by Tan Aik Keong (AK)
I recently joined a live radio interview, a collaboration between Oriental Daily and AiFM, to talk with listeners about DeepSeek and its far-reaching impact on the AI industry. DeepSeek has drawn massive attention in AI circles — not just as an innovative product, but as something that's lowered the barrier to using AI and pushed the whole ecosystem toward being more accessible, open and democratised. So what exactly is DeepSeek? How is it reshaping the market? And where is it headed? Here's a closer look.
Low cost, high efficiency
DeepSeek was founded in China in 2023 by Liang Wenfeng, and has released DeepSeek V3 and DeepSeek R1, both particularly strong on Chinese-language understanding, coding and maths. Its core edge is low cost and high efficiency: DeepSeek's training and inference costs run dramatically lower than competitors like OpenAI — around $5.5 million, versus an industry norm often north of $100 million — making AI access genuinely more attainable. The app itself went viral fast, topping the Apple App Store charts in both the US and China, with its market influence expanding quickly. It's also broadly compatible, running on major cloud platforms including NVIDIA, Microsoft Azure and AWS, which significantly widens its usefulness.
Open source, and a smarter architecture
Technically, DeepSeek combines open-source access with an innovative architecture. It isn't the first open-source AI player — Meta's LLaMA and Mistral are open source too — but its optimised training and inference cost pushes AI accessibility further still. It uses a Mixture-of-Experts (MoE) architecture with 671 billion parameters across its expert network, but activates only 37 billion at inference — outperforming traditional dense neural networks on both cost and efficiency. It also uses automated reinforcement learning in place of traditional supervised fine-tuning, streamlining the training pipeline, and ships distilled models that can run on ordinary computers without needing expensive compute.
A price war, and an open-vs-closed split
DeepSeek's rise has pushed AI pricing competition into overdrive — OpenAI recently launched a free version of o3-mini partly in response. The industry is also splitting into an "open source versus closed" landscape: DeepSeek, Qwen and Meta's LLaMA represent the open-source camp, while OpenAI and Google Gemini hold to closed models. That competition doesn't just shape AI software — it's rippling into AI hardware too, and as open-source AI spreads, GPU demand patterns may shift meaningfully. On the MMLU benchmark, DeepSeek performs strongly on maths and coding, and leads comparable open-source models specifically on Chinese-language processing — filling a real gap in the market.
Ethics can't be an afterthought
AI is moving fast, but safety and ethics can't be sidelined. As DeepSeek and similar products spread, AI ethics has become a genuine focus area. Malaysia has already introduced its AI Governance and Ethics (AIGE) guidelines, offering a reference point for safe AI use, built around inclusivity, fairness, accountability, safety, transparency, reliability and human wellbeing. For companies to use AI responsibly, that means setting up an AI committee, drafting AI ethics guidelines, protecting data privacy, and building a prompt library to keep AI interactions consistent — alongside AI education (MYMahir, prompt-engineering courses) to build public understanding, and ongoing AI audits and monitoring to keep systems within ethical and legal bounds.
Where AI goes from here
AI's development has moved through several stages — from early image recognition, to today's generative AI, and on toward agentic AI and physical AI next. The path toward AGI can be roughly mapped across five levels: conversational AI at the base, then reasoning AI, then autonomous AI, then innovator AI, and finally organiser AI. DeepSeek's vision is for AI to become a tool everyone can use on equal footing — not something reserved for a handful of tech giants. In this new era, there's no good reason not to embrace it.
The future is already here, and AI is set to become our ally, not our competitor. Let's embrace it, and build a smarter world together.
Part of the AK AI Corner column. Originally published in Oriental Daily (东方日报) on Feb 8, 2025.