When Ye Shengwei first asked the new Chinese artificial intelligence (AI) chatbot DeepSeek to write an article for him, he was amazed by its quality.
"It provided in-depth analysis and told me the reasons why it was generating that content," Mr Ye said.
"It gave me some surprisingly good points for the topic and drew some useful conclusions."
Mr Ye, 35, works for a major tech company as a software product manager and so had already been using Western AI chatbots to draft articles for work.
The Chinese government's "Great Firewall" blocks access to Western websites for ordinary citizens.
Mr Ye said he thought ChatGPT was the best AI tool but after using the home-grown Chinese chatbot, he said he might change his preference.
"ChatGPT's output still requires me to manually edit and polish sentence by sentence, which still is time-consuming, and it tends to give me vague conclusions," Mr Ye said.
"If DeepSeek can make my writing more efficient, I'll keep using it."
'The best Lunar New Year gift'
Since the release of its new AI model R1 on January 20, DeepSeek has been dominating global headlines, shaking the stock market, wowing computer scientists and stoking pride in China.
Creator High-Flyer Quant — a previously little-known AI hedge fund company — claims it was built in the face off US restrictions on the latest technology for a tiny fraction of the cost of AI models such as ChatGPT, and that it is much more efficient.
Its release coincided with the return to the White House of Donald Trump, who called it a "wake-up call" for American companies.
Chinese netizens shared their excitement on social media, with one describing it as "the best Lunar New Year gift".
"I think DeepSeek is fully capable of doing the job, and it's much clearer in logic, which is very suitable for Chinese students," one Chinese social media user posted.
"I've already unsubscribed from ChatGPT and plan to stick with DeepSeek."
Censorship and data security concerns
While the Chinese government, state media, and users have lavishly praised DeepSeek, others have expressed concerns about censorship and potential surveillance by Beijing.
In-built restrictions limit DeepSeek's responses to questions about subjects that are sensitive in China, such as the Tiananmen Square massacre in 1989.
Some, including Australian cabinet ministers, have also warned that people should be careful when using DeepSeek and be aware of their data security.
"It's fine to talk to the app but perhaps don't give it personal information that you don't want the rest of the world to know," Housing Minister Clare O'Neil told Seven's Sunrise program.
Science Minister Ed Husic told ABC News there remained a lot of unanswered questions, including over "data and privacy management".
"I would be very careful about that. These types of issues need to be weighed up carefully," he added.
Nicholas Davis, an AI expert at the University of Technology Sydney, said a lot of platforms censored content, and in China "the government regulates what material is shown".
"All AI models have been fine-tuned to shape the outputs in certain directions," he said.
According to Professor Davis, scientists have successfully removed the censorship in some open-sourced Chinese models and, theoretically, it could be done with DeepSeek as well.
"An important question is, in the case of DeepSeek, are people worried about specific topics of censorship? Or do people care that it is a Chinese company?"
Professor Davis said data security was a bigger concern than censorship and one that was relevant to all the AI platforms.
"Personally, a bigger issue is the risks created by typing personal information [into an online large language model (LLM)]," he said.
"For example, I would never put any sensitive work information into [an internet] browser-based large language model, because I know that companies like Anthropic and OpenAI can see and potentially use what I'm writing.
"I feel just as concerned when it comes to DeepSeek. If I worked in the government, I may feel even more concerned about using DeepSeek.
"That's not a question about censorship. That's a question about, 'What happened to my data?' and 'How could that data be used?'
"However, if everyone is using DeepSeek and only DeepSeek, yes, censorship is a critical social problem because free speech supports democracy.
"But I would argue that privacy is the bigger issue right now.
"People might easily think, 'I'm going to use this AI to organise my finances.'
"With any LLM, this would expose sensitive information that could come back to bite you."
A breakthrough 'we should be engaging with deeply'
Professor Davis said he had been testing and benchmarking DeepSeek, which can be installed on users' computers to be used locally instead of via the internet.
He said it was "extremely good" and he was impressed by its efficiency.
"If I'm on a long plane trip, now I can run an extremely capable model directly on my MacBook during the flight," he said.
"I don't need an internet connection and ask a server to run ChatGPT or [Anthropic's AI chatbot] Claude for me. I can just do it locally.
He said that despite people's concerns, open-source models such as DeepSeek could assist with broader understanding of the technology.
"If it's all controlled and managed entirely and only within large tech companies, that means there is less opportunity for independent experts and the community to play, to test and to understand it," he said.
"DeepSeek is a really interesting genuine breakthrough that we should be engaging with deeply."
