TRANSCRIPT
New competition in the world of artificial intelligence technology.
It was the number one downloaded free app on Apple's iPhone store on Monday - and Chinese startup DeepSake says their AI chatbot was developed at a fraction of the cost of its US counterparts.
This has American companies, like OpenAI, worried.
Dean of the Faculty of Computing, Law and Psychology at The University of Buckingham, Harin Sellahewa explains.
"DeepSeek has the advantage of standing on the likes of ChatGPT because they have done the work for a number of years already. So that's a key difference. Plus, DeepSeek is open source, which mean developers can create other innovative products and services that relies on DeepSeek compared to... there are some restrictions on what you can do with the likes of ChatGPT."
The startup DeepSeek was founded in 2023 in Hangzhou China and released its first AI large language model later that year.
DeepSeek began attracting more attention in the AI industry last month when it released a new V3 AI model that it boasted was on par with similar models from US companies such as ChatGPT maker OpenAI, and was more cost-effective in its use of expensive chips to train the system on troves of data.
But it was a research paper published last week that really caught the world's attention, following the release of an updated model R1.
It revealed how this version showed advanced 'reasoning skills' - such as the ability to rethink its approach to a math problem - and was significantly cheaper than a similar model sold by OpenAI called o1.
A report by The Conversation said DeepSeek's V3 model cost $8.9 million in contrast to OpenAI's $160 million $100 million.
Mr Sellahewa says the program works to more efficiently direct your prompt or question to specific areas of your language model.
"It's that when you want to solve a problem, you're asking specific area of AI models which is where you gain the efficiency. So, it's like you're not using your entire brain to solve a problem. You're asking for a specific part of the brain to solve a problem."
The expensive chips used by DeepSeek are from California-based company Nvidia - which saw its stock drop 17 per cent on Monday.
But the company said in a statement that DeepSeek's work was 'an excellent AI advancement' and utilised widely available models with full control over the export.
One key feature that distinguishes DeepSeek from competitors like OpenAI is that its models are open source - meaning anyone can access and modify components of the model, as researcher at Oxford Internet Institute Kayla Blomquist says.
"So much of the information about the research going into this, the model weights, how it's constructed and the processes that they've used to treat it as well as the code that users can locally download to run and tweak on their end - that is not something that we see with equivalents like ChatGPT. That is all how behind closed doors."
It's not just the US that's curious about this readily accessible and more affordable technology.
Australian politicians and experts have come out to both praise and condemn the AI model.
Newly announced chief scientist Tony Haymet says he views AI as a great opportunity, and a great export opportunity in particular to deliver AI with renewable electricity.
Science Minister Ed Husic is more cautiously optimistic of the benefits for Australia.
"Can I just emphasise if we take anything out of today, it's to shake the misconception that China has not been involved in the development of artificial intelligence until the last minute. Reality is, you will be well aware, some of their country's goals have been by 2025 that they will be a world leader in artificial intelligence. How do we know that? They set it in the middle of last decade. They set those targets."
Others, including Housing Minister Clare O'Neil, are citing potential national security concerns.
She spoke to Channel 7's Sunrise program.
"You mention those national security comments, and you know, we would just urge Australians to exercise real caution about the personal information that they're giving away. 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 about you."
Opposition finance spokeswoman Jane Hume is calling on the government to publicly share any security concerns.
Senator Hume also spoke to Channel 7.
"It's the highest download app in Australia, it's overtaken Chat GTP, we would urge caution on this one. Take our advice from the security experts, and we'd hope that the government will make that advice explicit, because we want to make sure that we keep Australians safe. There's plenty of opportunities as Clare said in AI both professionally and personally. But we would urge caution on this one."
It's clear the technology is innovating on pre-existing models, as DeepSeek reveals its language model's train of thought which it then uses for further training without having to feed new sources of data.
They claim to have used a lower number of Nvidia chips than previously expected, and said by 2022 they had amassed 10,000 of the California-based company's high-performance chips.
But there's skepticism around DeepSeek's resources, with analysts from Bernstein Research on Monday highlighting that their total training costs for its previous V3 model were unknown and much higher than what the startup said was used for computing power.
Experts and users must wait to see if the model performs better than its predecessors or is simply more efficient and cost-effective.
