DeepSeek is rewriting the AI playbook

The emergence of Chinese artificial intelligence (AI) startup DeepSeek has stunned the tech world; the company’s AI chatbots are said to rival the capabilities of major competitors like ChatGPT and Gemini but were built at a fraction of the cost and released fully open-source.

DeepSeek’s unveiling to the world served as a Wall Street wrecking ball, wiping off nearly US$1 trillion from major US technology stocks.

Experts agree that DeepSeek has delivered genuinely impressive technical innovations for the AI industry, but not everyone is sold on the technology. Countries including Australia have banned its use on all government devices and systems – citing security risks – with other nations expected to follow suit.

While this decision feeds into a deeper narrative about AI and data security, it should not overshadow the most profound takeaway from DeepSeek’s breakout success.

The startup’s unexpected and sudden rise to the top is testament to the fact that building smaller, yet powerful AI models like DeepSeek’s R1 and OpenAI’s o1, which focused on step-by-step reasoning and problem-solving, don’t require more money or data centres, but rather human ingenuity and willingness to experiment.

DeepSeek should serve as a case study for the Australian government to pour more funding into homegrown AI research projects, whether it’s universities, government agencies or Aussie tech startups. It could usher in a new era where more players can enter the AI ecosystem to build powerful tools with limited resources. In fact, these small models, built on top of the AI chatbot giants (think ChatGPT), are even more powerful in aiding complex cognitive tasks, such as math, logic, and research reporting.

Despite its Chinese association, DeepSeek is just one amongst many brain trusts around the world already working within the new AI paradigm: turning existing gains in machine intelligence into more powerful, more precise tools for day-to-day cognitive tasks.

At its core, it is just a group of well-supported, bright researchers and engineers free to experiment, investigate and improve on existing open-source AI model. Much like a better-funded version of an academic research group at a university.

Universities often face tight computational constraints, and research students are bright minds accustomed to working with limited resources. Like DeepSeek, they seek powerful solutions by working ‘smart’ rather than ‘big’.

This means improving model architectures and computational efficiency, and creating expert models for specialised contexts. DeepSeek’s breakthrough proves that transforming research prototypes into real-world applications – from commerce to healthcare, to scientific discovery – may require just a fraction of the resources commanded by US tech giants.

This revelation opens a compelling pathway for nations like Australia to cultivate its own competitive AI ecosystem that’s capable of going toe-to-toe with the Silicon Valleys of the world, but this can only be achieved through more financial and institutional support.

AI chatbots have already released the equivalent of a decade’s worth of intelligence for society to absorb. The true challenge lies not in resisting this tide but in determining how wisely and safely we integrate it.

Our ability to navigate this transformation will depend largely on how we develop and deploy these smaller, more specialised models in the coming years.

Now is the time for Australia to capitalise on our own research expertise in AI and expand our footprint in this rapidly growing sector.

This article first appeared at The Canberra Times.