questions about how to enforce AI terms of service across borders. Additionally, the case underscores the growing need for regulatory oversight in AI, as seen in Italy’s investigation into DeepSeek’s data practices. With AI technology advancing rapidly, governments and tech companies will likely face increasing pressure to establish clearer guidelines on data privacy, fair competition, and the ethical training of AI models. (accept all reasonable answers)